Главная > Документ

1

Смотреть полностью

SAFETY OF ALTERNATIVE APPROACHES TO CHILDBIRTH

A DISSERTATION

SUBMITTED TO THE DEPARTMENT OF SOCIOLOGY

AND THE COMMITTEE ON GRADUATE STUDIES

OF STANFORD UNIVERSITY

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

Peter F. Schlenzka

March 1999

© Copyright by Peter F. Schlenzka 1999

All Rights Reserved

I certify that I have read this dissertation and that in my opinion it is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy.

W. Richard Scott, Ph.D., Principal Adviser

I certify that I have read this dissertation and that in my opinion it is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy.

Byron Wm. Brown, Jr., Ph.D.

I certify that I have read this dissertation and that in my opinion it is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy.

Morris Zelditch, Jr., Ph.D.

Approved for the University Committee on Graduate Studies:

abstract

This study examines the obstetric profession's claim and society's belief that the hospital-based obstetric approach to childbirth is safer with respect to perinatal mortality than the natural approach to childbirth as administered by midwives and some physicians in free-standing birth centers or at home. Using merged birth certificate and hospital discharge data for California for 1989 and 1990, it was possible to identify that part of all pregnancies that could be considered a candidate for natural childbirth and to identify a comprehensive risk profile for each pregnancy. Birth setting was used as a proxy for the childbirth approach. Intrapartum transfers to the hospital were allocated to the originating birth setting. Perinatal mortality was compared with two statistical approaches: indirect standardization using only birthweight, sex, race, age, education, and insurance as risk adjusters, and logistic regression controlling for all risk factors available in the database.

The results for the indirect standardization and the logistic regression show that low-risk women who opt for a natural childbirth in an out-of-hospital setting will experience a slightly though not significantly lower perinatal mortality than low-risk women who opt for a hospital birth under the management of an obstetrician, including the unfavorable results for transfers from home to hospital. Our data also suggest that even for the high-risk levels of our study population the natural approach produces the same perinatal mortality outcomes as the obstetric approach. Our analyses of the California data for 1989 and 1990 do not support the claim by the obstetric profession that for the large majority of low-risk women hospital birth is "safer" with respect to perinatal mortality.

Given no differences in perinatal mortality it must be noted that the natural approach shows significant advantages with respect to lower maternity care cost as well as reduced mortality and morbidity from unnecessary cesareans and other obstetric interventions, and significant benefits from avoiding negative long-term consequences from unnecessary obstetric interventions and procedures. These advantages of the natural approach are of such a large order of magnitude as to raise serious doubts concerning the appropriateness of conventional "obstetric" treatment for low-risk childbirth.

Acknowledgements

I want to thank the following individuals and acknowledge that without their help and contributions I would not have been able to carry out this fairly large research project: Linda V. Walsh, CNM, Ph.D., Associate Professor, University of San Francisco School of Nursing, Director of Midwifery Services of the Lucile Slater Packard Children's Hospital at Stanford, and Chair of the Division of Research, American College of Nurse Midwives (ACNM), who steered me through the medical aspects of this project and checked the plausibility of the results; R. Harold Holbrook, Jr., MD, Associate Professor of Gynecology & Obstetrics, Stanford University School of Medicine, who was instrumental in designing the variables for the medical risk factors; Eric Minch, Ph.D., Research Associate, Department of Genetics, Stanford University School of Medicine, who advised on and ran the statistical analyses; Susan K. Schmitt, Social Science Research Assistant, who was instrumental in developing the complex database; Ciaran S. Phibbs, Ph.D., Center for Health Care Evaluation, VA Medical Center, Menlo Park, CA, who generously made the VA computer system available to me as Stanford was unable to provide the required large computer capacity for a doctoral student; Emmett B. Keeler, Ph.D. and Joan Keesey of RAND Health, Santa Monica, CA, who provided me with their linked birth certificate and hospital discharge dataset for California births for 1989 and 1990.

Special thanks go to Ruth Cronkite, Ph.D. (Center for Health Care Evaluation, VA Medical Center, Menlo Park, CA), Henci Goer, Eric Minch, and Linda Walsh who were kind enough to give their time to read the manuscript and provide many helpful comments.

My dissertation committee consisted of W. Richard Scott, Ph.D., Professor of Sociology and my advisor, Byron Wm. Brown, Jr., Ph.D., Professor of Health Research and Policy, and Morris Zelditch, Jr., Ph.D., Professor of Sociology. This dissertation would not have been possible without their understanding and willingness to embark on this interdisciplinary project. Their coaching and their trust during the sometimes rocky progress provided a constant source of encouragement and guidance.

I am particularly grateful to my wife Julie for her patience, unwavering support, and some forceful nudging in the final stretch. There is no doubt that a dissertation deprives the family of the husband and father for quite some time.

This study, using a merged database with California birth certificate and hospital discharge data, was approved by the California Health and Welfare Agency, Committee for the Protection of Human Subjects, on June 3, 1994 (Project No. 94-06-01).

TABLE OF CONTENTS

Page

ABSTRACT iv

ACKNOWLEDGEMENTS vi

LIST OF TABLES xi

LIST OF FIGURES xiv

INTRODUCTION 1

From Home to Hospital - The Medicalization of Childbirth 4

- The Obstetric Approach to Childbirth 4

- The Natural Approach to Childbirth 6

Is the Obstetric Approach "safer"? 9

- Perinatal Mortality of Hospital Birth Might not Be Superior 10

- Obstetric Interventions: Necessity and Effectiveness 20

. Differences in Obstetric Care: Are the Interventions Necessary? 21

. Effectiveness of Interventions 29

. How Obstetric Interventions Can Be Counterproductive 34

  • Excursion: Predictability of Low-Risk Pregnancies 39

. Predictive Accuracy of Obstetric Risk Assessment Instruments 41

. The Dutch 'Indication List' 43

. National Birth Center Study (NBCS) 45

The Need to Take a New Look at the "safety" Issue 49

- Importance of the Issue 50

- The Re-Emergence of Midwifery 52

- Newly Available Data Overcome Some Previous Limitations 54

OBJECTIVE OF THE STUDY 58

- Objective and Scope of the Study 59

- Birth Setting as a Proxy for Alternative Approaches 61

TABLE OF CONTENTS (cont.)

Page

COMPARING PERINATAL MORTALITY - RESEARCH DESIGN 67

- Data Sources 67

. Birth Certificate Information (Birth Cohort File) 67

. RAND Database with Hospital Discharge Data 68

. Birth Center Information 70

. Quality of the Data 71

- Dataset for this Research 77

. Outcome and Control Variables 78

. Derived Variables 85

. General Exclusions and Size of Database 95

- Methodological Framework 105

. Exclusion Criteria for Birth Centers and Home Birth 106

. Power Calculations 110

. Indirect Standardization (Risk Adjusted Outcome Measure) 112

. Logistic Regression 113

- Limitations of the Study Design 123

. Observational Design and Selection Bias 123

. Desirable Information Missing in Birth Certificate Reporting 125

COMPARING PERINATAL MORTALITY – RESULTS 126

- Low Perinatal Mortality in the Low-Risk Populations 126

- Risk Factors for Low-Risk Pregnancies 129

. General Risk Factors 134

. Medical Risk Factors 135

. Interaction Effects 137

- Perinatal Mortality Comparisons 141

. Indirect Standardization – Results 142

. Logistic Regression – Results 149

TABLE OF CONTENTS (cont.)

Page

THE OBSTETRIC APPROACH IN A LARGER PERSPECTIVE 153

- A Large Cost to Society is Potentially at Stake 154

. Cost Savings from Unnecessary Obstetric Interventions 156

. Cost Savings from a Shared Maternity Care System 159

. Long-term Consequences of Perinatal Interventions 162

- Dominance of the Obstetric Approach 169

- Issues for Further Investigation 171

CONCLUSIONS 174

APPENDICES 177

GLOSSARY 193

REFERENCES 195

list of tables

Page

Table 1 Births and Perinatal Mortality Rates (PNMR) at Places of 13

Birth and Different Levels of Risk in Britain,

1958 Perinatal Mortality Survey

Table 2 Antenatal Prediction Score: Weights Given to Different 14

Risk Factors - 1970 Survey of British Births

Table 3 Labor Prediction Score, Singleton Pregnancies: Weights 14

Given to Risk Factors - 1970 Survey of British Births

Table 4 Proportion of Births and Perinatal Mortality Rates 15

(PNMR/1000 births) by Antenatal Prediction

Scores (APS) - 1970 Survey of British Births

Table 5 Percentage of Births and Perinatal Mortality Rate (PNMR) 15

by Labor Prediction Score (LPS) and Place of Birth –

1970 Survey of British Births

Table 6 Comparison of Interventions for Physicians and 21

Midwives at LA County and USC Women's Hospital

Table 7 Comparison of Interventions for Physicians and 23

Midwives at the University of Michigan Women's Hospital

Table 8 Comparison of Interventions for Obstetricians and 25

Midwives in Washington State

Table 9 Comparison of Interventions for Hospital Versus Home Birth 27

in a Swiss Study

Table 10 Comparison of Obstetric Interventions for Natural Approach 28

(Pithiviers) versus Obstetric Approach (US Hospitals)

Table 11 The Effectiveness of Obstetric Interventions According to the 31

Cochrane Pregnancy and Childbirth Database

Table 12 Comparison of Perinatal Mortality Over Time for 50

Selected Countries

Table 13 Quality of Non-Medical Birth Certificate Data is Rather High 73

Table 14 Variables Used in this Research on Safety of Alternative 80

Approaches to Childbirth

Table 15 Risk Factors Available for this Study Compared to Risk 83

Factors Used in Major Risk Assessment Instruments

Table 16 Birth Settings Based on the Birth Certificate Information 86

about Planned and Actual Birthplace

Table 17 Significant Underreporting of Medical Risk Factors on Birth 92

Certificate

Table 18 Study Population and Exclusions 96

Table 19 Excluded Cases with Inadequate Prenatal Care or Missing 100

Information on Prenatal Care and their Perinatal Mortality

Table 20 Non-Preventable Deaths Among BIRTHSET = 12 Cases After 102

Applying the Exclusion Criteria for Our

Low-Risk Study Populations

Table 21 Conservative Exclusion Criteria 107

Table 22 Liberal Exclusion Criteria 108

Table 23 Power Calculations for the Birth Setting Categories 111

with Smaller Numbers

Table 24 Transformation into Categorical Variables for Logistic 119

Regression Models

Table 25 Univariate Logistic Regression Results for the Four 131

Different Sets of Pregnancy Populations

Table 26 Only 20 Two-Way Interactions Met the Inclusion Criteria. 138

Only Three of those Did not Contain the Risk Factor

Premature Labor

Table 27 Correlation Models for Testing the Significance of 140

Higher-Order Interactions and the Higher-Order

Interactions Chosen for the Logistic Regression Models

list of tables (cont.)

Page

Table 28 Indirect Standardization Results for All Four Risk Levels 143

of the Study Population

Table 29 Indirect Standardization: Results for Study Population after 145

Conservative Exclusions

Table 30 Indirect Standardization: Results of for Study Population after 146

Liberal Exclusions

Table 31 Indirect Standardization: Results for Study Population =2,500 147

Grams (After Excluding Birthweight <2,500 Grams)

Table 32 Indirect Standardization: Results for Entire Study Population 148

(Without Exclusions)

Table 33 Logistic Regression Results for All Four Risk Levels of the 150

Study Population (Models with All Variables and Interactions)

Table 34 Logistic Regression Results for All Four Risk Levels of the 152

Study Population (Models with Antenatal Variables only and Interactions)

Table 35 Comparison of Cesarean Sections for the Different Birth 157

Settings (Indirect Standardization Using the Six Risk Adjusters)

list of figures

Page

Figure 1 Postulated Chain of Diagnostic and Therapeutic 38

Interventions in Perinatal Services

Figure 2 Wormerveer Study Population with Referrals and 44

Outcomes

Figure 3 Comparison of Cesarean Section Rates and Perinatal 49

Mortality Around 1985 for Selected Countries

Figure 4 Definition of Perinatal Mortality 77

Figure 5 50% of All Infant Deaths Occur Within the First Week 78

of Life, 31% on the First Day of Life

Figure 6 Differences for Study Population in Characteristics 113

Between Home, Center, and Hospital Populations:

Education and Insurance Coverage

Figure 7 Differences for Study Population in Characteristics 114

Between Home, Center, and Hospital Populations:

Race and Age of Mother

Figure 8 Differences for Study Population in Characteristics 115

Between Home, Center, and Hospital Populations: Parity

Figure 9 Conventional Statistics Comparing Perinatal Mortality 128

Obscure the Rather Low Mortality for Low-Risk Women

INTRODUCTION

"Out of instinct and desire, we make children. Dependent on the genius of our bodies, we grow them in our wombs. If and when we defer to that same genius during birth, if and when we invite and cultivate power, we find that women can give birth 85 to 90 percent of the time without complication and unreasonable bravery" (Armstrong and Feldman 1990, p. 24).

Depending on the definition, 60-80% of all pregnancies can be considered low risk. Various research suggests that the medicalization of childbirth and the move of childbirth from home to hospital might not have improved the outcomes for these low risk pregnancies. The reason for this counterintuitive phenomenon is seen in today's prevalent medical model of hospital-based maternity care, which focuses on obstetric intervention. For low-risk women these obstetric interventions are considered not just unnecessary but also counterproductive to positive outcomes of childbirth (Kloosterman 1984; Tew 1990; World Health Organization 1985b). This study contrasts the non-interventionist natural approach to childbirth, as administered by midwives and some physicians in free-standing birth centers or at home, with the interventionist obstetric approach, as practiced by obstetricians in hospitals. While more and more research shows significant advantages for the natural approach, the discussion about the pros and cons is still relegated to the fringes of the maternity and child health field. At the core of the controversy about the benefits of the medicalization of childbirth is the issue of perinatal mortality and the obstetricians' claim that obstetrician attended hospital birth is safer.

The evaluation of this claim suffers from the fact that randomized clinical trials with perinatal death as the outcome measure are no longer feasible (see the discussion "Should there Be a Trial of Home versus Hospital Delivery in the United Kingdom?" in the British Medical Journal 1996, 312:753-757). The low perinatal mortality rate, particularly among the low risk pregnancies, would require prohibitively large sample sizes. Experienced delivery systems for natural childbirth and home deliveries have virtually disappeared in most industrialized countries. Neither women nor physicians would consent to randomization (Eskes 1992; Institute of Medicine and National Research Council 1982; and personal communication with Marsden Wagner, Regional Office for Europe of the WHO, who unsuccessfully tried to set up a randomized study in the Netherlands in 1985b). And finally, it is rather questionable whether one could achieve the conditions for a "fair" trial given the vested interests of obstetricians and patients in today's hospital based obstetric model (Springer and van Weel 1996). Wiegers et al. point out that in areas where a patient's choice has a profound effect on outcomes, the elimination of this choice, necessary in a randomized comparison, will result in unreliable estimates of true differences. The woman's choice itself may influence her level of anxiety and apprehension, and in obstetrics levels of anxiety have been shown to predict obstetric complications (Wiegers et al. 1996, p. 313). Thus studies to analyze perinatal mortality for low-risk women will be limited to observational designs, and the quality of these studies will depend on their ability to deal with confounding factors and self selection bias.

Compared to other causes of death within our society perinatal mortality cannot be considered a large cost to society, although the impact on the family in the individual case is devastating. The fact, however, that the perinatal mortality issue seems to block constructive discussion and investigation into alternative approaches to childbirth might well result in considerable and unnecessary cost to society.

This dissertation has therefore two objectives:

1. The primary objective is improving our understanding of perinatal mortality by developing an observational research design, which allows comparing women of equal risk profile when analyzing the two different approaches to obstetric care. This research tests the hypothesis that for low-risk women the non-interventionist natural approach to childbirth, as administered by midwives and some physicians in free-standing birth centers or at home, is as safe as the interventionist obstetric approach in hospitals, using perinatal mortality as the outcome measure for "safety".

2. A secondary objective is to begin outlining a broader framework for comparing these two competing approaches with respect to their overall social and economic cost to society. It is proposed that there is sufficient preliminary research evidence suggesting that the interventionist hospital-based obstetric approach may lead to undesirable outcomes with significant social cost, and that society would be ill served if the comparison of these two competing approaches to childbirth is limited to the issue of perinatal mortality.

The focus of this study is clearly on the perinatal mortality controversy. But the study does want to make the point that perinatal mortality is just one aspect of a larger context and within that context it might not play a dominant role.

FROM HOME TO HOSPITAL - THE MEDICALIZATION OF CHILDBIRTH

Since the turn of the century the birthplace for children in the United States, as in most industrialized countries, has moved from home to hospital. In 1900 less than 5% of the babies were delivered in the hospital, in 1940 it was 50%, and today it is close to 99% (Declercq 1993). The vast majority of women giving birth in American hospitals are dressed in hospital gowns and placed in hospital beds. Eighty-three percent are hooked up to an electronic fetal monitor, and in many hospitals 100%. They receive intravenous injection to induce labor (17%) or augment labor (17%). Sixty percent receive an epidural, a local anesthesia, and a large number1 receive some other form of pain medication. Forty-three percent have an episiotomy, a surgical incision of the vagina to widen the birth outlet. One in five babies are delivered by cesarean section, the second most common hospital-based surgical procedure among women. Another 10% of babies are delivered with the help of forceps or vacuum extraction. (See Table 10 for the sources for all the above intervention statistics).

THE OBSTETRIC APPROACH TO CHILDBIRTH

While the physician strives for a safe birth and a healthy baby, this highly medicalized approach to childbirth (the obstetric approach) is based on medicine's belief that every birth has a high potential for pathology.

The female body is viewed as an abnormal, unpredictable and inherently defective machine. During pregnancy and birth, the unusual demands placed on the female body-machine render it constantly at risk of serious malfunction or total breakdown. (Davis-Floyd 1994, p. 1127)

In the hospital-based obstetric model a pregnancy can only be recognized as low risk or normal in retrospect. Accordingly, we find in a 1985 issue of the New England Journal of Medicine a note on the potential advantages of universal prophylactic cesarean sections (Feldman and Freiman 1985). Along this line is also the invention of "active management of labor" (originally designed at the National Maternity Hospital in Dublin, Ireland), or what Rooks calls "birth by the clock" (Rooks 1997, p. 318), to shorten and control labor for primiparous women2. The Dublin protocol promises delivery after 12 hours of labor. Amniotomy (artificial rupture of membranes) is performed one hour after admission. If cervical dilation is not progressing at 1 cm per hour, then oxytocin is administered intravenously and the dosage is increased until the mother has the desired frequency of contractions. Forty percent of the women in the Dublin trial required oxytocin to progress "normally"; the cesarean section rate remained the same as before (see Goer 1995, pp. 83-85). Looking at the cycle of maternity care from prenatal care through labor and delivery to postpartum care, Enkin observes:

Because of time constraints, [obstetric] specialists caring for women with both normal and abnormal pregnancies have to make an impossible choice: to neglect the normal pregnancies in order to concentrate their care on those with pathologies, or to spend most of their time supervising biologically normal processes, in which case they would rapidly lose their specialist expertise (Enkin et al. 1995, p. 15).

It is no wonder then that the obstetric approach focuses on the pathologies in the labor and delivery phase, and the physician tends to take charge in the patient-doctor interaction and sees himself as the decisionmaker (Rooks 1997, p. 130). The mother concludes her birthing experience by saying: "Thank you, doctor, for delivering my baby" - as if it was not she who gave birth to the child.

THE NATURAL APPROACH TO CHILDBIRTH

The natural approach to childbirth, as Kloosterman, the doyen of Dutch obstetrics, puts it

... is based on the belief that human parturition belongs in principle to such activities as breathing and laughing ... (Kloosterman 1984, p. 115).

The word "natural" derives from "nature", meaning the essential qualities or properties of a thing. The core definition of "natural" in the context of childbirth is "implanted, existing, or present, by nature; inherent in the very constitution of a person; innate; not acquired or assumed" (Simpson and Weiner 1989, p. 242). The ability to give birth is something that is inherent in a woman's constitution. It is innate and not an acquired skill. In this context then birth is seen as a biosocial process that is part of daily life (Wagner 1994, p. 32). Midwifery is rooted in the natural approach. Pregnancy and birth are considered fundamentally healthy processes which have many normal variations; it is a normal part of life, not a medical condition (Steiger 1987, p. 5). Midwives are more likely to accept greater variation as still within the "range of normal" as long as the woman and fetus are tolerating labor well. Only when complications occur which are beyond the midwife's expertise, is the woman transferred to obstetric care. Treating normal labors as though they were complicated can become a self-fulfilling prophecy (Rooks 1997, p. 131). Midwives come to the field predisposed to respect the natural birth-giving ability of women's bodies (Rooks 1997, p. 128). Within this framework the French physician Odent and his midwives developed their concept of "the undisturbed birth", based on their observation that labor seemed to progress most smoothly and safely if women were left to and fully supported in their instinctive innate knowledge. Birthing as an involuntary bodily process cannot be helped, the point is not to disturb it (Odent 1994, p. xix). Midwives view childbearing as an emotionally, socially, culturally, and spiritually meaningful life experience, one of life's major experiences (Zander 1984, p. 129). "Although the birth of a baby is the ultimate goal of pregnancy and labor, the process itself has intrinsic value and great importance in the life and psyche of each woman" (Steiger 1987, p. 6). It can be a most empowering act of creation in a woman's life, or as Kitzinger describes it in her introduction to Odent's3 book:

Birth, like death, is a universal experience. It may be the most powerful creative experience in many women's lives. It can either be a disruption in the flow of human existence, a fragment having little or nothing to do with the passionate longing that created the baby, or it can be lived with beauty and dignity, and labor itself can be a celebration of joy (Kitzinger in Odent 1984a, p. xix)

Midwives are involved in the entire childbearing process with strong emphasis on prenatal care, but also breast-feeding, postpartum recovery, and the first steps of mothering the infant. They perceive part of their role to be to "empower" the pregnant women (Rooks 1997, p. 343), especially important for young, unmarried women, substance abusers, and the socially underprivileged, who are generally at higher risk of complications4. Their philosophy about their protecting, supporting, and empowering role is also reflected in their use of words: they "attend birth" and "catch babies", while it is the mother who bears and, through her labor, delivers her baby into the world (Rooks 1997, p. 130).

IS THE OBSTETRIC APPROACH "SAFER"

Over the last couple of decades several studies have argued that the move of childbirth from home to hospital has not necessarily improved the outcomes for low risk pregnancies, which are the majority of all pregnancies. The reason, so it is proposed, is today's prevalent medical model of maternity care which focuses on obstetric intervention. For low-risk women these interventions are considered not just unnecessary but also counterproductive to positive outcomes of childbirth (Kloosterman 1984; Tew 1990; World Health Organization 1985b).

We will first look at the "safety" argument, or the issue of preventing perinatal mortality, which was and is at the core of the medicalization of childbirth (Zander 1984). We will then compare both approaches with respect to their use of obstetric interventions and summarize the present understanding of the effectiveness of these interventions. Finally, in a short excursion, we will evaluate the predictability of low risk pregnancies, and see whether predictability is a necessary condition for the natural approach to childbirth.

For the purpose of this study we assume that in the highly industrialized countries, particularly in Western Europe, the risks of childbirth and the available obstetric technologies are similar to those in the United States - although the population demographics as well as the application of these technologies might vary among these countries. We furthermore assume that in all these countries midwives are constrained to a non-interventionist approach to childbirth while the obstetric profession adheres to the general philosophy of the interventionist obstetric model. Based on these assumptions, we include literature and studies from the U.K., the Netherlands, Australia, and Scandinavia, along with the material from the United States, as relevant for identifying the issues. The proposed research design for the United States data will benefit from this inclusion since alternative approaches to childbirth are more prevalent in these countries than in the United States.

PERINATAL MORTALITY OF HOSPITAL BIRTH

MIGHT NOT BE SUPERIOR

Since the turn of the century the birthplace for children in the United States has moved from home to hospital. Less than 5% of the babies born in 1900 were delivered in the hospital and today it is close to 99% (Declercq 1993). Over the same period the maternal and neonatal mortality has declined dramatically (Institute of Medicine and National Research Council 1982; Tew 1990). This relationship of increasing hospitalization and decreasing mortality rates has been interpreted as causal, and advances in obstetrics as well as the increased obstetric interventions are considered responsible for the improved outcomes for mothers and babies (Institute of Medicine and National Research Council 1982). But a closer look at the data suggests that we might not have a causal relationship here. McKeown (a physician by training) has provided a detailed and convincing analysis of the contribution of medicine to the decline of general mortality, concluding that the decline in the second half of the 19th century and up to 1971 was due to rising standards of living (mainly diet), improvements in hygiene, and generally living in a healthier environment (reduced exposure to infection). There was a small effect from immunizations. A significant effect from medical therapies, however, could not be identified with certainty (McKeown 1976). McKinlay and McKinlay come to a similar (although tentative) conclusion for a preliminary analysis of general mortality data for the United States (McKinlay and McKinlay 1990).

The United States, The Netherlands, and Britain experienced dramatic reductions in maternal mortality starting in the mid-1930s but had very different systems of maternity care. In the United States, obstetricians increasingly delivered most of the babies in hospitals. In the Netherlands professional midwives delivered most of the babies in homes. Britain had both home and hospital births and used midwives, general practitioners, and specialist obstetricians. These differences had no apparent effect on the decline of the maternal mortality rate (see Maine, 1991 as quoted in Rooks 1997, p. 31).

Comparable analyses for the decline of perinatal mortality have not been done, but the same factors which account for the decline of the general and maternal mortality are probably also responsible for the decline of perinatal mortality. Tew analyzed perinatal mortality and hospitalization in England and Wales for a recent period (1969-1981) and had to conclude that

... if hospitalization had increased proportionally less, perinatal mortality would have fallen proportionally more (Tew 1990, p. 263).

In other words, the increase in hospitalization slowed down the already observable decline in perinatal mortality. While such statements do raise a red flag, we need to remember that these relationships are associations, not causal relationships.

While obstetric intervention is undoubtedly highly beneficial for high-risk pregnancies, its benefits for low-risk pregnancies, which constitute by most definitions 60-80% of all pregnancies (Selwyn 1990), has never been proven. Tew points out that during this move of birth from home to hospital

...no-one in any country ever set out to test the hypothesis that hospitalizing birth made it safer (Tew 1990, p. 26).

Randomized trials did not justify the hospitalization of childbirth in the beginning, nor were the outcomes systematically compared later (Campbell and Macfarlane 1987; Tew 1990). While comparative observational studies generally found higher perinatal mortality rates for hospitals, the available data did not allow comparisons of women of similar risk categories, and so these findings were plausibly explained byc the significantly higher risk profile of the hospital population (see the 1958 and 1970 surveys of the Royal College of Obstetricians and Gynaecologists: Butler and Bonham 1963; Chamberlain et al. 1978; Chamberlain et al. 1975). However, several observational studies carried out during the last two decades suggest that out-of-hospital birth is as safe as hospital birth for women with comparable low-risk profiles (Kloosterman 1984; Mehl et al. 1976; Tew 1977b; Van Alten, Eskes and Treffers 1989). This raises the question whether the medical model of hospital-based obstetric intervention has actually improved the outcomes of childbirth for low-risk women in the U.S.

Since the mid-seventies, there has been a growing body of literature on the safety aspects of different approaches to childbirth (using birth settings as the proxy). The U. K. had taken the lead in this research, partly due to her relatively late move toward hospitalization, but mainly because of the discussions within the National Health Service (NHS) which needs to periodically decide the extent of resources that will be allocated to out-of-hospital birth or whether to pursue a policy of covering only hospital birth. To inform this discussion, the National Birthday Trust, a charity, sponsored two major epidemiological surveys of the Royal College of Obstetricians and Gynaecologists in 1958 and in 1970. Both reports (Butler and Bonham 1963; Chamberlain et al. 1978; Chamberlain et al. 1975) concluded that advancements in obstetrics make hospital confinement mandatory. The fact that the data showed higher perinatal mortality rates for obstetric hospitals as compared to general practitioner units (GPUs) or home deliveries was explained by the higher risk profile of the hospital population.

It was Marjorie Tew5 who initiated the debate about safety of alternative birth settings by re-analyzing the published and unpublished data of these two surveys. She showed that in both studies birth in obstetric hospitals was significantly less safe than in general practitioner units or home birth, after controlling for risk. In the 1958 survey, general practitioner units and home (GPU/home) showed a significantly (p<0.0005) lower perinatal mortality than the hospital for every level of risk (Tables 1, last two columns).

The 1970 survey used an antenatal and a labor prediction score, where the labor prediction score includes the antenatal prediction score (Tables 2, 3).

Again, for each level of risk we find a significantly lower perinatal mortality for GPU/home in 1970 (Tables 4 and 5, last two columns). The risk factors employed in these surveys reflected the obstetric profession's screening criteria at that time. They are not a complete list by today's standards (see Goodwin, Dunne and Thomas 1969; Keirse 1982), but they are congruent with those factors considered important today.

A surprising finding was that birth at home and in General Practitioner Units (GPU) was not only safer for low-risk pregnancies, but also for the high-risk cases (Tew 1990, pp. 241-245). However, Tew's analysis is based on the actual place of birth and not the planned place of birth, because the data did not allow this distinction. This means that the home birth category includes unplanned home births as a result of precipitous labor. These births have a larger share of preterm and low birthweight babies with presumably higher mortality. On the other hand the intrapartum transfers from home to hospital are included in the hospital category, and these births have a higher rate of complications and presumably a higher mortality. We do not know whether the effects of these two subgroups cancel each other out, or whether the mortality of the transfer group is so much larger than the mortality of the unplanned home birth that it is solely responsible for the overall higher mortality of the hospital group, as found in the two British surveys. Tew's critics (Campbell and Macfarlane 1986; Campbell and Macfarlane 1987) point out that her retrospective observational analysis design does not allow her findings to be conclusive, but they also point out that there is no evidence that hospital deliveries are safer. There is a large body of literature in the U.K. surrounding these findings and the safety debate (Alberman 1984; Alberman 1986; Campbell and Macfarlane 1990a; Campbell and Macfarlane 1990b; Campbell and Macfarlane 1990c; Campbell, Macfarlane and Cavenagh 1991; Chalmers 1978; Chamberlain 1984; Chamberlain 1988b; Chamberlain and Zander 1992; Kitzinger and Davis 1978; Zander 1984; Zander and Chamberlain 1984), but no major new studies have been done6. The safety debate in the U.K. was until recently characterized by what Luke Zander called

... the reluctance of the obstetric establishment to consider the implications of objective evidence which runs counter to their preconceived assumptions -- without refuting it on statistical grounds (Zander 1984, p. 128).

Another early participant in this safety debate was the Regional Office for Europe of the World Health Organization (WHO). The WHO commissioned in 1979 a Perinatal Study Group to examine the "problems surrounding birth and birth care" and organized a conference in 1985 in Fortaleza, Brazil on this topic. The recommendations (World Health Organization 1985a) and the report (World Health Organization 1985b) strongly argue for a non-interventionist approach to childbirth. The WHO was particularly adamant about the safety issue and worried that the spread of hospitalization for childbirth to the developing countries, with the resulting increase in inappropriate interventions, might be detrimental to outcomes [World Health Organization (WHO), 1985 #6].

For the Netherlands, as the only country with a sizable proportion of natural childbirths (home birth as proxy), a few comparative studies exist. Treffers looked at the relationship of perinatal mortality and the degree of hospitalization across regions and could not find any correlation (Treffers and Laan 1986). This led him to conclude that in a homogeneous population there is no evidence that hospitalization for childbirth improves perinatal mortality. Scherjon compared perinatal outcomes between Denmark and the Netherlands and concluded that the higher rate of hospitalization in Denmark did not result in better perinatal outcomes than the considerably lower rate in the Netherlands (Scherjon 1986). Tew and Damstra-Wijmenga analyzed Dutch national perinatal statistics from 1986 and found that perinatal mortality rates were much higher for obstetricians in hospitals than for midwife-attended home care or midwife-attended hospital care, at all levels of risk when controlling for gestation, maternal age and parity (Tew and Damstra-Wijmenga 1991). But as with Tew's analysis of the British data, these Dutch data do not allow the identification of intrapartum home to hospital transfers or unplanned home birth due to precipitous labor. They do not compare planned hospital birth with planned home birth; rather, their comparison is based on the actual place of birth.

A well designed Swiss study of matched pairs of deliveries planned for home versus deliveries planned for hospital showed the same perinatal mortality (one case) in each group, but the total population of 874 cases is too small for a statistically meaningful comparison of perinatal mortality (Ackermann-Liebrich et al. 1996). We will look at other outcome measures from this study later.

A recent meta-analysis of planned home birth vs. planned hospital birth of studies published after 1970 found six studies (from Australia, Netherlands, Switzerland, UK, two from the US) which met the selection criteria and concludes that

Perinatal mortality was not significantly different in the home and the hospital groups in any individual study, nor was the difference statistically significant in the pooled analysis. The pooled estimate for the odds ratio was 0.87 (0.54-1.41) in favor of home birth (Olsen 1997, p. 6).

While Macfarlane, in a very thorough commentary, discusses the limitations of this meta-analysis due to the considerably different study designs, structure of delivery systems, or observation periods, she acknowledges that "the material assembled makes this review a valuable contribution" in the search for the "safest place of birth" (Macfarlane 1997, p. 16).

Due to its early move toward hospitalization, the United States has long had a small number of planned home births. The increase of home births during the late 60s and the opening of free-standing birthing centers in the 70s led to several observational studies which, however, did not start a debate about safety (Mehl 1978). The establishment of the birth centers, however, raised concerns that births were increasingly occurring in out-of-hospital settings and led to the formation of the Committee on Assessing Alternative Birth Settings by the Institute of Medicine and the National Research Council in 1980. The committee was charged with determining which methodologies were needed to evaluate the current childbirth settings in the U.S.. The Committee made detailed suggestions on how to answer this question (Institute of Medicine and National Research Council 1982). A direct result of the Committee's work was the National Birth Center Study which analyzed approximately 12,000 births covering the majority of birth centers existing in the United States at that time, but which unfortunately did not have a direct control group, relying for comparisons on studies of low risk pregnancies in hospital settings (Rooks et al. 1989).

In her recent book, Rooks (for a detailed discussion see Rooks 1997, Chapter 11) subjects all those retrospective and prospective observational studies from the United States (that were large enough to address perinatal mortality for out-of-hospital births and that allowed for some sort of control group) to a thorough analysis (Anderson and Murphy 1995; Mehl-Madrona and Madrona 1997; Rooks et al. 1989; Sullivan and Beeman 1983; Tyson 1991). While all these studies have their shortcomings, she concludes that

... there is substantial evidence that birth attended by either midwives or physicians in homes and free-standing birth centers in this country can be as safe as in-hospital birth when a system of quality assurance is in place (like competency, proper screening, proper and accepted transfer protocols, reliable obstetrical back-up) (Rooks 1997, pp. 383 and 465)

Dr. C. Arden Miller, chairman of the Department of Maternal and Child Health at the University of North Carolina School of Public Health and a past-president of the American Public Health Association, testified in a 1980 hearing on the obstacles to nurse-midwifery practice. He stated, "If one looks for reasons why this country is deprived in many areas of the services of midwives, one has to look in the political and economic arenas. The answer is not to be found in terms of health outcomes. All of the studies I know confirm that the health benefits rendered by nurse-midwives stand up to scientific scrutiny exceedingly well" (House Committee on Interstate and Foreign Commerce, 1980, as quoted in Rooks 1997, p. 297).

We have to conclude that the evidence from previous studies, although limited, suggests that for low-risk women there is the possibility that the obstetrician-managed hospital birth might not be safer, i. e. might not have a lower perinatal mortality, than the non-interventionist natural approach to childbirth.

OBSTETRIC INTERVENTIONS: NECESSITY AND EFFECTIVENESS

The obstetric approach is characterized by a set of obstetric interventions aimed at improving the safety for mother and child. While these interventions do have their place in those cases where certain complications occur and when judiciously applied (Goer 1995, p. 3), it is questionable whether their "routine" application improves the outcome for mother and child in the case of low risk pregnancies. We will first see whether the use of these interventions for low risk women differs significantly between the natural and obstetric approach to childbirth. Then we will briefly look at what is known today about the effectiveness of these interventions. Finally, we will present the argument why, for low risk women, these interventions might be counterproductive to good outcomes.

Differences in Obstetric Care: Are the Interventions Necessary?

When comparing the natural to the obstetric approach with respect to obstetric interventions, it is important to ensure that we are comparing women of similar risk profiles. If we find, in well controlled studies, significant differences in the use of obstetric interventions between the two approaches, but no differences in outcome, we will conclude that there may be cases where such obstetric interventions are not necessary to achieve good outcomes.

In a double blind clinical trial at the Los Angeles County and USC Women's Hospital, 492 low risk women who qualified for the hospital's Normal Birth Center were randomly assigned to either the midwifery service in the birth center or to the physician service in the maternity ward (Chambliss et al. 1992). While there were no differences in the demographics of the two groups or in neonatal outcome, the physicians had significantly higher intervention rates than the midwives (see Table 6).

It should be noted that the Normal Birth Center did not use any oxytocin or epidurals. The 10% intrapartum transfers from the birth center to physician management did receive oxytocin or epidurals only at the maternity ward (though the transfers are counted in the midwifery group). The difference in operative deliveries (forceps, vacuum, cesarean) was due to the large difference in forceps and vacuum deliveries (18 cases for physicians versus zero cases for midwives), while the difference in cesarean sections was not significant (1 case for physicians versus 5 cases for midwives). This study showed that for low risk pregnancies the physicians in this large teaching hospital achieved the same low cesarean section rate as the midwives in the in-hospital birth center7. It should be noted that the physician service, managing a very heavy load of mainly Hispanic women, had, by American standards, a rather low overall cesarean section rate of approximately 9% prior to this study. The cesarean section rate of 2.1% for the midwifery service in this study was consistent with the 2% that the service had experienced in the past.

A similar but retrospective matched pairs study compared low-risk women, who qualified for the university hospital's nurse-midwifery service, with matched cases managed by physicians in the hospital's maternity ward (29 matched pairs, delivered in spring of 1985) (Mayes et al. 1987). There were no differences in complications or maternal or neonatal outcomes, although the sample was too small to statistically identify outcome differences. Cesarean sections were excluded. The physicians had significantly more interventions than the midwives (Table 7). Since the women were not randomly assigned, the few differences in the groups risk profiles (smoking, marital status, occupation) could represent selection bias and be responsible for some of the differences in interventions.

A small (148 cases) British randomized controlled trial compared an in-hospital birthroom with the labor ward in the same London maternity hospital (Chapman et al. 1986). The same group of community midwives cared for the women in both settings. The difference was the setting and the fact that neither epidural analgesia nor continuous electronic fetal monitoring was available in the birthroom. The study sample was composed of low-risk women who qualified for the birthroom and had consented to randomization8.

With no differences in neonatal outcomes, the labor ward group had significantly higher rates of

- Analgesia (excluding epidural). While all women entered the study with the understanding that they might be allocated to the birthroom where epidurals would not be available, six women in the labor ward group had epidurals.

- Suturing, suggesting more severe perineal injuries (no difference in episiotomies).

Satisfaction with the birth experience was clearly higher among those women in the birthroom group. While the birthroom group unanimously wanted their next baby again in the birthroom, only five of the labor ward group wanted another delivery in the labor ward (three of whom had had an epidural). This study does not compare obstetricians with midwives. Instead it shows that the same group of midwives, operating in a different environment with slightly different protocols, was more prone to intervene in the obstetric labor ward setting than in the birthroom setting.

In another randomized controlled trial at the Glasgow Royal Maternity Hospital 648 healthy women were assigned to midwife-managed care and 651 women to shared care (care divided among midwives, hospital doctors, general practitioners) (Turnbull et al. 1996). There were no differences in fetal or neonatal outcome. The midwife-managed group had significantly fewer inductions and electronic fetal monitoring (EFM) and a lower rate of episiotomies, which was not offset by a greater frequency of perineal tears. There was no difference in pain relief used or in augmentation of labor. Midwife-managed women experienced significantly less antenatal hypertension and antepartum hemorrhage. They also expressed significantly greater satisfaction. Again we observe that although the players in the system remain the same, a small shift in control alone, from shared care to midwife-managed care, leads to significantly fewer interventions and better maternal outcomes.

A retrospective observational study compared practice patterns of obstetricians9 to nurse-midwifery care in Washington State during 1988-89 (Rosenblatt et al. 1997). While providers were picked randomly from the urban Washington State population of the respective provider category, the patients could not be randomized, i. e. the patients had chosen their provider and were not randomly allocated to providers. For each provider a random sample of his patients was selected. The investigators restricted the study to women who were low risk at entry into obstetrical care and equivalent with respect to their medical, obstetric, and sociodemographic risk factors (total of 1,322 cases). There were no significant differences with respect to birth outcomes. The obstetricians had significantly higher intervention rates (Table 8, where we only compare obstetricians to midwives and have excluded family practitioners for the purpose of this analysis). The authors argue that although they "were unable to randomize patients, it is unlikely that unmeasured differences in patients caused all the observed differences in practice style" (Rosenblatt et al. 1997, p. 350).

All the above studies compare the natural approach as managed by certified nurse-midwives to the obstetric approach. All these certified nurse-midwives were trained and were providing care in a hospital setting and culture. It is safe to assume that their approach to natural childbirth is still heavily influenced by the obstetric protocol. But even when comparing "obstetrically oriented" certified nurse-midwives to obstetricians we can see clear differences with respect to interventions. Studies of midwives providing care in free-standing birth centers or in the home show considerably lower intervention rates. There are, however, no such studies with patient randomization or satisfactory controls and the observed differences may be in part due to selection bias.

A Swiss prospective cohort study with matched pairs provides a glimpse into the homebirth setting - in a country with 99% hospital birth (Ackermann-Liebrich et al. 1996). A group of 214 women planning home delivery during 1989-92 were matched on medical, obstetric, and sociodemographic risk factors to a group of 214 women planning birth in a hospital. There were no differences with respect to complications or neonatal outcomes10. The hospital group had significantly more interventions (Table 9).

The authors conclude that while the study does not have sufficient power to exclude possible differences in rare complications/events, the lower rate of interventions in home births meant a lower risk of subsequent complications for the mother. They do point out that their matching could not control for differences in attitudes: "Usually it takes a certain tenacity for women to realize a home birth in a health care system in which this is considered irresponsible" (Ackermann-Liebrich et al. 1996, p. 1317).

Finally, let us try to construct a comparison that looks at a truly natural childbirth protocol and includes women of all risk and thus avoids the issue of randomization or controls for risk factors. We discussed earlier the natural childbirth approach as practiced by the French physician (surgeon) Odent in Pithiviers. His public hospital accepted all women in the region without screening out any high risk pregnancies. While keeping all births in the hospital, they changed the hospital setting and the protocol dramatically. The delivery room became a birthing room with adjacent hot tub. Their concept of "the undisturbed birth" ruled out any oxytocin for induction or augmentation of labor. The protocol had no place for routine or elective analgesia or epidurals. There was no use of forceps, just vacuum extraction. When the midwives could not manage the complications, the surgeon Odent stepped in. Since this approach could theoretically be transferred to any hospital in the United States, it seems fair to compare Pithiviers' intervention rates and results with data for the United States hospitals (Table 10). Pithiviers' outcomes match the United States hospitals' outcomes. However, comparing the zero or low intervention rates from Pithiviers with the high and still rising intervention rates for United States hospitals, one can only question whether the vast majority of obstetric interventions in the United States is necessary.

From all the above data we conclude that there seems to be sufficient evidence to suggest that when low-risk women in the United States choose to have their birth managed by an obstetrician/physician, they subject themselves to a significant probability of unnecessary obstetric interventions. The next chapter suggests that these interventions might not just be unnecessary but to a large degree ineffective or harmful.

Effectiveness of Interventions

The following brief overview of the effectiveness of the major obstetric interventions is based on the findings from the Cochrane Pregnancy and Childbirth Database as published in A Guide to Effective Care in Pregnancy and Childbirth (Enkin et al. 1995), and Goer's book, Obstetric Myth Versus Research Realities (Goer 1995), an excellent review of the obstetric research literature. The Cochrane Pregnancy and Childbirth Database11 is a major research effort which originated out of the British safety debate and led in the late 1970s to the formation of the National Perinatal Epidemiology Unit (NPEU) in Oxford. A group of leading obstetricians, started by Iain Chalmers, are collecting (on an ongoing basis) the results from empirical investigations in obstetrics in this database, and are subjecting them to meta-analysis. The American statistician Mosteller calls this effort12

The most advanced current example of a basis for practicing medicine founded on both empirical evidence and theory (Mosteller 1993).

Enkin et al. have grouped the interventions into six categories (see Table 11) working from two basic principles:

Firstly, that the only justification for practices that restrict a woman's autonomy, her freedom of choice, and her access to her baby, would be clear evidence that these restrictive practices do more good than harm, and secondly, that any interference with the natural process of pregnancy and childbirth should also be shown to do more good than harm. We believe that the onus of proof rests on those who advocate any intervention that interferes with either of these principles (Enkin et al. 1995, p. 389).

While the Cochrane project analyzes obstetric interventions from screening and antenatal care to postpartum care on a very detailed level, we have limited our overview to the more aggregated categories of major interventions as discussed in the previous section. This required us to group together some of their more narrowly specified interventions and draw some inferences from the text regarding how they match up with our more aggregated major intervention categories. We also present the Cochrane project's evaluation of some frequently used obstetric procedures, some of which were not covered in the studies in the previous chapter. We observe that, with the exception of post-term induction (>41 weeks), none of the obstetric interventions falls into the clearly beneficial or likely beneficial categories (Table 11). We do see two components of natural childbirth protocol, vacuum extraction instead of forceps and mobility and choice of position, in the clearly beneficial category though.

Following are the key excerpts from Goer's (1995) evaluation of the effectiveness of these obstetric interventions.

Induction, postdates: Little evidence that modern postdates management offers benefits and considerable evidence that it does not (ibid., p. 182). Routinely inducing creates more problems than it solves. Surveillance testing in order to induce selectively introduces risks. Letting nature take its course is generally best, although that is not risk free either (ibid., p. 183).

Induction, after prelabor rupture of membranes at term: Major benefits for expectant management (watching and waiting) in women with no signs of infection (ibid., p. 204). Inductions lead to vaginal exams, internal monitoring, fetal distress, and cesarean sections, which themselves have much higher infection rates (ibid., p. 205).

Augmentation, early use of oxytocin, active management of labor: The indiscriminate use of oxytocin does no good and may harm the baby (ibid., p. 87). Active management is not needed at all (ibid., p. 86). Mobility and upright positions (avoidance of supine position) seem to improve labor progress [more] (ibid., p. 88).

EFM: Compared with intermittent auscultation, EFM benefits neither high-risk nor low-risk infants. It increases the odds of cesarean or instrumental delivery (ibid., p. 135).

Epidural anesthesia: Epidurals substantially increase the incidence of oxytocin augmentation, instrumental delivery, cesarean for dystocia, the need for episiotomies (ibid., p. 279), and bladder catheterization (ibid., p. 253). They cause maternal complications (ibid., pp. 254-255), abnormal fetal heart rate, and may cause adverse neonatal physical and behavioral effects (ibid., p. 255). Innovations in procedure have not decreased epidural-related problems (ibid., p. 254).

Episiotomy: Routine episiotomy serves no medical purpose (ibid., p. 278). Episiotomies do not prevent tears, heal better, or are less painful than tears. They do not prevent birth injuries nor fetal brain damage. They increase blood loss and may lead to infections (ibid., p. 279). Other techniques have been found effective for reducing perineal trauma (ibid., p. 280).

Cesarean section: Cesarean rates [in the United States] depend on factors having nothing to do with medical indications. These factors are: individual philosophy and training, convenience, the patient's socioeconomic status, peer pressure, fear of litigation, and possibly financial gain. In many cases of cesarean section, the true culprit was obstetric management practices [unnecessary interventions], not medical indication. Far from resulting in better perinatal outcomes, babies born by cesarean fare worse. Cesarean section substantially increases maternal risk (ibid., p. 25).

The Cochrane project was initiated by obstetricians. The main initiators, however, were very open minded obstetricians skilled in epidemiological approaches to evaluation of care practices and the project is characterized by a very objective approach. Goer's values are rooted in the natural childbirth movement. Her excellent abilities in abstracting the obstetric research papers cuts through the strengths and weaknesses of these papers and isolates the substantiated pros and cons in a very systematic way.

It is interesting to observe that both these research efforts come to more or less the same conclusions with regard to the effectiveness of obstetric interventions. The Cochrane project uses more guarded wording so as not to alienate the obstetric readership. Goer takes a more aggressive stance in light of the apparent disadvantages of the obstetric approach and her sense of urgency for needed change.

Against this critical evaluation of the effectiveness of obstetric interventions, Enkin summarizes the Cochrane project's position as follows:

As technical advances became more complex, care has come to be increasingly controlled by, if not carried out by, specialist obstetricians. The benefits of this trend can be seriously challenged. Direct comparisons of care given by a qualified midwife with medical back-up with medical or shared care show that midwifery care [the natural approach to childbirth] was associated with a reduction in the range of adverse psychosocial outcomes in pregnancy, and with reductions in the use of acceleration of labour, regional analgesia/anaesthesia [epidural], operative vaginal delivery, and episiotomy. No differences have been demonstrated in the rates of labour induction, pharmacological analgesia, or caesarean section. Midwifery care also resulted in fewer babies weighing less than 2500 grams, needing resuscitation, or needing admission to special care units.

and he concludes:

It is inherently unwise, and perhaps unsafe, for women with normal pregnancies to be cared for by obstetric specialists, even if the required personnel are available (Enkin et al. 1995, p. 15).

While we have at this point looked very narrowly at the medical benefits of obstetric interventions, Rooks reminds us that there is more to these obstetric interventions than just medical consequences.

In addition to the known risks associated with each procedure, their use affects the intricate physiologic dynamics of labor, the environment in which women labor and give birth, and, ultimately, the experience and meaning of labor and childbirth for the mother. (Rooks 1997, p. 311).

How Obstetric Interventions Can Be Counterproductive to Good Outcomes

Odent considers the physiology of the process of delivery to be an intricate interplay of the mother's state of consciousness, the conditions of optimal secretions of oxytocin and endogenous opiates, and maternal postures (Odent 1984b). Labor and delivery seem to be made "easier by reducing the control exerted by the upper brain [neocortex]" (ibid., p. 78b) in favor of the body brain [hypothalamus], and allowing the women to shift to a deeper level of consciousness. "Given the right kind of environment [semi-dark small place, little noise, few people, free mobility, warm water] - where the woman can feel free and uninhibited - she can naturally reach [an instinctive] level of response deeper within her than individuality, upbringing, or culture" (Odent 1994, p. 13). It "seems evident [to him] that the instinctive state that enables women to labor spontaneously is connected with a particular hormonal balance" (ibid., p. 14) and that the "fluctuations of hormonal secretion during labor cannot be dissociated from the state of consciousness" (Odent 1984b, p. 79).

"It has also become more and more obvious that endorphins play an important role in the complex hormonal equilibrium that makes a spontaneous delivery possible. Neurohormones with morphinlike functions, these "endogenous opiates", act as natural painkillers, not only protecting against pain but also suppressing anxiety and inducing a general feeling of well-being" (Odent 1994, p. 14).

The third area of concern is maternal postures.

"We know, for example, that some postures can trigger a change of the consciousness level [e. g. praying position on knees]. Conversely, reducing neocortical control and liberating the instinctive brain induces the capacity to find more physiological postures spontaneously. In [his] view this capacity for finding a physiological posture is an expression of a successful regression; so, too, are the capacity for screaming without restraint and the capacity for opening one's perineum. An obstetrical posture may be considered as physiological when there is no compression of the vena cava, when gravity is employed at least in an intermittent fashion, and when the pelvic skeleton becomes fully mobilized" (Odent 1984b, p. 80).

If we assume this to be an adequate description of the physiology of the labor and delivery process, then we can see immediately how today's protocols with routine or prophylactic obstetric interventions may lead to a vicious cycle with undesired bad outcomes.

The bright and foreign hospital environment with new faces moving in and out whenever the shift changes, with limited space for moving around and the supine or lithotomy position with legs in stirrups the preferred birthing position, with narrowly defined ranges of what constitutes normal progress and what requires intervention, with "birth by the clock" or active management of labor, attached to electronic devices and with everything around her reminding the woman constantly that she is a candidate for emergency delivery - all that seems hardly to be an environment supporting the woman to "shift to a deeper level of consciousness". Or as Odent reminds us, "We have to accept that we are mammals and that all mammals have some particular strategy so as to not feel observed when giving birth" (Odent 1994, p. xiii).

And then induction of labor starts the vicious cycle, which Rooks (1997, p. 313) describes as follows:

"Induction forces a very complex system into action before it is really ready to begin. When labor starts spontaneously, some of the changes that need to occur during the process happen before the woman is aware of being in labor. When labor is forced through the use of oxytocin, this preparation is not in place. As a result, more labor is needed to accomplish the work. The various changes may not take place in the right sequence; for example, the uterus may begin to contract strongly while the cervix is still thick and closed. In addition, the individual contractions tend to arise more sharply; this makes them more painful, and harder for women to cope with. There is always an increased risk of fetal distress with use of oxytocin; EFM is necessary, and an intravenous infusion is required, [both interventions now limiting the woman's mobility severely]. The proportion of all U.S. births in which labor was induced increased from 9.0 percent in 1989 (NCHS, 1992) to 16.9 percent in 1996 (Ventura et al. 1998)".

Augmentation of labor through intravenous infusion of synthetic oxytocin (pitocin) is used to expedite labor if progress is deemed to be inadequate. And again the interventive cycle is started: EFM is necessary, EFM wires and IV limit the woman's mobility, oxytocin increases pain of contractions (stronger and more frequent contractions), leading to administration of analgesia or epidural, which in turn reduce or blot out feelings, the intuitive sense for contractions, and depresses the pushing reflexes (Armstrong and Feldman 1990, p. 43), leading to increased cesarean sections or forceps delivery. There is an approximately five-times-increased risk of having a forceps delivery when a woman has had an epidural (Kitzinger in her foreword to Odent 1994, p. xvii). And epidural and forceps are accompanied by episiotomy and the following suturing.

All these interventions severely curtail a woman's mobility and she typically ends up in the traditional lithotomy position which "works against the physics and physiology of childbirth" (Armstrong and Feldman 1986, p. 111). It is "in fact the worst possible alternative physiologically for both mother and child: when a woman lies on her back, the enlarged uterus compresses the major blood vessels [vena cava], which in turn diminishes the amount of oxygenated blood entering and leaving the placenta. In addition, such a position makes it impossible for a woman to take advantage of gravity to facilitate delivery" (Odent 1994, pp. 9-10). In addition, women are prohibited from eating or drinking during labor, reducing their energy and strength when it is most needed (Armstrong and Feldman 1990, pp. 40-41).

The WHO study group of childbirth in Europe laid out such a chain of diagnostic and therapeutic interventions (Figure 1) and concluded

In Europe we are still far from determining what is appropriate technology for birth. Not only have individual techniques come into widespread use without proper evaluation, but there is growing evidence that at present there is an increasing involvement in chains of technological diagnostic and treatment interventions. Such a chain can be visualized as shown in Fig. 3. In the quest for appropriate technology for pregnancy and birth and following birth, it is necessary not only to evaluate individual techniques prior to use but also to take a broader approach in evaluating the way in which events and procedures interact (World Health Organization 1985b, p. 97).

Odent, being aware of the complex neurohormones driving the birthing process, concludes that

in order for the body's natural powers to come into play, they must be left alone. To give women painkilling drugs and synthetic hormones (artificial oxytocin) during birth, as is common practice in most modern hospitals, will destroy the hormonal balance on which spontaneous labor depends. Certainly pain itself can slow down labor, but when drugs are not used, the body can defend itself effectively and naturally against it. Indeed, it has been found that the longer and more difficult a woman's labor, the higher her endorphin level will be ... Overall, the existence of these complex neurohormones confirms our belief in the interconnection of all aspects of sexual life, and - since the balanced secretion of hormones is a delicate matter, highly responsive to external conditions and psychological states - presents another powerful argument against medical interference in, or disturbance of, the physiology of the labor process (Odent 1994, p. 15).

EXCURSION: PREDICTABILITY OF LOW-RISK PREGNANCIES

An important question in the 'safety' debate is whether it is possible to identify in a prospective fashion a low-risk group of women, for whom birth with a natural childbirth caregiver, be it in a hospital or in an out-of-hospital setting, does not increase the odds of a negative outcome. For out-of-hospital settings there will, of course, always be the need for unexpected emergency transfers to the hospital during the intrapartum. The comparison of outcomes for the out-of-hospital natural childbirth approach to the in-hospital obstetric approach will have to include the outcome for these transfers into the statistics of the natural childbirth group. The question then is, do obstetric risk assessment instruments have a sufficiently high predictive power to distinguish low-risk (including the transfer cases) from high-risk pregnancies.

It should be noted that risk assessment is a dynamic process with accuracy increasing with additional information over time. A crucial point is the onset of labor, at which time a decision not to proceed with the planned out-of-hospital birth but to transfer to the hospital does not create any additional transfer risks. New information during the intrapartum (or postpartum) may require a transfer. While the majority of well-managed intrapartum transfers do not pose any additional risks (Springer and van Weel 1996, p. 1277), there are emergency intrapartum transfers which clearly increase the risks for planned out- of-hospital births.

Obstetric risk assessment instruments are used to assign women to the appropriate level hospital in regionalized systems, to choose women for alternative birth settings and alternative providers with low intervention approaches, and to assist in clinical decision making about necessary testing or interventions (Knox et al. 1993; Selwyn 1990). A variety of obstetric risk assessment instruments have been developed (Creasy, Gummer and Liggins 1980; Edwards et al. 1978; Ernest et al. 1988; Goodwin, Dunne and Thomas 1969; Halliday, Jones and Jones 1980; Hobel et al. 1973; Holbrook, Laros and Creasy 1989; Kaminski and Papiernik 1974; Knox et al. 1993; McCarthy, Schulz and Terry 1982; Rumeau-Rouquette, Kaminski and Goujard 1974; Sokol et al. 1977), and although there is a core of recurring risk variables there does not seem to be a standard set of variables. Multivariate analysis seems to suggest that the set of predictive variables also depends on the specific outcome to be predicted (like perinatal mortality or morbidity, preterm labor/delivery, preterm or term low birthweight, intrauterine growth retardation) (Selwyn 1990). Another issue is where to set the cut-off or threshold for "high risk" designation. Decreasing the threshold will increase the instrument's sensitivity, but will decrease its predictive power (Selwyn 1990) and could easily end up in screening everybody into the "high risk" category. Finally, is obstetric risk assessment more like screening in parallel, where one administers several tests and those positive in any one of them screen positive, or does "high risk" designation require several factors or certain combinations of factors to test positive?

Like all (medical) decision making, obstetric risk assessment is an interactive process between the decision-maker with his or her experience and the instrument. For the purpose of this study we need to focus on the outcome of this interactive process, i. e. can we find situations where risk assessment instruments have been widely and successfully employed? After briefly looking at the predictive accuracy of some smaller scale applications of risk assessment instruments, we will then review the Dutch system which still has two distinctly different approaches to childbirth and thus requires a systematic and nationwide applied allocation procedure. We will close this part of our review with a brief look at the experience from the National Birth Center Study (NBCS) for the United States.

Predictive Accuracy of Obstetric Risk Assessment Instruments

Herman et al. and Selwyn have analyzed the major risk assessment instruments13 with respect to their predictive accuracy (Herman, Irwig and Groeneveld 1988; Selwyn 1990). We will here focus on those systems predicting perinatal mortality (versus prediction of low birth weight or pre-term birth). Herman et al., using receiver operating characteristic (ROC) curves14 for comparing different scoring systems, show that the Goodwin system (Goodwin, Dunne and Thomas 1969) achieves a very high predictive accuracy of 0.95 (Herman, Irwig and Groeneveld 1988, p. 837). They observe that the inclusion of obstetric history and late antenatal and intrapartum risk factors increases predictive accuracy. As to the weighting of risk factors they suggest that odds ratios obtained from multiple logistic regression be used. Most of the analyzed systems, including Goodwin's system, use experience based clinical weights. It should also be noted that to derive stable scores for perinatal mortality one needs rather large patient populations, which were probably not available in all of the analyzed studies (Knox et al. 1993, p. 198)

Knox et al., building on Herman's findings, developed a scoring system based on logistic regression analysis (Knox et al. 1993). Application of the system at the onset of labor resulted in a predictive accuracy of 0.9115. Eighty-seven percent of poor outcomes (defined as perinatal mortality or >5 days in neonatal unit) were accurately identified when choosing an allocation threshold of only 16% of the women to the high-risk group. Analyzing the poor outcomes in the low-risk group, the 13% false negatives, they found that most were unpredictable, with 55% of those cases due to congenital anomalies. This leaves 5.9% of preventable poor outcomes not predicted by the scoring system when using a high-risk threshold of 16% of the pregnancies.

These findings suggest that obstetric risk assessment instruments with very high predictive accuracy already exist and that there are very specific suggestions about how to improve these instruments. However, given the predominance of obstetrician managed hospital birth in the United States (as well as in the other developed countries), none of these systems has been used on a large scale for identifying low-risk pregnancies and allocating them to a different obstetric approach than the high-risk pregnancies. The only developed country with a high share of natural childbirth is the Netherlands, and it is therefore illuminating to see how the Netherlands manages this allocation process.

The Dutch 'Indication List'

In the Netherlands home birth seems to have stabilized around 33% of all births. Midwives attend approximately 46% of all births (all home births and 33% of hospital births), general practitioners 11%, and obstetrician specialists 43% (Eskes 1992; Treffers et al. 1990). The organization of obstetric care in the Netherlands is based on the selection for risk factors (Eskes 1992, p. 161). Midwives and general practitioners select pregnant women on risk factors using an officially approved list of medical indications for exclusion for home delivery, Kloosterman's Indicatielijst (Keirse 1982). The insurance system will only pay for a hospital birth with an obstetrician specialist when specialist care is indicated (Keirse 1982; Kloosterman 1978; Kloosterman 1984). In addition to home delivery women can also choose to give birth at a hospital (at low additional cost), still under the primary responsibility of the midwife or general practitioner.

The prospective Wormerveer16 study (Van Alten, Eskes and Treffers 1989) gives some insight into the predictive quality of this risk assessment approach. 8,480 women residents of Wormerveer, or approximately 85% of all pregnant women in the catchment area during the period 1969-1983, were booked for delivery by a midwife either at home, or in a small birth center, or in some cases in a hospital. After excluding 500 cases for various reasons (see Figure 2) the remaining study population was 7,980 women, or 80% of all pregnant women during this period. Seventeen percent of these women needed referral to obstetric care in the hospital before the onset of labor (antepartum) and 8% needed transfer during delivery (intrapartum). Seventy-five percent (or 60% of all pregnant women) were delivered at home or in the small midwife-operated birth center (maternity unit). Combining the home

deliveries with the transfers during delivery (intrapartum and postpartum) to the category that was considered low-risk by the midwives (83% of the study population), we observe for this group a total perinatal mortality of 0.23%. The 17% of women transferred to obstetric care before the onset of labor (antepartum) due to complications or concurrent illness during pregnancy shows the expected significantly higher perinatal mortality of 5.17%. Unfortunately the study does not show outcomes for those 15% of women referred to obstetric care in early pregnancy and not booked with the midwifery service. But comparing the Dutch national average of 1.45% perinatal mortality for this period with the 1.11% for the study population, it is probably safe to assume that the early pregnancy referrals must have a significantly higher perinatal mortality. The clearly superior outcomes of the low-risk group versus the high-risk group (the referrals before the onset of labor) led Kloosterman to conclude

... that selection during pregnancy makes it possible to demarcate a low-risk group with a good prognosis, a group that in more than 90% of cases can deliver spontaneously without any interference (Kloosterman 1984, p. 122).

The perinatal mortality rate of the Netherlands has consistently been among the low end of developed countries (see Table 12), suggesting that their system of maternity care seems to operate quite effectively and supporting Springer's and van Weel's claim that

The key to the consistently good results of home births in Dutch primary care settings is meticulous selection of women at low risk of obstetric complications. This results in equal or better obstetric outcome compared with hospital birth, and fewer interventions, for a large number of women in the community (Springer and van Weel 1996, p. 1276).

National Birth Center Study (NBCS)

The largest study on out-of-hospital births done in the U.S. is the National Birth Center Study (NBCS) conducted by the National Association of Childbearing Centers (NACC) between 1985 and 1987 and covering 16,589 women who started prenatal care in one of 89 birth centers in 35 States (Rooks, Weatherby and Ernst 1992a; Rooks, Weatherby and Ernst 1992b; Rooks, Weatherby and Ernst 1992c; Rooks et al. 1989). The study does not report about the effectiveness of applying risk assessment instruments for referrals and transfers of these birth centers. But we can indirectly infer from the study (Rooks, Weatherby and Ernst 1992a) whether the screening practices employed resulted in admitting a low-risk group to intrapartum care at the birth centers and whether this group had favorable perinatal outcomes. Of the women who started prenatal care at a birth center, 14% needed a referral to obstetric care and planned hospital birth because of a prenatal complication. Another 10% moved or wanted to give birth somewhere else and 5% had induced or spontaneous abortions or other reasons. This left 71% or 11,814 women who were admitted to intrapartum care at a birth center.

Are birth centers chosen by a particularly low-risk group? A comparison of all subjects who entered the NBCS with all childbearing women in the U.S. with regard to a set of variables generally associated with perinatal risk shows some favorable (lower risk) and some unfavorable differences. The major differences are that the birth centers have a slightly larger proportion of women with low socioeconomic status (as measured by a SES index as well as by education) and a higher share of highly educated women, whereas the middle class seems to be underrepresented. Black women are also underrepresented while Hispanic women are overrepresented. Birth center women report that they smoke less and drink less, but they start prenatal care rather later in the second trimester with fewer in the first trimester. Overall one cannot conclude that this looks like a particularly low risk group based on these selected risk factors (no medical risk factors included).

How do the findings from the NBCS compare to the Wormerveer study (Rooks, Weatherby and Ernst 1992c, pp. 371,380; Van Alten, Eskes and Treffers 1989, p. 658)? Since the NBCS did not follow up on the referrals during pregnancy (antepartum), we have to limit the comparison of transfers and perinatal mortality to those women admitted to the intrapartum. Looking at the intrapartum transfer rates, it seems that the Wormerveer rate of 8.0% is below the NBCS rate of 15.8%, although we do not know whether the Womerveer study also included the infant transfers. Perinatal mortality, however, seems to be comparable. At first glance, the NBCS mortality rate of 0.13% is below the Womerveer perinatal mortality rate of 0.23%. But the NBCS rate does not appear to include fetal death before term. Using our 1989 and 1990 California data we can extrapolate a perinatal mortality rate including fetal deaths. Given that roughly 55% of the perinatal deaths were fetal deaths during the period from 20 weeks of gestation until term, that would mean a perinatal mortality rate for the NBCS population of 0.29% if we include fetal deaths, almost identical to the Wormerveer study's 0.23%. In both cases the mortality rate for the transfers is, as to be expected, significantly above the mortality for the non-transfers: eight times as high for the Wormerveer study and 15 times as high for the NBCS. This suggests that in both cases the risk screening is effective. These findings from the NBCS suggest that providers of low-intervention maternity care in U.S. birth centers appear to be able to successfully screen out women at high risk and make referral and transfer decisions that ultimately lead to favorable perinatal outcomes for this group.

What can we conclude from these studies about the predictability of low-risk pregnancies? Goodman, in a recent and somewhat polemic argument with the Canadian midwifery movement about their claim that home birth is safe under certain conditions, proposes that even the best fetal risk scoring system cannot avoid a "surprisingly high incidence of major complications" during the intrapartum. His argument implies that the resulting emergency transfers from home to hospital have a significantly worse outcome than the similar emergency cases inside the hospital, but he does not support this claim with any evidence (Goodwin 1997, p. 1182) 17.

Knox et al. have shown that their risk scoring system leaves only 5.9% of preventable poor outcomes not predicted by the scoring system when using the relatively low high-risk threshold of 16% of the pregnancies (Knox et al. 1993, p. 198). Weigers et al. go even a step further. Comparing low-risk Dutch women who had booked with a midwife and then chose to deliver at home to those women who had booked with a midwife and then chose to deliver in a hospital, they showed that choosing to give birth at home has at least as good outcomes as choosing to give birth in a hospital. This comparison included for both categories the emergency transfers to care by an obstetrician in the hospital. In addition to perinatal mortality they included a rather comprehensive set of outcome measures (Wiegers et al. 1996).

It seems reasonable to conclude that there are risk scoring systems with a sufficiently high predictive power to select a group of low-risk women who will have favorable outcomes in out-of-hospital settings, even when intrapartum emergency transfers to hospitals are included in the statistics for the group, as they should be. But risk scoring systems with high predictive power, although desirable, do not appear to be a necessary condition for successfully allocating women between a natural or obstetric approach. The Dutch example shows that an official set of sound rules, competent primary caregivers (midwives), and a well-functioning referral system can produce excellent results for the whole population.

THE NEED TO TAKE A NEW LOOK AT THE "SAFETY" ISSUE

It is probably fair to say that the United States has not yet experienced a comparable "safety debate" as have the U.K. or The Netherlands. This is surprising given that the United States, while having clearly higher obstetric intervention rates (e.g. cesarean sections) than comparable highly developed countries, has at the same time a significantly higher perinatal mortality rate (Figure 3).

Rooks argues that this rate "would be much worse, except that we lead the world in the ability to save the lives - although not always the full health and potential - of babies who are born so premature, small, or damaged that they would not survive without the most sophisticated, expensive neonatal intensive care" (Rooks 1997, p. 2 and p. 99). The United States has not been able to improve its international ranking in perinatal mortality since the seventies (Table 12). It must be noted here that since the mid-eighties the perinatal mortality rate in the Netherlands has been declining at a slower rate than the rates in neighboring developed countries, dropping the Netherlands from rank 6 in 1980 to rank 20 in 1990 (see Table 12), the same level as the United States. Possible reasons are considered to be the increasing share of hospital births and a significant increase in the average age of the mothers at birth (Wiegers, van der Zee and Keirse 1998). The issue needs further clarification.

IMPORTANCE OF THE ISSUE

Compared to other causes of death within our society, perinatal mortality cannot be considered a large cost to society, although the impact on the family in the individual case is devastating. Extrapolating from the California birth certificate and hospital discharge data18 which we are using for our research, the United States had roughly 47,000 perinatal deaths in 1990. Using the same data, 71% of all pregnancies can be considered candidates for benefiting from a natural approach to childbirth. Those pregnancies would have been responsible for only 16,000 perinatal deaths. And if we go a step further and look at the low risk study population (after liberal exclusions), the typical candidates for the natural approach to childbirth, then these 60% of all pregnancies would have accounted for only 3,800 perinatal deaths. Segmenting all pregnancies in this way shows clearly that for the large number of low risk pregnancies perinatal mortality is not really an issue - although we must keep in mind that in every pregnancy there is some element of anxiety about perinatal mortality and it is a devastating experience for everybody involved when it happens. Our previous review of the necessity and effectiveness of obstetric interventions suggests, that the issue for these low-risk pregnancies is rather that there is strong evidence that they are subjected to considerable unnecessary and potentially harmful obstetric interventions. It is the costs of these interventions and their long-term consequences which can be considered an unnecessary and high price society pays for subscribing to the obstetric approach to childbirth. We will discuss

these costs in more detail in the last chapter. The order of magnitude may be illustrated by one estimate developed by the Public Citizen organization, that eliminating the unnecessary cesarean sections, only one of the unnecessary interventions, would save the United States $1.3 billion per year (Gabay and Wolfe 1994). That number does not include any costs associated with long-term consequences of unnecessary cesareans.

We focus in this study on perinatal mortality because this issue seems to be an emotionally charged central argument for blocking constructive discussion and investigation into alternative approaches to childbirth, and thus prevents society from rethinking its approach to childbirth.

THE RE-EMERGENCE OF MIDWIFERY

There are a few developments that suggest that the debate about the "safety" of alternative birth settings in the U.S. might intensify. First, we observe an ever so slight questioning of the indiscriminate application of the obstetric model19 accompanied by an increase in midwife attended births in hospitals (Declercq 1992) and out-of-hospital births. It is probably not a coincidence that Roger Bulgar, in his presidential address as the outgoing president of the Association of Health Services Research (AHSR) on June 27, 1993, chose a study showing significantly better outcomes for birth protocols with an emotional support-giver during deliveries in a maternity hospital to illustrate his call for improving patients' outcomes through better care-giving (Bulgar 1993; Kennell et al. 1991). Second, we can observe a gradual embracing of the organizational form of the birth center by the maternity hospitals, a development driven not by safety or quality considerations but by economic cost considerations and, to a smaller degree, some women's preferences20. But once these in-hospital birth centers have grown and established their own constituency, competition between these midwife-managed birth centers and the traditional obstetrician dominated delivery ward in the same hospital is likely to develop. Finally, the pressures to reduce health care expenditures and the emerging emphasis on alternative care models (see Clinton's original health care reform plan) will cause payors to investigate cheaper forms of perinatal care. Among the outcomes could well be a push for midwives and birth centers and possibly even home birth. All these developments will threaten the obstetric profession, which in turn can then be expected to put the "safety" issue at center stage.

The push for increased use of midwives in the United States is not new. In 1985 the Institute of Medicine (IOM) recommended that the nation rely more heavily on nurse-midwives to provide care to "hard-to-reach" high-risk women and that state laws be amended to support nurse-midwifery practice (Institute of Medicine 1985). Those recommendations were based on studies that had found that fewer low-birthweight babies were born to socioeconomically high-risk women who had received their care from nurse-midwives (Levy, Wilkinson and Marine 1971). In 1995 the independent consumer advocacy group Public Citizen published Encouraging the Use of Nurse-Midwives: A Report for Policymakers. It recommended expanding nurse-midwifery education with a goal of gradually adjusting the ratio of certified nurse-midwives (CNMs) to obstetricians so that nurse-midwives become the primary provider of care to low-risk pregnant women. The report and recommendations were based on a Public Citizen Health Research Group study of 41 nurse-midwifery practices with in-hospital births and 41 nurse-midwifery practices in free- standing birth centers. A major finding of the study was lower cesarean delivery rates associated with care by nurse-midwives (Gabay & Wolfe, 1995).

Of particular interest to the whole "safety" debate for out-of-hospital births and the role of midwives are the developments in the United Kingdom. The United Kingdom recently reversed its anti-home-birth policy in favor of a policy intended to give every woman the right and ability to choose the type of birth care she wants, "including the option, previously largely denied to them, of having their babies at home, or in small maternity units." This policy change resulted from the findings of a year-long, in-depth investigation by a parliamentary committee, which concluded that the nation's previous policy of encouraging all women to give birth in hospitals could not be justified on grounds of safety. The committee had asked the obstetricians to provide evidence that their care was superior to midwifery care, which they could not (House of Commons Health Committee, 1992 as quoted in Rooks 1997, p. 353). This policy change also calls for a lead role and greater responsibility for midwives to hold their own caseloads and provide care throughout the antenatal, intrapartum, and postpartum period (Turnbull et al. 1996, p. 213).

A further example for the new emergence of midwifery is Canada, where midwifery was banned. After intense discussion, several Canadian provinces (B.C., Ontario, Alberta) established in recent years midwifery as a self-regulating profession with its own regulatory body, the College of Midwives, and the Health Ministry announced that the government would fund midwifery care provided by registered midwives. The Health Ministry also announced the Home Birth Demonstration Project (HBDP), a pilot project to determine not whether home births should be allowed, but rather how they should be administered and incorporated into the health care system (Rice 1996).

NEWLY AVAILABLE DATA OVERCOME SOME PREVIOUS LIMITATIONS

As previously discussed, the major shortcoming of research on the safety of alternative birth settings for low-risk women has been the lack of randomized trials. There are also only a very small number of case control studies (see Mehl et al. 1976). While these studies can draw conclusions about interventions and maternal and infant morbidity, the number of cases studied is too small for the analysis of perinatal mortality. This addresses the second major shortcoming of previous research as it relates to perinatal mortality, i.e. the small sample sizes. Large sample sizes are required for two reasons: (1) low perinatal mortality incidence in a healthy population where the mortality rates are expected to be less than 0.5%, and (2) low incidence of out-of-hospital birth around 1% (except for The Netherlands). Yet a third problem is that data bases that provide large enough sample sizes, such as public birth and death statistics, generally do not include information on obstetric risk factors or morbidity outcomes. The linking of the birth and death information with medical or personal information, e.g. regarding socioeconomic status, is difficult if not impossible and in all cases costly (Macfarlane 1989; Treffers et al. 1990). Since this information is essential to control for obstetric risks and compare women of similar risk across the different birth settings, previous studies have always fallen short of proper risk control.

Another data problem that tends to distort the results is the inability to distinguish between the planned and the actual place of birth (Alberman 1984; Campbell and Macfarlane 1987). Public data sources always report the actual place of birth. As a result the outcomes for home birth include all emergency births at home that were originally planned for the hospital, and most of these are in the high-risk category. Given today's small share of planned home birth, these cases could significantly distort home birth outcomes. Conversely, hospital outcomes include all the emergency transfers from planned home birth, also a high-risk category. However, these events have little effect on the outcome of the large hospital population because of the small number relative to all hospital births.

The California data base on births and perinatal deaths was revised in 1989, and when merged with the information from the hospital discharge abstracts for each hospital birth it allows for the first time to address many of the limitations of previous research. The proposed research design overcomes in particular the following previous limitations:

- The large sample size of more than 1,000,000 California births provides sufficient power for the analysis of perinatal mortality, despite the low perinatal mortality rate and the small percentage (1%) of out-of-hospital births.

- The crucial information about the planned place of birth in addition to the actual place of birth allows the identification of emergency home births that were originally planned for the hospital as well as the intrapartum emergency transfers from home or birth center to the hospital. We now can allocate births to birth settings according to the mother's original plans.

- Information about the cause of death allows us to eliminate those deaths from the analysis that are non-preventable and that are independent of the type of maternal care the woman received (e.g. certain congenital abnormalities).

- The large number of medical risk factors reported on the U.S. birth certificate allows us to control, i.e. statistically adjust or correct, for basically all risk factors assumed to affect perinatal outcomes. The reporting of these medical risk factors has been improved through formatted check-boxes on the birth certificate for complications and procedures of pregnancy and concurrent illness, complications and procedures of labor and delivery, and abnormal conditions/clinical procedures of the newborn. Earlier certificates had included a number of open-ended questions that were poorly completed (Taffel, Ventura and Gay 1989). Furthermore, by merging the birth certificate with the hospital discharge abstract, we can augment the information on medical risk factors from the birth certificate with the more reliable and complete information from the discharge abstract. This alleviates the previously insurmountable problem of underreporting of medical risk factors on the birth certificate.

OBJECTIVE OF THE STUDY

The discussion of alternative approaches to childbirth is often presented in terms of "home (or out-of-hospital)" versus "hospital" birth. In the United States, it is then quickly reduced to the question of what is safer for the woman and the baby: a birth with an intrapartum complication managed by a lay midwife at a home location or by an highly trained obstetrician in a tertiary hospital. As important as this sub-issue may be in the individual case, it is only one aspect of a much more complex issue. Proponents of the natural approach to childbirth argue that the price we pay for iatrogenic problems caused by unnecessary use of obstetric interventions during labor and delivery of women under obstetrician care might be larger than the benefits of avoiding poor outcomes due to intrapartum transfers which would be unnecessary if we did not have any out-of-hospital births (Rooks 1997, p. 296). While we will take a close look at the safety (perinatal mortality) question, including the transfer issue, we will also look at the larger issue of unnecessary interventions and their social and economic cost. We make this point here to alert the reader not to lock him/herself into this very narrow aspect of "home (transfer) versus hospital".

Since economic evaluation has become an increasingly important part of medical and health research, the U.S. Public Health Service, Department of Health and Human Services, convened in 1993 a Panel on Cost Effectiveness in Health and Medicine. In 1996 the Panel released its report with recommendations for improving cost-effectiveness analysis (CEA) in health and stressed that CEA needs to take a more comprehensive view by framing the analysis from a social perspective, assessing effectiveness and costs to society at large, and highlighting all impacts of an intervention and not just those which pertain to a narrow perspective (Gold et al. 1996; Manning 1998). The search for the appropriate approach to childbirth needs to follow these recommendations.

OBJECTIVE AND SCOPE

Given the infeasibility of a randomized trial to determine the "safety" issue, the focus must shift to developing an observational study design that provides us with the most definitive answer to the safety issue with the presently available data and the most appropriate statistical methods. The originality of the study lies in capitalizing on newly available public vital statistics data merged with hospital discharge abstracts, which for the first time allow for the comparison of birth outcomes for women with similar risk profiles who plan to deliver in different birth settings.

Against this background, this dissertation has two objectives:

1. The primary objective is to address the perinatal mortality controversy by developing an observational research design, which compares women of equal risk profile utilizing two different approaches to obstetric care. It needs to be stressed that the statistical modeling design is aimed at controlling for the potential risk factors. Our objective for this research is not to establish whether causal relationships exist between the various possible risk factors and perinatal outcomes, or in other words, determine which risk factors are the best predictors of poor outcomes, nor to identify those models which have the highest goodness of fit and predictive efficiency. This research rather tests the hypothesis that the non-interventionist natural approach to childbirth, as administered by midwives and some physicians in free-standing birth centers or at home, is as safe as the interventionist obstetric approach in hospitals, when using perinatal mortality as the outcome measure and after adjusting for as many of the possibly interfering or confounding factors as the available data base will allow. Given the low perinatal mortality rates and the impact that some missing or incorrectly allocated deaths could have on outcomes, special emphasis is placed on developing a high quality database and ensuring that the assumptions made in this process are available to the reader and do not unfairly bias the results. This study clearly strives to provide the most "definitive" answer to the perinatal mortality question for the two competing approaches to childbirth based on all 1989-90 California births and the data presently available for these births.

2. A secondary objective is to begin developing a broader framework for comparing the overall social and economic costs to society for these two competing approaches. It is proposed that there is sufficient preliminary research evidence suggesting that the interventionist hospital-based obstetric approach may lead to undesirable outcomes with significant social cost, and that society would be ill served if the comparison of these two competing approaches to childbirth remains restricted to the issue of perinatal mortality. The study examines the possibility that these social and economic costs might be several orders of magnitude larger than the usual cost associated with perinatal mortality in light of any differentials in perinatal mortality rates.

The scope of the study is limited to two years of California births, 1989 and 1990. California births make up roughly 14% of all U.S. births. While there are a host of poor outcomes associated with childbirth, the study is limited to perinatal mortality as the outcome measure, because of its dominant role in preventing an open discussion of alternative approaches to childbirth. The study limits itself to following a study plan laid out at the beginning (definition of birth setting categories, general exclusions, and liberal and conservative exclusion criteria for the low-risk populations). It does not follow some of the very interesting leads which arose during the analysis of this rich database - particularly the issues surrounding the apparent "protectiveness" or insignificance of some of the commonly accepted risk factors for the lower-risk population segments. In considering the natural and obstetric approach in a larger perspective this study is limited to just raising the issues and presenting some preliminary research findings.

BIRTH SETTING AS A PROXY FOR ALTERNATIVE APPROACHES

TO CHILDBIRTH

While the obstetric approach usually requires the hospital facility with its equipment, the natural approach is not limited to any particular birth setting. We saw that the general surgeon Odent and his midwives developed their natural approach of the "undisturbed birth" for women of all risk categories directly within the public hospital in Pithiviers/France. The large Normal Birth Center at the LAC/USC Medical Center Women's Hospital, Los Angeles, which also accepts women with no prior prenatal care, is managed exclusively by midwives and is located three floors above the high-risk labor and delivery ward of a tertiary care teaching hospital. At the Farm in Summertown, TN, Ina May Gaskin developed a large midwifery practice with direct-entry midwives and a couple of general practitioners, having a small birth center but focusing mainly on home birth. Her video on assisting a vaginal breech birth is appreciated by physicians and midwifes as a much needed teaching tool for a fast disappearing skill (Rooks 1997, p. 142). These are three examples of the wide spectrum of natural childbirth settings, and all have shown unquestionably good results.

It is this low-interventionist natural approach to childbirth, generally administered by midwives, which we want to contrast with the interventionist obstetric approach. In cases where the natural approach to childbirth is administered in out-of-hospital settings, the Dutch experience shows that a necessary condition for safe and effective operation of the natural approach is the seamless integration with the obstetric care system. This requires:

1. A universally accepted screening system that ensures that out-of-hospital birth is limited to low-risk pregnancies. Women also need to be made aware of these criteria.

2. Proper protocols for transferring women or babies with serious complications rapidly to hospitals. Transfers should not be seen as failures, neither by the mother and her family, nor by the midwife.

3. Good communication and working relationships between the attendants of the out-of-hospital birth and the physicians and nurses who will take care of the transferred woman or baby in the hospital. This probably requires hospital privileges for midwives who work in birth centers or offer home birth services (see also Rooks 1997, p. 348).

We are aware that, with the demise of natural childbirth and midwifery over the last half century, in present-day America these conditions generally do not hold. At issue is not the quality of the midwives in the United States. Given the low demand for midwives and that midwifery is not a high status profession in the United States, only women with a high level of dedication and determination will find their way into this profession. The training of certified nurse midwives (CNMs) is restricted to academic settings. For the lay midwives, or direct-entry midwives as they are now generally called, Rooks documents the increasing professionalization over the last twenty years (Rooks 1997, ch.9). It is the interface of all out-of-hospital care givers, not just the midwives but also the physicians, with the obstetrician-managed hospital system that does not work well. When examining the California data for 1989-90, we will be looking at the natural approach to childbirth in the United States under the current medical climate. We will compare the hospital birth system, which commands all the resources and dominates almost 99% of the market, with a struggling, suboptimal natural childbirth system which serves 1% of the market. However, if our hypothesis about the relative safety of these two approaches holds for these data, then a more institutionalized natural childbirth system would only strengthen our findings.

In reality, there is a spectrum of approaches and settings (Campbell and Macfarlane 1994; Macfarlane 1997). In the US we have home birth administered by direct-entry (lay) midwives, Certified Nurse Midwives (CNMs), or occasionally by family practitioners and obstetricians. The same group of providers attends births in free-standing birth centers. We have the in-hospital birth centers, physically inside the hospital, generally run by CNMs but under the supervision of obstetricians. Next, we have in some instances, hospital-owned birth centers in a different location than the hospital maternity ward and presumably under less influence by obstetricians. And finally there is the hospital maternity ward where obstetrician control protocols with CNMs and nurses in supporting roles.

In the US, home birth occurs outside the mainstream maternity care system, with non-regulated selection criteria (if any), and a questionable availability of backup services for intrapartum transfers. In both England and The Netherlands home birth is part of the standard maternity care system. In England only 2% of births occur at home, while in The Netherlands one-third of all births occur at home. Women planning home births in England work with their general practitioners (so far as they support them) and community midwives, who are also part of the National Health Service. There is no systematized selection process. General practitioners are in general not supportive of home birth and also have become de-skilled over time (Davies et al. 1996). Transfers in this system do not constitute any problems. Most normal deliveries in the hospital are conducted by midwives under the supervision of obstetricians (Macfarlane 1997, p. 15). In The Netherlands midwives provide independent maternity care for women with uncomplicated pregnancies and select out and transfer those women who require care from an obstetrician by using a nationwide accepted set of selection criteria. Since 1970, women booking with midwives can choose whether they prefer to give birth at home or with the midwife in the hospital. Of the 46% of births which are attended by midwives, one half occur at the hospital.

While the natural approach to childbirth can be implemented in any setting, the interventionist obstetric approach requires hospital facilities. Unfortunately the data available for California do not identify directly whether it was a birth with the natural or obstetric approach, but the data do provide information about the planned and actual place of birth. We therefore had to settle on the place of birth (birth setting) as a proxy for the type of perinatal care administered. By philosophy, education, and regulation, midwives are focused on non-interventionist care. Obstetric interventions like anesthesia and instrumental delivery are prerogatives of obstetricians. Home and birth centers don't lend themselves to interventionist care, and most obstetric interventions can only be administered in the hospital environment. While there are some interventionist midwives and family practitioners attending in birth in home and birth centers who will quickly transfer to the hospital, and while there are non-interventionist physicians and midwives in hospitals, our analysis assigns all births planned for home and birth centers to the natural approach to childbirth and all births planned for the hospital to the obstetric approach.

With this assumption we group all midwife-attended births in in-hospital birth centers into the obstetric hospital category. If our hypothesis that the natural approach is as safe as the obstetric approach holds, then this misallocation will have no significant effect on our interpretation of the statistical results. If, however, it turns out that the natural approach (the home and birth center setting) is less safe than the obstetric approach, then this misallocation will result in a smaller advantage for the obstetric approach than actually exists - although the small number of births in in-hospital birth centers in 1989 and 1990 will probably have no significant effect given the large hospital population.

Looking at a recent study, however, we have to assume that this allocation of midwives who administer births in in-hospital birth centers to the hospital category will favor the hospital category. This study examined differences between certified nurse midwife (CNM) and physician delivered births in hospitals (MacDorman and Singh 1998). 94% of all CNM delivered births in 1991 occurred in the hospital, the majority probably in in-hospital birth centers. The study used birth certificate data for the entire United States. It was assumed that any under reporting of medical risk factors on the birth certificate would apply to both groups in the same way, as birth certificates for hospital births are probably filled out by the same hospital record staff regardless of attendant. The study was limited to singleton, vaginal deliveries at 35-43 weeks gestation. Due to lack of appropriate information the study could not address the issue of transfers antepartum or intrapartum from midwifery to obstetric care for cases with complications. The study controlled for the following health risk factors: age, race, education, marital status, birth order, start of prenatal care, and gestation. The following medical risk factors showed statistical significance and were included: hydramnios/oligohydramnios, abruptio placenta, breech/malpresentation, fetal distress, precipitous labor, premature rupture of membranes, and seizures during labor. The risk of experiencing a neonatal death (<28 days) was 33% lower for CNM attended than physician attended births, and the risk of delivering a low birthweight baby was 31% lower for CNMs.

COMPARING PERINATAL MORTALITY - RESEARCH DESIGN

Given the low perinatal mortality rates and the effects that even very small distortions in the data can have on the results, the reliability of the data sources and the quality of the data are of crucial importance. After establishing the quality of the data we will describe how the variables in our dataset were derived and which cases were excluded from the analysis. The section discussing methodology presents the power calculations and then describes the indirect standardization and logistic regression model. The chapter closes with a discussion of limitations of this observational study design.

DATA SOURCES

This study takes advantage of the 1989 revision of the birth certificate and a special database which RAND had developed for a research project about childbirth outcomes.

Birth Certificate Information (Birth Cohort File)

In 1989 a revision of the U.S. Standard Certificate of Live Birth was put in use greatly expanding the possible scope of research on pregnancy outcomes (Luke and Keith 1991; Taffel, Ventura and Gay 1989). The following changes in the California birth certificate are relevant for this study, which attempts to compare outcomes by birth setting while controlling for risk factors:

- Distinction between the different places of birth. e.g. hospital, birth center, home.

- Distinction between the planned place of birth and the actual place of birth.

- Improvement in the reporting of the medical risk factors by using formatted check-boxes versus open-ended questions.

Based on the information provided on the birth certificate (Appendix 1) the Maternal and Child Health Branch of the California Department of Health Services creates the California Birth Cohort file (Maternal and Child Health Branch 1993). This file combines live birth and fetal death records. It should be noted that the information items on the fetal death certificate mirror the information items on the birth certificate. The death records of infants who die within the first year of life are then matched and linked to their respective birth record, thus allowing us to analyze the mortality experience of a cohort of infants from birth to one year of age. These linked Birth Cohort Files for 1989 and 1990 are the basis for our analyses.

RAND Database with Hospital Discharge Data

The medical risk factor information that can be derived from the birth certificate, however, was not deemed sufficient. Validation studies for these information items on the birth certificate show that there is significant under reporting of medical risk factors and complications on the birth certificate (Buescher et al. 1993; O'Reilly 1991; Piper et al. 1993; Watkins et al. 1996; Woolbright and Harshbarger 1995). This under reporting would provide us with a lower risk profile than the true risk profile for the hospital population, whereas for home birth and birth centers the reported and the true risk profile would most likely be rather congruent, since the care givers will have screened out all high-risk cases and referred them to hospital care.

The RAND organization (Santa Monica, CA) had been awarded in 1990 a major research project by the Agency for Health Care Policy and Research (AHCPR), the Management and Outcomes of Childbirth PORT (Patient Outcomes Research Team). The focus of this research was on cesarean sections and approaches to reduce the high cesarean section rate. For this project RAND linked the 1989 and 1990 California Birth file to the California Office of Statewide Health Planning and Development's (OSHPD) hospital discharge abstracts of the mother and the infant (Keeler et al. 1995). This allowed RAND to augment the incomplete information about medical conditions/risks and complications on the birth certificate with more comprehensive and accurate information from the hospital discharge abstracts. Because the two databases do not share a unique identifier, a probabilistic linkage model using hospital of birth, birth date, admission period, sex, race, and ZIP code of mother's residence was applied. While in this RAND database 96.3% of the birth certificates could be linked to a maternal discharge record (Phibbs et al. 1996), only 80% of the singleton births could be linked to an infant discharge record21. Linking the birth certificate to the infant discharge record was more difficult because the mother's birth date, a crucial variable for linkage, is not on the infant discharge record. In hospitals with large numbers of births the birth date, sex, race, admission period, and mother's ZIP code were often not sufficient to distinguish among babies born on the same day. The biggest linkage problems were created by the four large Los Angeles County hospitals (approximately 7% of all California births) where, also due to higher rates of missing data, the match rates for infants ranged from only 44% to 55%. However, this linkage problem for infant discharge records is of less concern for our research since we limit our data requirements to the birth certificate and the mother's discharge record.

Birth Center Information

For the place of birth the Cohort file, which forms the basis of our dataset, uses the California Maternity Hospital Code (University of California at Santa Barbara 1989). Non-hospital births are coded either as "Home or Residence" or they fall into the categories "Other Not-in-Hospital", "Hospital in CA with no Maternity Hospital Code ", or "Unknown Place of Birth or Out-of-State Birth" (Maternal and Child Health Branch 1993, pp. 7-18). While this coding provides a separate code for home birth, free-standing birth centers receive a Maternity Hospital Code like any hospital. For our study, free-standing birth centers with Maternity Hospital Codes were identified by inspecting the name of the facility and checking whether the "Hospital Ownership" was coded as "Child Bearing Centers". The resulting listing of cases was compared against information obtained from the National Association of Childbearing Centers and the American College of Nurse-Midwifes (ACNM) on accredited smaller birth centers in 1989 and 1990, which allowed us to identify some additional free-standing birth centers. We also obtained from the ACNM a list of nurse-midwife practices in California operating during these two years. Our hope had been to match the names provided by the ACNM with the names of the birth facility as provided by the field "California Maternity Hospital Name" on the birth certificate. However, this field only shows the name of the facility if the birthplace was actually coded with a maternity hospital code. Nurse-midwife practices without a maternity hospital code, which would be the vast majority of nurse-midwife practices, could not be identified from the Cohort file.

The natural childbirth/birth center category includes only births in free-standing birth centers, not births in in-hospital birth centers which had to be allocated to the obstetric/hospital approach. While in-hospital birth centers have grown significantly over the past years, there were few in 1989 and 1990. There is no distinguishing identifier in the Maternity Hospital Code for in-hospital birth centers; they are an integral part of their respective hospital's statistics. We did not try to infer births in in-hospital birth centers by checking whether the birth attendant/certifier was an MD or a midwife (Maternal and Child Health Branch 1993, p. 23). The fact that an MD signs as attendant/certifier on the birth certificate does not mean that he or she was the care giver throughout the pregnancy. All cases that started out as midwife administered in-hospital birth center cases but later required intervention by an obstetrician will show the obstetrician as the attendant on the birth certificate. These cases would then mistakenly be allocated to the "hospital" category, unduly burdening the hospital birth category (interventionist obstetric approach) with complicated birth cases which actually originated from midwives in the in-hospital birth center category. The remaining in-hospital birth center category would then be left with only uncomplicated birth cases and thus a more favorable outcome profile. Leaving in-hospital birth center cases in the obstetric approach/hospital category, however, and not allocating them to the natural approach where they belong, will affect the results and we will take that into account when discussing the results.

Quality of the Data

Crucial to the validity of our research outcomes is the completeness and accuracy of the information of our two data sources, birth certificate and hospital discharge abstracts. The birth certificate is a legal document which must be filed with the state or local registrar within 10 days of the birth of the child. The hospital is held responsible for filing. For out-of-hospital births the attending physician or midwife will generally file the birth certificate directly with the county birth registrar. The hospital's birth recorder obtains the general information from the mother by use of a work sheet (frequently bilingual) which the mother fills out (Appendix 2), and a complementary interview with the mother when the birth recorder collects the work sheet. The birth recorder derives the medical information from a labor and delivery summary (Appendix 3), usually filled out by the labor and delivery nurse, and from the patient file. The hospital discharge abstract is based on the same labor and delivery summary and on the patient file and is part of the same database that provides billing information. We can assume that the medical information on the discharge abstract (the diagnosis and procedure codes we are interested in) is rather complete and accurate since it has consequences for the billing.

Since birth certificate data are extensively used in the public health field for planning and evaluating maternal child health programs (Buescher et al. 1993, p. 1163), they have repeatedly been analyzed for their reliability. All these studies show that the medical information on the birth certificate is significantly less accurate/reliable than the medical information which can be obtained from the hospital discharge abstracts (Buescher et al. 1993; O'Reilly 1991; Piper et al. 1993; Watkins et al. 1996; Woolbright and Harshbarger 1995). For the medical information we, therefore, will construct new derived variables based on information from both data sources, the birth certificate and the discharge abstract (see pp. 90-93).

Here we are concerned with the quality of the sociodemographic information and the general obstetric risk factors on the birth certificate. We have two sources for evaluating the quality of these information items for a specific birth certificate: (a) agreement with the corresponding maternal hospital medical records, and (b) agreement with the mother's questionnaire response in the National Maternal and Infant Health Survey (1988).

Table 13 summarizes our findings on data reliability: While most of the items show a very high degree of accuracy and completeness, there are a few less reliable items. However, there are some ways to improve completeness and accuracy for these items.

- Planned Birthplace. This is an item available only on the California birth certificate and we could not find any validity analysis. Stanford Hospital's Health Information Management Service maintains that, for Stanford birth certificates, this information is accurate and not missing (Smith 1997). For our entire California dataset we have only few cases with missing information on actual or planned birth setting (all these cases will be excluded from the analysis): for BIRTHSET=13-16 with 851 cases we do not know the actual birthplace; for BIRTHSET=12 with 1,040 cases and a significantly worse perinatal mortality we have no information about the planned birthplace. We know the hospital as the actual birthplace but we cannot allocate these cases to our birth settings (for further discussion see p. 89 and p. 101).

- Method of Delivery. Besides a very small number of missing values, the literature assessing validity reports a disagreement between birth certificate and medical records of 7-8%, which we also find in our California data, which show a disagreement rate of 10%. We recoded those cases letting the hospital discharge information augment and, where necessary, override the birth certificate information.

- Race/Ethnicity. The reporting of this variable is highly accurate. In those cases where the race of the mother was missing, we used the race of the child for race of the mother when available..

- Years of Education. The lower agreement rate in the NMIHS might be due to the distinction of academic and vocational training in the NMIHS questionnaire and to additional education of the respondent since the birth event (Schoendorf et al. 1993, pp. 5-6). Since we intend to use education as a proxy for socioeconomic status, we decided to look at mother's and father's education and then construct an education variable for the family using the higher of the two education values. This procedure also reduces the number of missing values somewhat. We still have missing values in 4,065 cases.

- Insurance Status. A recent study found "excellent agreement between insurance information in (California) birth certificates and interview data" (Braveman et al. 1998). The study concludes that "insurance information appears to be an appropriate resource for examining both the extent of coverage for maternity care and the association between prenatal care utilization and insurance status". We followed the study's recommendation to collapse "HMO/capitated" and "private" insurance into one category. We have missing values in 1,747 cases.

- Length of Gestation. In general this information is less accurate. However, looking at full-term births (38-42 weeks) versus births of less than 38 weeks, the NMIHS found a substantially higher agreement rate of 91.1% (Schoendorf et al. 1993, p. 9). Since our research focuses on low-risk pregnancies, the full-term birth category will dominate our population. The gestational age is calculated from the date of last menses as reported by the mother on the birth certificate, and not the gestational age as reported in the medical records, since doctors seem to have a tendency to routinely record standard gestations of either 36 or 40 weeks (David 1980, p. 971). Length of gestation is also needed for Kotelchuck's prenatal care utilization index, which we will be using in this study. For calculating the appropriate number of prenatal care visits, Kotelchuck infers missing or unreasonable gestational age from the birthweight (Kotelchuck 1994). This approach finds support in Williams' research, who found that perinatal mortality rates are much more sensitive to birthweight than to gestational age (Williams et al. 1982, p. 624).

- Prenatal Care. This is the other information item which is less accurate. However, once one distinguishes the responses not by monthly break-down but by trimester, then those women having initiated prenatal care in the first trimester show a rather high agreement rate of 93.8% in the NMIHS (Schoendorf et al. 1993, p. 8). Since our research eliminates women with inadequate prenatal care, i. e. where care was initiated after the fourth month, we are mainly dealing with that part of the population for which the reporting seems to be rather accurate (McDermott et al. 1997). It should be noted that a very small study of low income and low risk women who received prenatal care in a county health department clinic showed poor accuracy when comparing the clinic records (which can be considered more accurate than hospital medical records) to the birth certificate information. Only 50.7% had complete agreement on the trimester when care began (Clark, Fu and Burnett 1997).

The generally lower level of agreement for length of gestation and prenatal care in the NMIHS compared to the validation studies using medical records might also be due to recall problems with a survey a year after the event. But there is also the question of whether medical records are the "gold standard" for accuracy. A British comparison of different sources of information suggests that mothers may be more reliable sources of information than medical records (Oakley, Rajan and Robertson 1990).

For our low risk study population with adequate prenatal care and more or less full term gestation, the overall quality of the birth certificate information about sociodemographics and general obstetric risk factors seems to be high. The same can be said about the medical risk factor information once it is augmented by the information from the hospital discharge abstracts.

DATASET FOR THIS RESEARCH

The basis for our research is the birth certificate information as reported in the Cohort file, which is available for every birth. The discharge information from the RAND database is only used to improve the medical risk factor information from the birth certificate for the hospital cases. After presenting the dataset for this research with its variables and discussing how these variables were constructed, we will describe a set of cases that lie outside the scope of our study and which will be excluded from our analyses.

Outcome and Control Variables

Our objective is to measure the effectiveness of alternative approaches to childbirth. While the quality of medical care can also be measured on the level of structure (adequacy of the environment) and process (proper medical practice), we will use for our research a "risk adjusted outcome measure" (Blumberg 1986; Williams and Wroblewski 1991). Perinatal mortality (PERIDTH) is our outcome variable (Figure 4).

The perinatal period is one of high risk of adverse outcome (Figure 5). More than half of all infant deaths (within the first year of life) occur within the first week and 31% occur on the day of birth. It is generally believed that some proportion of deaths during the perinatal period is preventable through effective obstetric care (Powell-Griner 1986; Tew 1990; Williams and Wroblewski 1991, p. 21)

...because the factors responsible for the bulk of deaths in early infancy are often present before delivery or are brought about by unfavorable conditions during the birth process (Williams 1979, p. 96).

Perinatal mortality is an internationally accepted indicator of the effectiveness of medical care and meets all the requirements of a good health services tracer. It is unambiguous and reliably reported (Williams and Wroblewski 1991, pp. 6-7). It has therefore been chosen as the risk adjusted outcome measure (RAMO)22 for the State supported Maternal and Child Health (MCH) Data Base, which was developed to measure the effectiveness of perinatal care across maternity hospitals in the State of California (Williams and Wroblewski 1991, p. 3).

Table 14 lists the variables used in this research and indicates whether they are derived from the birth certificate alone or whether they are derived from the birth certificate and the discharge information from the RAND database. The variables starting with LOG... are the categorized form of the variable as it will be used for the logistic regression models (for more detail about the categorical variables see Table 26).

The independent variables can be grouped into sociodemographic variables, general obstetric risk factors, obstetric history; and the medical risk factors reported on the birth certificate can be divided into antepartum and intrapartum risk factors. In Table 15 we compare the independent variables (risk factors) included in our database with the risk factors included in other major risk assessment instruments (Goodwin, Dunne and Thomas 1969; Keirse 1982; Knox et al. 1993; Vedam and Kolodji 1995). We observe that our set of risk factors (independent variables) is rather inclusive. Some of the obstetric history risk factors are not available in our database. There are four risk factors (marital status, smoking, previous birth >4,500 grams, and pregnancy within last year) for which there is information in our database, but because of questionable reporting reliability or because we originally did not expect to need them they were not included.

While the list of medical risk factors is reasonably complete, additional information about the severity of the condition would have been helpful in deciding whether the condition actually constitutes a risk for the pregnancy. Lacking that information, including every risk factor whether severe or mild probably results in a study population with a higher risk profile than was actually present. This affects mainly the hospital population since the riskier cases in the home and birth center population have been screened out and referred to the hospital by the care providers. This effect will not dilute our findings: if our hypothesis holds then the fact that the hospital population is actually of somewhat lower risk than the way we are treating it in our analysis strengthens the findings rather than weakening them.

Derived Variables

For the purpose of our research we constructed the following variables derived from our data sources:

5. Birth Settings (BIRTHSET). While our objective is to compare the non-interventionist natural approach to childbirth, as administered by physicians and midwives in free-standing birth centers or at home, to the interventionist obstetric approach in hospitals, we cannot identify these two approaches directly in our database. As discussed above, however, the birth setting can be considered a proxy for these differences in obstetric care. The birth certificate provides information about the actual place of birth and the planned birthplace. The following birth settings (Table 16) are an attempt to construct birth settings as homogeneous as possible with respect to the type of obstetric care typically delivered at that respective birth setting. In cases of uncertainty we either excluded the cases or allocated them in such a way as to rather err in favor of the hospital category.

a. Birth planned for home and actually occurred at home (BIRTHSET=1). It is assumed that this birth setting reflects the non-interventionist natural approach. Even if the birth at home is attended by an obstetrician or family practitioner23, it is assumed that these physicians subscribe also to a non-interventionist natural approach to childbirth.

b. Birth planned for home but actually occurred in the hospital, or transfers form home to hospital (BIRTHSET=3). It is assumed that these births were originally booked with a non-interventionist caregiver for home delivery but required eventually a transfer to a hospital.

c. Birth planned for free-standing birth center and actually occurred in the birth center (BIRTHSET=5). It is assumed that this birth setting reflects the non-interventionist natural approach. It is assumed that physicians who practice in birth centers subscribe to the non-interventionist natural approach to childbirtH34.

d. Birth planned for birth center but actually occurred in the hospital, or transfer from birth center to hospital (BIRTHSET=7). It is assumed that these births were originally booked with a non-interventionist caregiver for delivery in a free-standing birth center but required eventually a transfer to a hospital. Unfortunately this does not seem to be a clean category of transfers from birth centers to hospitals and, contrary to the category of home transfers (BIRTHSET=3), we will not use it to make a statement about the outcomes for birth center transfers. This category very likely includes also cases which were originally planned for birth in an in-hospital birth center (and not just in free-standing birth centers), explaining the unexpected large number of cases we find in this category25. Since the hospital code does not distinguish between the hospital's in-house birth center and the maternity ward, we probably have a group of cases in this category which were planned for in-hospital birth center and actually occurred in the in-hospital birth center as well as those which needed to be transferred from the in-hospital birth center to the hospital's maternity ward. However, all cases in this category were planned for the non-interventionist natural approach in a birth center and thus can be aggregated into our overall birth center category without distorting the findings in any way. Nevertheless, we do not have a clean category for transfers from free-standing birth centers to hospitals.

e. Birth planned for hospital and actually occurred in the hospital (BIRTHSET=9). It is assumed that this birth setting reflects the interventionist obstetric approacH36. The vast majority of all births fall into this category. In our statistical analysis we will use this category as our reference category.

f. Birth planned for hospital but actually occurred at home, or unplanned home birth (BIRTHSET=11). It is assumed that these are births originally booked with an obstetrician for the hospital but where precipitous labor resulted in a birth at home.

g. Birth planned for hospital but actually occurred in a birth center (BIRTHSET=10). Since these births (223 cases, see Table 32) occurred in free-standing birth centers (not in-hospital birth centers) and since free-standing birth centers are unlikely fall-back locations for precipitous labor of births planned for the hospital, we assume that the information about the hospital as the planned birthplace is inaccurate. We assume that these births were also planned for a free-standing birth center and thus for the non-interventionist natural approach and belong to the birth center category.

h. Planned birth location unknown, birth actually occurred at home (BIRTHSET=4). This is a small number (66 cases, see Table 32) with no perinatal deaths and we assume that it belongs to the non-interventionist natural approach and to the home birth category.

i. Planned birth location unknown, birth actually occurred in the hospital (BIRTHSET=12). This is a problematic category since, although small in absolute numbers (1,040 cases or 0.13%, see Table 18), it has by far the highest perinatal mortality. The prenatal care was adequate since we have already excluded all cases with inadequate prenatal care. However, for these cases we cannot distinguish between births booked originally with an obstetrician for a hospital setting (planned hospital) and birth booked originally with a natural childbirth caregiver for home or birth center but eventually requiring a transfer to the hospital and care by an obstetrician (planned home or planned birth center). Since any allocation of these cases would be guess work, we are forced to exclude this category from our analysis (see further comments on p. 101).

k. Finally we have a few birth settings (BIRTHSET=13-16) where we have partial information on the planned birth place, but do not know the actual birth place. This is a small number of cases (851 cases or 0.10%, see Table 18) with a perinatal mortality not significantly different from the hospital reference category. We exclude these cases from our analysis.

While we will calculate our information for all separate birth settings, for the first level of our analysis the birth settings will be aggregated to the following three categories:

- Non-interventionist natural approach = "HOME" (BIRTHSET= 1,2,3,4),

- Non-interventionist natural approach = "BIRTH CENTER" (BIRTHSET=5,6,7,8,10),

- Interventionist obstetric approach = "HOSPITAL" (BIRTHSET= 9,11).

2. Medical risk variables are used to correct the observed under reporting of medical risk factors on the birth certificate by complementing the information on the birth certificate with information from the hospital discharge abstracts. For each medical risk factor on the birth certificate for the reporting categories "Complications and Procedures of Pregnancy and Concurrent Illness" and "Complications and Procedures of Labor and Delivery" we identified the respective ICD-9 diagnosis and procedure codes (Appendices 4 and 5). Three sources contributed to the development of these tables: (1) R. Harold Holbrook, Jr., MD, Associate Professor Gynecology & Obstetrics, Stanford University School of Medicine, (2) A listing of High Risk OB Definitions developed by the Business Development Department of Stanford Health Services in 1993 with the clinical input from Maurice L. Druzin, MD, Professor Gynecology & Obstetrics, Stanford University School of Medicine, (3) Cristian Centiu, manager Coding/Data Quality, Health Information Management Services, Stanford Health Services. For studying readmission rates, the California Hospital Outcomes Project has, under the guidance of Patrick S. Romano and Harold S. Luft, developed a set of risk factors for postpartum maternal complications based on the discharge information (California Hospital Outcomes Project 1996). These Romano-variables are part of the RAND dataset and we used them in addition to the ICD-9 coding. We thus construct two sets of variables:

a. Complications and Procedures of Pregnancy and Concurrent Illness (XXRISK00 - XXRISK31). The birth certificate provides this risk factor category, which contains 30 possible conditions (plus a category "unknown" and "none"). For every case this information from the birth certificate goes directly into the corresponding XXRISK variable (1 or 0). If that results in a "0" value for a hospital case, we then check the hospital discharge information in the RAND dataset for this case and if we find any diagnosis or procedure codes related to this condition, then we enter a "1" for the respective XXRISK variable. Table 17 shows that there is little agreement between the two sources (column "Both"27) and, with very few exceptions, considerable underreporting for these medical risk factors on the birth certificate ("Discharge Abstract alone" reports many more cases with that condition than "Birth Cohort alone").

b. Complications and Procedures of Labor and Delivery (XXCOMP00 - XXCOMP32). The birth certificate provides this second risk factor category, which contains 31 possible conditions (plus a category "unknown" and "none"). For every case this information from the birth certificate goes directly into the corresponding XXCOMP variable (1 or 0). If that results in a "0" value for a hospital case, we then check the hospital discharge information in the RAND dataset for this case and if we find any diagnosis or procedure codes related to this condition or any relevant Romano-variable, then we enter a "1" for the respective XXCOMP variable. Table 17 shows again that there was little (though slightly more) agreement between the sources and considerable underreporting for the complication variables.

3. Race (RACE_M). Mother's and infant's race differ on the birth certificate. Only in those cases where the father is white, the mother's race determines the race of the child. In all other cases the father's race determines the race of the child (Maternal and Child Health Branch 1993, p. 16 and Appendix B). Since this research focuses on medical conditions related to perinatal outcomes, we assumed that the mother's race was the more influencing factor (see also Schoendorf et al. 1993, p. 2). In those cases of missing values for the race of mother where we had a value for the race of child, we used race of child for race of mother. We used the information on the birth certificate about Hispanic origin of the mother to separate a "Hispanic" ethnicity category out of the "White" category. We then use the following four race categories: White, Hispanic, Black, Other.

4. Insurance Status (INSUR). Since perinatal research seems to suggest that poor women are at higher risks of adverse outcomes, and since there is also some discussion about medical treatment being dependent on insurance coverage or socioeconomic status, we use the information from the birth certificate about the principal source of payment for prenatal care and the expected source of payment for delivery to construct an insurance variable with the following three categories:

a. Uninsured, including the following codings: self pay, no charge

b. Low Insurance Coverage, including: Medi-Cal, Worker's Compensation, Title V, other Government programs, other non-governmental programs, Medically Indigent, CPSP

c. Privately (Fully) Insured, including: Blue Cross/ Blue Shield, Private Insurance Company, HMO, Medicare.

5. Method of Delivery (XXDELMTH, XXCSECT, XXPRCSEC). For hospital cases, where we have the information from the birth certificate and from the mother's discharge abstract, we found about 10% of the cases where the information from the two sources was not congruent. We assumed that the discharge abstract, which is generally linked to the computerized billing system, will be more accurate. We decided to let the discharge abstract augment and where necessary override the birth certificate in case of different reporting.

General Exclusions and Size of the Database

The scope of this study is not the entire population of births in California in 1989 and 1990, but rather that population of low-risk women for whom birth at home or in a birth center could be considered an option28. However, since (a) there is some ambiguity about what constitutes a high risk pregnancy, and (b) there is the suggestion that the non-interventionist natural approach to childbirth might actually be advantageous even for high risk pregnancies, we will define our "Study Population" as the totality of pregnancies for which different types of obstetric care might possibly make a difference with respect to birth outcomes. We will then use a "conservative" and a "liberal" set of exclusion criteria to arrive at our low-risk population.

The study population is derived by applying a set of general exclusions to all births in California by California residents. These exclusions reflect cases which are principally not considered for birth center or home birth, which are not preventable death and/or independent of the type of obstetric care the women receives, or where information vital for our analysis was missing and the exclusion of these cases with missing values would not distort our findings (Table 18).

1. Multiple Births. The study population is singleton births. Multiple birth (2.2% of all births) is generally considered an exclusion criterion for home or birth center birth. Birth certificate and hospital discharge data could not be matched for multiple births.

2. Unmatched Mothers. There was a group of birth cases where RAND could not match a mother's discharge record to a particular birth certificate. This group accounted for 3.7% of the cases and had a perinatal mortality rate that was lower than the entire population, although not significantly. It should be noted that we also keep all those cases in our study population where the mother's discharge record could be matched to the birth certificate, while that of the infant could not be matched. Our focus is on risk factors of the mother which allow us to identify low-risk pregnancies, and for the outcome variables which we have chosen we need to draw only on the discharge record from the mother but not on the discharge record for the infant.

3. Birthweight <500 grams. Such low birthweight cases, which have almost no chance of survival, are principally excluded from home birth or birth in birth centers.

4. Fetal Death <20 Weeks Gestation. Since the objective of this research is to analyze different obstetric care approaches and their effects on perinatal outcomes, we are excluding all early fetal deaths, which are generally supposed to be independent of the type of obstetric care.

5. Congenital Anomaly as the Cause of Death. These deaths are considered not preventable and independent of the type of obstetric care. Of all exclusions due to congenital anomaly as cause of death, where the mother was not already excluded because of inadequate prenatal care (INDEXSUM=0 or 1), only four cases occurred in our home or birth center categories, all the rest were in the hospital category. All four were considered non-preventable by medical experts.

6. Induced Abortion as the Cause of Death. These are certain perinatal deaths, independent of the type of obstetric care.

7. Inadequate Prenatal Care. Since we want to compare properly executed different approaches of obstetric care, we will drop from consideration all cases where we cannot assume properly executed (adequate) prenatal care. This is also important because there is a higher likelihood of finding women with no or very little prenatal care in the hospital population, while women who chose natural childbirth are generally more likely to be under the supervision of a caregiver at an early stage. The birth certificate provides information about the start of prenatal care and the number of prenatal care visits. Kotelchuck's Adequacy of Prenatal Care Utilization (APNCU) Index (Kotelchuck 1994) combines the two pieces of information into one index. No prenatal care, any prenatal care initiation after four months, or prenatal care initiated early in pregnancy but with less than 50% of the visits recommended by the American College of Obstetricians and Gynecologists (ACOG) are considered inadequate (INDEXSUM=1). We are also excluding those 2.2% of cases where the information about prenatal care was missing (INDEXSUM=0). By excluding these cases with inadequate prenatal care and missing information about prenatal care, we are excluding roughly one thirds of all perinatal deaths. We notice that the exclusions in the Home+Center category (the out-of-hospital category) have a higher perinatal mortality than the exclusions in the Hospital category (Table 19). To test whether these exclusions create a bias, we ran logistic regressions29 on the excluded cases to identify the main effect for the out-of-hospital category (LOGLOC_N, Home+Center) and its significance. For the total exclusion we find a slightly though not significantly positive effect (higher perinatal mortality) for the out-of-hospital category (LOGLOC_N, Home+Center: standard coefficient of 0.0075 with p=0.4294 and odds ratio 1.149). Separating the effects of births at home (LOGLOC_H) and in birth centers (LOGLOC_C) we find that birth center has a significant protective effect (odds ratio of 0.243) while home birth has a significant positive effect (odds ratio of 1.625). Both effects when combined cancel each other out to the above non-significant effect. The Home+Center category (natural childbirth) contains 266 cases that the Kotelchuck index classifies in our dataset as cases with "missing information about prenatal care". These 266 cases account for an overproportionate number of perinatal deaths (56). It is questionable whether these case would actually qualify for the natural approach category because county birth registrar offices generally will not accept incomplete birth certificates from natural care providers. These cases are most likely home births without any care provider. We conclude that the exclusion of cases with inadequate prenatal care or missing prenatal care information does not bias our findings by favoring the natural childbirth category.

8. Birthweight Missing or Unknown. Because birthweight is such a critical factor in risk adjusting perinatal mortality rates, we are excluding these 124 cases. They account for only 0.067% of all births. While the perinatal mortality rate (PMR) is very high, it is not different for the hospital or out-of-hospital cases.

9. Race Missing or Unknown. Because race is supposedly a critical factor in risk adjusting perinatal mortality rates, we are excluding these 1,894 cases. All cases where race was missing were hospital cases. They have the same low PMR (0.580%) as the entire study population (0.560%).

10. Cases with Unknown Birthplace. The information about the actual and planned birthplace is crucial for constructing our birth setting variables. We exclude the rather small number of cases (851) where we do not know the actual birthplace (BIRTHSET=13-16). The PMR for this category is significantly higher than the one for our study population. However, if we subject this category to our indirect standardization analysis, we find that it actually has a lower Indirectly Standardized Perinatal Mortality Ratio (ISPMR) than our reference category hospital births. This means the observed mortality in this category was lower than what we would have expected given this category's risk profile. In addition we exclude the small category where birth occurred at the hospital but where we have no information about the planned birthplace (BIRTHSET=12). This category has a very high PMR and a number of perinatal deaths large enough to affect results. However, once we exclude all non-preventable deaths30, birthweight and race missing (already dealt with above), and birthweight <2,500 grams, cases which are not affected by the type of obstetric care or should not be included in the natural birth category, then we are left with only 24 presumably preventable deaths, all fetal deaths (Table 20). If we apply to these cases the liberal and the conservative exclusion criteria, we are left with 12 perinatal death cases which we are excluding from our low-risk population. The California Office of Vital Records suggests that these cases are probably all hospital cases: Since all perinatal deaths in a hospital go through a case review, it is unlikely that the patient records would not show that this death was a transfer case from an out-of-hospital setting, thus limiting the hospital's responsibility.

After a total of 331,932 general exclusions (28.9% of the original population of births), our database after exclusions, the "Study Population", then has 815,927 births. The perinatal death rate for the Study Population is now significantly smaller (from 1.16% down to 0.56%) since we excluded 65.8% of the perinatal deaths because these cases belong to a population not considered eligible for the natural childbirth approach in birth centers or at home, or are considered not preventable or independent of the type of maternity care the women received.

The sets of conservative and liberal exclusion criteria identifying the low risk part among our remaining Study Population will be discussed in the next chapter.

Given the large database, the number of cases with missing values in our Study Population is very small, partly as a result of our general exclusions, but also a tribute to the quality of the birth certificate data. Since all the perinatal deaths associated with missing value cases on the level of our Study Population occurred in the hospital category, the way we are treating the missing value cases could not possibly have a favorable effect on the outcomes for the natural childbirth category. We decided to allocate the missing value cases always to the high-risk category of our categorized variables, to rather err in favor of the lower-risk categories within the respective variable. We observe that the occurrence of missing values is highly correlated with perinatal deaths. The reason is probably that the hospital birth recorders do not want to unnecessarily bother an already distressed mother with paperwork and questions about the birth/death. This is in line with the observation in the field that county birth registrars, while not accepting incomplete birth certificates from midwives or out-of-hospital physicians, obviously accept incomplete birth certificates from hospitals.

The issue of missing values for the general risk factors (sociodemographic variables, general obstetric risk factors, and obstetric history) of the Study Population was handled in the following way31:

1. Birthweight and race missing: The few cases were excluded as discussed above. Their exclusion could not bias the results.

2. Length of gestation: For cases with missing or unreasonable value for gestation the Kotelchuck program was used which imputes gestation from birthweight for those cases. Since the Study Population does not contain any cases with missing birthweight, we have no missing value for gestation.

3. Prenatal care: The prenatal care index from the Kotelchuck program was used. There are no further cases with missing value in the Study Population since in the general exclusions we excluded all cases with INDEXSUM=0 (missing). As discussed above, if these cases would create a bias it would be in favor of the hospital category.

4. Mother's age: There were 58 cases missing with a perinatal mortality rate (PMR) of 60.345%. All these deaths belong to the hospital category. These cases were kept in the Study Population and it was assumed they belong to the higher-risk age category >34 years.

5. Education: There were 4,065 cases missing with a PMR of 2.583%. All these deaths belong to the hospital category. These cases were kept in the Study Population and it was assumed they belong to the higher-risk category of only "some high school education".

6. Insurance coverage: There were 1,747 cases missing with a PMR of 4.236%. All these deaths belong to the hospital category. These cases were kept in the Study Population and it was assumed that they belong to the higher-risk category of "uninsured".

7. Parity: There were 382 cases missing with a PMR of 0.875%. All these deaths belong to the hospital category. These cases were kept in the Study Population and it was assumed that they belong to the higher-risk category of "first child".

METHODOLOGICAL FRAMEWORK

The quality of our analysis depends on how well we achieve our objective of actually comparing women of similar risk profiles with respect to their outcomes (perinatal mortality) in the competing approaches to childbirth (proxy: birth settings). After having identified the overall study population for which differing approaches to childbirth could potentially have different effects on outcomes (see Table 18), we then exclude in three steps successively more women constructing three lower-risk study populations. While excluding more and more risk categories from the study population does leave the remaining population segment more homogeneous with respect to risk, but the population segments in the different birth settings might still have significantly different profiles with respect to the remaining risk factors. To control for these risk factors we will use two approaches: indirect standardization and logistic regression. We will use indirect standardization with a set of six risk adjusters to compare our birth settings. In doing so we follow the commonly used approach of risk-adjusted outcome measures for comparing hospital outcomes. Still, there are more than these six potential risk factors on which the birth setting populations might differ significantly. To control for all potential risk factors available in our database we will subject our different risk-level study populations to logistic regression. This will allow us to identify whether there is a significant main effect for the out-of-hospital birth settings after controlling for all potential risk factors. The logistic regression will also provide us with some information to what degree the risk factors included in our study actually influence perinatal mortality for the different risk-level populations of California pregnancies.

Exclusion Criteria for Birth Center and Home Birth

The first step is to define the low-risk population of pregnancies, which we want to investigate with respect to their perinatal mortality outcome. Practitioners in the home and birth center settings (midwives and physicians) are using certain exclusion criteria to screen out any potentially risky pregnancy, which will then be referred to a hospital based obstetrician. It should be noted that this screening-out is not a one-time activity at the initiation of prenatal care, but rather an ongoing observation process where any newly emerging risk factor may lead to a referral. Such transfers may occur early in the pregnancy (antepartum), during labor when circumstances require it (intrapartum), and also in the postpartum period.

Since there are no national level practice guidelines for birth at home or in birth centers, we decided to use three sets of exclusion criteria32. First a set

of "conservative exclusions" (Table 21) in which a large set of risk factors would be used to screen out woman from the natural childbirth population. For the "conservative" set of exclusion criteria we used as a starting point the exclusion criteria of one of the largest midwife-managed in-hospital birth centers in California, the Normal Birth Center (NBC) at Los Angeles County and USC Medical Center, Women's Hospital (Greulich et al. 1994, p. 196). Second, a significantly smaller set of "liberal exclusions" (Tables 22) where we use the term "liberal" in the sense of liberal admission to the natural childbirth approach. Finally a risk level where only birthweight below 2,500 grams would be considered ineligible for the natural childbirth approach.

Since our data do not allow us to distinguish the severity of an exclusion criterion like anemia, diabetes, or concurrent illnesses, we are forced to exclude even "mild risk" cases, which in practice are usually not excluded. Similarly, while the Los Angeles exclusion criteria accept documented low transverse uterine scars and undocumented uterine scars with a history of uncomplicated vaginal birth after cesarean (VBAC), we had to exclude all pregnancies with previous c-sections. This limitation of our data leads to unnecessarily excluding a group of low-risk pregnancies. Given that for home and birth center cases we only have the birth certificate information with its general under reporting of risk factors, we will probably keep a certain number of "mild risk" cases in the natural childbirth category while excluding similar "mild risk" cases from the hospital category. This makes the natural childbirth category slightly more risky than it actually appears in our data. This treatment will not dilute our findings: If our hypothesis holds, then if the natural childbirth population is actually of somewhat higher risk than the way we are treating it, our analysis errs in the direction of being conservative. Doing so serves to strengthens our findings rather than weakening them.

Since we are determining ex post whether any of the exclusion criteria were relevant for a specific case, we are also capturing all those cases where a risk factor manifested itself only late in the labor stage. We end up with four different risk-level populations for our analyses. They are, in order of their presumed potential of risk for complications:

1. The entire "Study Population" without any exclusions.

2. The "Study Population =2,500 grams", where we have excluded only all low birthweight cases with less than 2,500 grams from the entire study population. There seems to be no controversy that these low birthweight cases should not be managed with the natural approach.

3. The "Liberal Exclusions" set, which reflects the opinion of many natural care providers that one does not need to be too restrictive in admitting women to the natural childbirth approach.

4. The "Conservative Exclusions" set, which reflects more stringent screening criteria and results in the population with the presumably lowest risk potential.

Power Calculations

Since the perinatal mortality rates are so small (average of 0.56% for entire study population, and between 0.04% and 0.29% for the low-risk segments), and the numbers of home or birth center births are also small relative to the large category of hospital births, we did power calculations (Kraemer and Thieman 1987) for the important birth settings of the liberal and conservative exclusions runs, the latter having the smallest number of cases. The power calculation requires us to determine the minimal value (critical effect criterion) by which the perinatal mortality of our home and/or birth center categories must differ from our hospital reference category to satisfy us that we have established the validity of our hypothesis of equal perinatal mortality (Kraemer and Thieman 1987, p. 10). We decided to use two critical effect criteria: (a) An "absolute" change of 0.1 percentage points in the mortality rate, which seems to be an appropriate order of magnitude when expecting mortality rates between 0.1% and 1.0%, and (b) a more conservative "relative" change of 10% of the mortality rate in case the rates exceed 1.0%. Since we are using the Z-test for significance testing in the indirect standardization, the appropriate statistic for the power considerations is also the Z-statistic. Using Kraemer & Thiemann's Master Table (with 1% significance level and for two-tailed test) we get the results shown in Table 23. For the absolute critical effect criterion of 0.1 percentage point change in the mortality rate, all the critical birth settings with smaller numbers of cases have a power above 99%, except BIRTHSET=3 in the conservative exclusion set which has a power of 87%. For the more conservative relative critical effect criterion of 10% change in the mortality rate we find very low power for

the home to hospital transfer category (BIRTHSET=3). There was no possibility to increase the sample size -we had only data for the two years 1989 and 1990. In the indirect standardization analysis later we find a highly significant difference in increased mortality (ISPMR) for this home to hospital transfer category (BIRTHSET=3) in the liberal exclusion set, despite the low power. In the conservative exclusion set, however, we do not find a significant difference and must conclude that for this category and the conservative effect criterion the sample size is not sufficiently large to validate the result of our hypothesis testing. Overall we can conclude that the number of cases in our aggregated home and birth center categories as well as in our individual birth setting categories is more than sufficient to provide the statistical power required for this analysis. The only exception is the home to hospital transfer category (BIRTHSET=3), when it is viewed as a separate category and when the conservative critical effect criterion is applied.

Indirect Standardization (Risk Adjusted Outcome Measure)

The objective of this research is to analyze different approaches to childbirth and their effects on perinatal outcomes. Since the different types of maternity care cannot be identified directly from the available data, we infer the type of care from the birth setting (BIRTHSET), which can be considered a proxy for type of maternity care. The standard method for comparing medical care outcomes across hospitals (in our case different birth settings) is to use risk-adjusted (or indirectly standardized) outcome measures (Williams 1979; Williams and Wroblewski 1991; California Hospital Outcomes Project, 1996). This study looks at perinatal mortality as the outcome measure.

We will mirror the approach of the California Maternal and Child Health (MCH) Data Base33 (Williams and Wroblewski 1991), which compares perinatal care across all maternity hospitals in California. In our analysis we will compare different birth settings as the treatment with perinatal mortality as the endpoint. The MCH Data Base uses combined risk factor specific perinatal mortality rates as a risk adjusted (indirectly standardized) outcome measure and uses birthweight, race and sex as risk factors. According to Williams, birthweight has been shown to be a good proxy for maternal-fetal health as well as a powerful predictor of perinatal outcome, accounting in combination with sex, race, and plurality34 for 70-80% of the variance in neonatal and perinatal mortality rates35 (Williams 1979).

Age and socioeconomic status (we use education and insurance coverage as proxies for socioeconomic status) are also considered to be associated with outcome variance. The natural childbirth population and the obstetric population in our database show significant differences with respect to these factors (Figures 6, 7). Therefore, we add age, education, and insurance as risk adjusters. While race is one of our risk adjusters, we control for parity only in the Conservative Exclusions run, where we exclude cases with high parity (Figure 8).

Following the approach of the MCH Data Base (Williams and Wroblewski 1991, pp. 21-32), we will group the risk factors birthweight (9 categories from 500 grams in steps of 500 grams up to =4,500 grams), race (4 categories: white, hispanic, black, other), sex, age (3 categories: <18, 18-34, >34), education (4 categories: <12 years of education, 12, 13-15, =16), and insurance (3 categories: uninsured, Medi-Cal, private insurance) into 2,592 different risk categories (or cells). For each cell the perinatal mortality rate is calculated based on that part of our sample which had planned to give birth in the hospital and actually gave birth in the hospital (BIRTHSET=9)36, providing us with an "expected perinatal mortality rate" for each risk cell. The birth setting "planned and actual birth in hospital" (BIRTHSET=9) is our reference category against which we want to test our hypothesis that birth at home and in birth centers are as safe as birth in hospitals. This hospital reference category accounts for 98.8 % of all births in our study population. The large number of observations for our reference category ensures that there are no empty cells (cells without any births) when calculating the expected perinatal mortality rates for the reference category. Even in the smallest dataset with conservative exclusions we do not encounter any empty cells.

We then weight these risk-cell specific expected mortality rates with the distribution of births over all cells of the particular birth setting, which yields the expected perinatal mortality rate (EXPMR). For each birth setting we will then compare the actually observed perinatal mortality rate (OBPMR) against the expected perinatal mortality rate (EXPMR) and create the Indirectly Standardized Perinatal Mortality Ratio (ISPMR)

ISPMR = OBPMR/EXPMR

An ISPMR equal to 1.0 means that the observed population segment (birth setting, BIRTHSET) has a perinatal mortality equal to that of our reference category (with planned and actual birth in the hospital, BIRTHSET=9). An ISPMR above 1.0 means that the observed population segment has a higher perinatal mortality and since ISPMR is a ratio we can directly measure the percent by which the perinatal mortality of that birth

setting differs from the mortality of our hospital reference category, after adjustment for differences in risk factors. An ISPMR below 1.0 means the population segment has a lower perinatal mortality. We will also calculate 99% confidence intervals for the ISPMR.

The Z-statistic (ZPMR) or standard normal deviate will then be calculated (Williams and Wroblewski 1991, pp. 27-28) for the ISPMR to determine whether the observed perinatal mortality of this birth setting is significantly different from the expected perinatal mortality if the birth setting would have had the same mortality experience as the hospital reference category (two-tailed test). ZPMR measures the probability of obtaining an ISPMR different from 1.0 due to sampling variability alone, where

ZPMR = (ISPMR - 1)/sqrt (var ISPMR)

where the variance of ISPMR is

var ISPMR = var PDTH/EDTH3

PDTH is the observed number of perinatal deaths and EDTH is the expected number of perinatal deaths. All through our analyses we are using a probability level (p-value) of =0.01 as the criterion for statistical significance, i.e. alpha=1%, two-tailed. For a more detailed discussion of the equations used in the MCH analyses see Williams (1991, pp. 30-32).

Logistic Regression

It should again be pointed out that the objective of this study is not to establish whether causal relationships exist between the various possible risk factors and perinatal outcomes, or in other words, which risk factors are the good predictors of poor outcomes. There is a large body of research into this subject37. Rather, after having used a set of generally practiced exclusion criteria and the augmented MCH risk factors for our risk-adjustment in the indirect standardization, we now want to subject our data to logistic regression including all risk factors of the dataset into the models as control variables.

Perinatal mortality as the dependent outcome variable is a binary variable. The majority of the risk factors, the independent control variables, are binary or categorical variables. The few continuous variables, like mother's age, education, gestational age, parity, and birthweight, were transformed into ordered categorical variables (Table 24). For setting up the appropriate categories we employ the commonly used cut-off points in risk assessment instruments and in the literature. The appropriate model for testing association for such a system with a dichotomous dependent and categorical independent variables is a logistic regression model (Hosmer and Lemeshow 1989; Menard 1995; SAS Institute Inc. 1990). We used SAS 6.12 (SAS Institute Inc. 1990).

As a first step we did a univariate logistic regression analysis for each variable. In the cases of categorical variables, where we needed to construct a set of dummy variables, all dummy variables were put into the same logistic regression model. For each variable, using perinatal mortality (PERIDTH) as the outcome variable, we obtained the standardized estimated coefficients38, the p-value of the likelihood ratio test for the significance of the coefficients, and the corresponding odds ratios with their 99% confidence interval. As indicators for the overall goodness-of-fit of the models we calculated the adjusted generalized coefficient of determination R2adj , the adjusted R2LA , a log-likelihood analogue to the linear regression R2 , and the c statistic, the area under the receiver operating characteristic (ROC) curve. While a relationship can be statistically significant (p=0.0001), the R2 may not be substantively significant (e.g. R2=0.005) for a large sample (Menard 1995, p. 19) and we may want to exclude this variable from the model building process. We decided on p=0.0001 and a rather inclusive R2adj =0.00139 as inclusion criteria.

When assessing the fit of the model in logistic regression we do not have a simple summary statistic for goodness-of-fit like the ordinary least square (OLS) R2 for linear regressions. Since in logistic regression there are several possible measures of difference between the observed and fitted values (Hosmer and Lemeshow 1989, p. 137), several summary measure have been developed (Hosmer and Lemeshow 1989, ch. 5; Magee 1990; Menard 1995, ch. 2). Each has advantages and disadvantages depending on whether the dependent variable is dichotomous or polytomous, the relationship of the number of covariate patterns to the number of subjects (n- or m-asymptotics), or being invariant to the parametrization of Theta. Following Menard we will use as an analogue to the linear regression R2 a measure of the proportional reduction in the absolute value of the log-likelihood, R2LA = (GM - 2k)/D0 = (D0 - DM - 2k)/D0 , where k is the number of independent variables and the adjustmen (Menard 1995, p. 22). In addition the use the generalized coefficient of determination R2adj = 1 - (D0/DM)2/N as provided by SAS, where N is the number of cases (Magee 1990, p. 252; SAS Institute Inc. 1995, p. 68). Since in logistic regression the upper bound on the R2 measure can sometimes be less than 1, we are using here the measures adjusted for the number of parameters in the model (Menard 1995, p. 22) or adjusted for the maximum R2 (SAS Institute Inc. 1995, p. 68). Both these measures have a "proportion of explained variation" interpretation, but because of nonlinearities and nonspherical errors there is not necessarily a unique R2 (Magee 1990, p. 252). We will also use the c statistic, which represents the area under the receiver operating characteristic (ROC) curve and can range from 0 to 1. The higher the number, the better the predictive power of the model (SAS Institute Inc. 1995, p. 91). Since in our analyses the values of the R2LA and the R2adj were consistently very close, we decided to show only the commonly used R2adj and the c statistic, both provided by SAS, in the presentation of our results.

In addition to the univariate analyses we analyzed all two-way interactions and included into the models those for which (a) both standardized coefficients were significant at p=0.0001, and (b) the R2adj >0.01. We are left with 110 two-way interactions, of which 82 include the low gestation < 36 weeks or the low birthweights <2,500 grams, which will be excluded from the natural childbirth category in most of our analyses. Furthermore, we analyzed a set of more complex interaction terms. None of these more complex interactions met the above inclusion criteria.

The logistic regression models were run on a Sun Ultra computer under SunOS 5.5.1 on Unix System V Release 4.0, with 500MB of RAM and 1GB of swap space. The larger models (entire database with and without interactions, and the study population dataset with interactions) turned out to be too large for this platform. They did not converge when using the gradient descent convergence method, the default in SAS 6.12. Either the models ran for so long that we had to terminate them, or the runs were aborted with the error message: Floating Point Overflow. After consulting with the SAS Institute we were advised to use the ridging convergence method40 for these three large datasets, which is faster and will be the default method in future SAS versions. The SAS Institute could not explain the long duration of these runs nor the error message and ascribed these phenomena to the fact that the size of our models was probably stretching the capacity of our computer or SAS. In addition, for the analysis of the entire database with interactions we had to use resampling. For the large hospital category we used 30 jackknife samples with 20% samples each (Efron 1982). The smaller home and birth center categories were included in their entirety. The coefficients of variation for those standardized regression coefficients which are significant were generally below 0.02, suggesting that this sampling design is acceptable.

When the three large models ran successfully with ridging (and, for the largest model, with resampling), we decided to re-run all smaller models also with the ridging method, although they had successfully run with the gradient descent convergence method. The results were identical with only an occasional difference in the fourth decimal. All the results we are presenting in this paper are obtained with the ridging method.

LIMITATIONS OF THE STUDY

We have discussed the impossibility of subjecting this research issue to a randomized clinical trial. In addition, possible selection bias and, to some degree, missing information on the birth certificate, which requires us to use proxies for the type of maternity care the woman received, placed some limitations on the study. While the missing information would allow a more in depth analysis, it does not in any way affect the accuracy of those findings presented here. The possible effect of still existing selection bias, however, cannot be quantified.

Observational Design and Selection Bias

A general limitation of any observational study design is that it can never give a definitive answer regarding causality, unless all factors possibly affecting the dependent outcome variable were included. It should be noted that the 1989 revision of the birth certificate allowed for a wide input from organizations and experts who are users of these data (Freedman et al. 1988). Therefore, it can be assumed that the present list of variables in the birth certificate reflects (a) the existing knowledge of which factors are expected to affect childbirth outcomes in a significant way, and (b) the expectation that these factors can be captured with the birth certificate reporting.

Although this study includes all those factors assumed to determine the risks of childbirth, the main effects of obstetrical care differences on perinatal mortality might still be the result of selection bias due to factors not included in the model. The population of low risk women might have subsets of women with even lower risk, and these women systematically might elect to give birth in birth centers or at home. For example, one could argue that the mother's confidence in her ability to go through with the birthing process, her ability to stay focused and relaxed during labor (see the study by Kennell et al. 1991), and her physical condition could be factors which might have a significant effect on the course of labor and the need for obstetric interventions. Given the relatively small number of women who opt for natural childbirth, we must assume that there are many women with these same characteristics also in the hospital group. However, the natural childbirth group might still have a relatively larger share of those women.

Regarding selection bias, we do observe that the group of women opting for the natural childbirth approach (home and birth center) differs from the group that opts for the obstetric approach (hospital). The natural childbirth group is generally older, has more children (see Figure 7), is predominantly white, has a higher educational level, but has also a higher rate of uninsurance (see Figures 5, 6). But in all cases, except for race, the differences within the natural childbirth group between women who give birth at home and those who give birth at birth centers are even larger than the differences between the natural childbirth group and the hospital group. We also know that we have at least two sub-segments in the birth center group: a low income segment which for economic reasons selects birth center services, and a more educated segment which wants to go back to a more natural way of life. All this suggests a high degree of heterogeneity within the natural childbirth group, ruling out any simple selection bias mechanism.

Desirable Information Missing in Birth Certificate Reporting

For future research on maternal and child health outcomes and quality of maternity care the birth certificate information needs to be improved in several dimensions. First, we need direct information about the type of maternity care the woman received instead of the place of birth as a proxy for the type of maternity care. Second, we need specific information about transfers of the woman and the infant. When did the transfers occur and what sites were involved? Third, we need more and better information about obstetric interventions and a broader range of poor outcome measures for the mother and the child. Perinatal death is a rare and extreme outcome. Most obstetric interventions have side effects. Presently we cannot reliably identify the obstetric interventions (e. g. induction or augmentation of labor, epidural or other anesthetic intervention, episiotomies) in the merged birth certificate/hospital discharge database nor poor outcome measures as a result of obstetric interventions. Finally, it would be very desirable to have at least some reliable information on life-style associated risk factors such as alcohol and drug addiction, smoking, exercise, or nutrition.

Since the Division of Vital Statistics at the National Center for Health Statistics had indicated that a new revision of the national birth certificate is just getting under way, the author took this opportunity to make some suggestions with respect to the above mentioned improvements (Appendix 6).

COMPARING PERINATAL MORTALITY - RESULTS

As would be expected from such a rich database, the analyses produced a multitude of interesting findings and points to areas for further investigation far beyond the scope of this study. In line with the objective and scope of our study we are limiting our analyses and interpretation of the results to the very narrow question of the relationship of different birth settings to perinatal mortality after controlling for the potential risk factors. After first presenting the low perinatal mortalities for the low-risk populations we will briefly discuss the risk factors and the relationship of each to perinatal mortality in our database. We will then turn to the core question of this study, the comparative perinatal mortality between the natural childbirth category and the hospital birth category. In a brief excursion we will show how these two categories fare with respect to cesarean section as an outcome variable.

LOW PERINATAL MORTALITY IN THE LOW-RISK POPULATIONS

After having excluded (a) all cases with inadequate prenatal care, since they don't allow us to compare the properly executed different types of obstetric care, (b) all perinatal death cases which were not preventable and/or independent of the type of obstetric care the women received, (c) all multiple birth cases which are principally not considered candidates for the natural approach to childbirth, and (d) a group of cases with critical missing values, whose omission, however, should not distort our findings, we are left with our Study Population (see Table 18, p. 96) which accounts for 71% of all California births for 1989 and 1990. For these pregnancies, different types of maternity care might possibly make a difference with respect to birth outcomes. We notice that the perinatal mortality rate for this part of the childbearing population is only 0.560%, or one fifth of the rate of the excluded 29% of the pregnancies, which have a perinatal mortality rate of 2.649%. This is due to the fact that almost a third of all perinatal deaths are either not preventable and/or not affected by the type of maternity care, and another one third are linked to a segment of women who have had no or inadequate prenatal care. If we further exclude all births below a birthweight of 2,500 grams from our study population, because these cases are not considered candidates for natural childbirth, then we are left with 68% of all pregnancies and a perinatal mortality rate of only 0.186%, compared to a perinatal mortality rate of 3.225% for the excluded 32% of California pregnancies.

Our first major finding is that only 11% of all perinatal deaths occur in the 68% of all pregnancies - those that can be considered low-risk and are considered candidates for the natural approach to childbirth. Furthermore, it has been shown that this group experiences a very low perinatal mortality. The conventional way of calculating perinatal mortality statistics for all births obscures this fact. Conventional statistics (a) do not distinguish between high-risk and low-risk populations but calculate mortality rates only for the entire population, and (b) when they distinguish between hospital and out-of-hospital birth, they do so by allocating according to the actual place of birth and neglect the unplanned home birth and transfer issue. If we subject our California data to the conventional way of presenting perinatal mortality statistics, then we see the familiar pattern: A perinatal mortality rate for home birth twice as high as that for hospital birth (Figure 9). We also see a very low mortality rate for birth centers, which is usually not presented. If we correctly allocate unplanned home birth and transfers and properly segment pregnancies into high-risk and low-risk populations, we see a completely different picture: The perinatal mortality rates for the low-risk pregnancies are a fraction of the "conventional" perinatal mortality rates, and the differences between the birth settings are very small.

We can draw two conclusions from this insight:

1. The large number of low-risk pregnancies has a very low perinatal mortality. When comparing alternative approaches to childbirth for these low-risk populations, perinatal mortality as an outcome measure will have only a very limited meaning. Other outcome measures are very likely to be much more important evaluation criteria. The discussion about alternative approaches to childbirth, therefore, needs to be put into a larger perspective and cannot be limited to the narrow "safety" question.

2. Eighty-nine percent of all perinatal deaths occur in high-risk pregnancies, which are not considered to be candidates for the natural approach to childbirth. An increase in natural childbirth among low-risk women should, therefore, not have a major impact on perinatal mortality as a whole. Discussions about effective steps to decrease perinatal mortality need to focus on phases and elements in the maternity care system that lie before the phase where a natural childbirth approach might affect outcomes. A general move toward midwifery (not within the scope of this research), however, would very likely have a positive effect on decreasing overall perinatal mortality. Midwifery care has been shown to lead to a reduction in low birthweight and inadequate prenatal care, two major factors in perinatal mortality.

Our analysis shows that the often emotional "safety" debate with regard to natural (or out-of-hospital) childbirth versus obstetric hospital birth is due to a lack of proper segmentation of the childbirth population. The aggregate data used up to now have obscured the fact that the population segment under discussion has very different characteristics from the overall population.

RISK FACTORS FOR LOW-RISK PREGNANCIES

Our second major finding concerns the risk factors, which are supposed to drive perinatal mortality, and their relevance to low-risk pregnancies. It seems that many of these risk factors do not affect perinatal mortality for low-risk pregnancies in a significant way, though they might still be significantly associated with other poor outcomes for mother and child, something we could not test with this database. There also seems to be a group of risk factors that trigger medical treatment, and do that so successfully, that it results in a protective effect at least with respect to perinatal mortality outcomes. Again these risk factors might still have a significant negative association with other poor outcomes. In general we have to realize that the ex-post analysis of risk factors, which are now buffered by medical treatment, has limited value for evaluating the predictive power of these risk factors (which, however, is not purpose of our research). We can only make a statement about the effects of these risk factors by viewing the risk factor including the medical treatment it triggers as a unit.

The objective of the logistic regression models is to test whether the out-of-hospital birth settings - home and birth center - have a main effect on perinatal mortality with respect to the reference category hospital - after controlling for the potential risk factors. When subjecting all potential risk factors in our database individually to univariate analysis to identify those which were significant and therefore needed to be included into our logistic regression models, we found the expected associations, for the most part, but also some puzzling results with negative coefficients, i.e. risk factors having a lowering effect on perinatal mortality (Table 25). These initially counterintuitive results of "protective" risk factors require further analysis and explanation and they are mentioned here as suggestions for further investigation. They are, however, not material to our analysis of perinatal mortality. To properly compare out-of-hospital births (natural approach) with hospital births (obstetric approach) we need to ensure that we look at women of comparable risk, i. e. we have to control for all factors which potentially might bias the outcome, whether these factors affect perinatal mortality positively (increasing perinatal mortality), whether they seem to be "protective", or even whether they may be redundant. For the purpose of our analysis we will keep them all in our models.

In the end we took a very inclusive approach when deciding whether to keep a risk factor in the model since we wanted to benefit from the control effect of as many risk factors as possible. For the general results as presented in this study we did not exclude any of the risk factors listed in Table 25, because they are all considered by the obstetric profession as risk factors, although some for whatever reasons do not seem to have any significance in our dataset. We then did some logistic regression runs with risk factor information that would be available only up to the onset of labor, the "antepartum risk factors", and with risk factor information as it becomes available during labor, the "intrapartum risk factors" (see Table 14). For those two sets of control variables we excluded those risk factors that were clearly not significant for any or most of the four risk levels of our Study Population and had R2adj values below 0.0005 (see risk factors shaded in Table 25). Table 25 also provides some information on which risk factors had positive effects (increase in perinatal mortality) and which one had a negative (protective) effect as well as whether these effects were significant at our usual significance level of p=0.01 or not. This information shows how the direction of the effect and the significance changes with the different risk levels of our Study Population. We consider this information helpful for further investigation into the role these risk factors should and can play in risk assessment instruments.

General Risk Factors

The general risk factors, including sociodemographic information, general obstetric risk factors, and obstetric history, show the expected significant positive effects (increasing perinatal mortality) for the entire California population, for our Study Population, and in general also for the Liberal and Conservative Exclusions sets.

1. Birthweight, particularly the low birthweights (LOGBWT05 - LOGBWT20), and gestation, particularly low gestation (LOGGST_1), have the dominant positive effect (high R2adj ). It should be noticed that long gestation (LOGGST_3) does not have the generally assumed positive significant effect in our California data (for the relevant reference categories see Table 24, p. 119).

2. Education (LOGEDU) and insurance coverage (LOGINS) (as the proxies for socioeconomic status) have significant positive effects across all levels of analysis.

3. While Blacks and Hispanics show increased risk, race "Other" (LOGETH_O), dominated in California by Asians, shows a "protective" effect.

4. While age over 34 (LOGAGE_3) increases mortality significantly for all risk levels of our study population, age below 18 (LOGAGE_1) no longer increases the risk in the low-risk segments of our study population.

5. The category with intermediate prenatal care (LOGCAR_2) has a significant "protective" effect for the entire California population (while non-significant effects for the other population levels). It seems that our general exclusions contain a significant group of healthy women, probably with birth experience (multiparous), who start prenatal care later and with reduced frequency, without jeopardizing their outcomes.

6. First pregnancy (LOGLIV_1) and high parity (LOGLIV_3) show the expected significant positive effects on mortality.

7. The significant protectiveness of female birth (LOGSEX=1) in the high-risk populations disappears in the low-risk populations.

We notice, as an emerging pattern for the Liberal and Conservative Exclusions sets, that risk factors that are clearly significant for the pregnancy populations before excluding high-risk cases tend to lose their significant effect in the low-risk populations. Once the high-risk cases are screened out, several of the assumed risk factors don't seem to be significant risk factors any longer - at least not when looking at perinatal mortality as the outcome variable.

Medical Risk Factors

The medical risk factors (XXRISK.., XXCOMP..) show a somewhat different and to some degree puzzling pattern. We do observe the same phenomenon of decreasing (or no) significance of the risk factors for the low-risk populations. The overall result is that for the low-risk Liberal and Conservative Exclusions populations almost none of the remaining medical risk factors (with only a couple of exceptions: XXCOMP19, XXCOMP20 XXRISK17, XXRISK29) has any longer a significant positive effect on perinatal mortality. We do, however, observe several "protective" risk factors which remain significant. This raises the question, whether for low-risk populations, as defined through the exclusion criteria, the remaining assumed medical risk factors, which are collected with the birth certificate, actually are helpful indicators for increased medical risk. As to the significant protectiveness of several of these assumed risk factors, maternity care experts suggest as one possible explanation that the observed presence of such a medical risk factor may cause the provider to actively manage this risk, possibly leading to successful risk control and favorable outcome with respect to perinatal mortality. While, under this hypothesis, these risk factors could show protective (negative) effects with respect to the extreme outcome of perinatal mortality, we would probably observe the expected significant positive effects (increased poor outcome) if we could look at other outcome measures, like the need for infant transfer to neonatal care or other poor outcomes. While we do have some information about other poor outcomes in the RAND database, we unfortunately do not have the corresponding poor outcome information for the out-of-hospital cases (which were not part of the RAND database and where we do not have the information from the hospital discharge abstracts).

For the high-risk population sets most of the medical risk factors show the expected significant positive effect on perinatal mortality. But for quite a few medical risk factors we do have the same significant "protective" effects as discussed above.

Overall, we observe that the general risk factors have generally much larger coefficients and R2adj values than the medical risk factors. The general risk factors low birthweight and short gestation have by far the highest impact. Among the medical risk factors, premature labor stands out as having the highest coefficient and R2adj.

Interaction Effects

While a single individual risk factor might not significantly affect perinatal mortality, the combination of two or more non-significant risk factors in a woman might have a highly significant affect on perinatal mortality. We, therefore, looked at all two-way interactions of risk factors and also some higher-order interactions.

Of all possible two-way interaction effects, only 110 passed our more conservative inclusion criteria of having highly significant standard coefficients (p=0.0001) for both variables in the interaction term and a R2adj =0.0141. If we further exclude those interactions where one of the variables in the interaction term is either birthweight <2,500 grams or gestation below 36 weeks, birth cases which will be excluded from our low-risk populations42, then we are left with only twenty two-way interaction terms (Table 26). Seventeen of them have premature labor (XXRISK21) as one of the interaction variables. Two are "protective" (negative sign), females with birthweight between 3,000 and 3,500 grams (LOGBWT30), and first child with birthweight between 3,000 and 3,500 grams (LOGBWT30). And the combination of more than adequate prenatal care (in itself an indication for complications) and abruptio placenta (LOGCAR_4 * XXCOMP11) has an increasing effect on perinatal mortality.

Looking beyond two-way interactions, we used a set of eleven correlation models (higher-order interactions) reflecting clinical research findings and community standards of care and considered to very likely result in poor outcomes (Table 27). We first ran each model with every possible 2-way, 3-way, ..., n-way combination and identified those interaction terms that were significant at p=0.05, a very inclusive selection criterion. Only nine three-way interactions met this criterion, none of the other higher-order interaction terms were significant at this inclusive level43. We then ran these nine interactions as though they were univariates and applied our inclusion criterion for univariates of p=0.0001 and R2adj =0.001. Six of the three-way interaction terms met these criteria and were included into our logistic regression models. It should be noted that five of the interaction terms derive their significance from low birthweight (<2,500 grams) or low gestation (LOGLOC_1), cases which the Liberal and Conservative Exclusion criteria exclude from the low-risk populations. The only three-way interaction included in the regression models for the low-risk populations is then XXRISK15 (bleeding before labor) * XXCOMP11 (abruptio placenta) * XXCOMP12 (placenta previa), from correlation model 4.

Contrary to our expectation, that combinations of several risk factors (interaction terms) are likely to increase the risk for perinatal mortality, only few interaction terms passed our rather inclusive set of selection criteria. Those who did, do not show any surprises. If they do not include a low birthweight or low gestation term, they include risk factors which are generally considered extremely high risk factors, such as premature labor, placenta previa, or abruptio placenta.

Including interaction effects into the logistic regression models does not have a large effect on the R2adj. We ran all models with and without interactions. The R2adj increase from including the interaction effects is very modest44: for our Study Population (without any exclusions) a 3.6% increase, for the Study Population =2,500 grams a 6.2% increase, Liberal Exclusions a 2.3% increase, and Conservative Exclusions a 1.7% increase.

In concluding our discussion of the results of the risk factor analyses we want to point out that our philosophy was to be inclusive, since our objective was to control for all potential risk factors. The fact that we observe the puzzling result of "protective" risk factors has no effect on our use of these risk factors for controls. However, the "protectiveness" of some risk factors and the widespread insignificance of the risk factors in the low-risk populations provides some interesting leads for further investigation.

PERINATAL MORTALITY COMPARISONS

The core of this study is the test of the hypothesis, that the natural approach to childbirth (birth setting Home+Center) is as safe, i.e. has the same perinatal mortality, as the hospital-based obstetric approach (birth setting Hospital) for low-risk pregnancies. Following are the results of subjecting our study population with its four risk levels to the two methodological approaches, namely indirect standardization and logistic regression, for testing this hypothesis. Both approaches allow us to control for risk factors to ensure that we are comparing women of comparable risk profiles.

The analysis of all California pregnancies in 1998 and 1990, potentially eligible for natural childbirth (the Study Population), gives a very clear picture: The hypothesis, that the non-interventionist natural approach to childbirth, as administered by midwives and some physicians in free-standing birth centers or at home, is as safe with respect to perinatal mortality as the interventionist obstetric approach in hospitals, cannot be rejected. The perinatal mortality is actually slightly lower for the natural childbirth approach, although not significantly so. This result holds, no matter which exclusion criteria are applied for identifying the low-risk pregnancies. And whether we use indirect standardization or logistic regression, we arrive at the same results. It should also be kept in mind that the assumptions we made in setting up the database generally favor the obstetric approach (hospital setting) - thus having the potential for strengthening our findings rather than weakening them. Our conclusion is that this consistent pattern of findings is a very strong suggestion that natural childbirth, including the problems that seem to be associated with intrapartum transfers in today's childbirth system in the United States, is as "safe" with respect to perinatal mortality as obstetrician-managed hospital birth for low-risk women.

Indirect Standardization - Results

When using perinatal mortality as a risk adjusted (indirectly standardized) outcome measure, we are using the same risk adjusters - birthweight, race, and sex - which the California Maternal Child Health (MCH) Data Base uses for comparing maternity care outcomes across California hospitals. In addition we add as risk adjusters age and education/insurance as a proxy for socioeconomic status. In Table 28 we can see that the Indirectly Standardized Perinatal Mortality Ratio (ISPMR) for the natural childbirth category (HOME+CENTER), i.e. birth at home and in a free-standing birth centers, is slightly but not significantly lower than the ISPMR for the obstetric approach (HOSPITAL). This is true for the population of low risk pregnancies when we apply the conservative exclusion criteria as well as when we apply the liberal exclusion criteria. This is also true when we look at our Study Population and exclude only births below 2,500 grams, cases which under all circumstances should have been transferred to obstetric care before labor. Interestingly it is also true for the entire Study Population, which includes all categories of pregnancy risks. This latter finding suggests, in line with Tew's conclusion (Tew 1990, p. 244), that care givers of the natural approach seem to manage even the high risk cases, whether they eventually require a transfer to obstetric care or not, with the same or possibly slightly better outcomes than the obstetricians manage these case with their interventionist approach in the hospital.

Once we differentiate within the natural approach between home birth (HOME) and birth center (CENTER), we find that for our low risk populations, Conservative Exclusions and Liberal Exclusions, both categories have slightly but not significantly lower ISPMRs than the hospital (HOSPITAL) category (Tables 29, 30). Looking at the entire Study Population and the Study Population with just births below 2,500 grams excluded, we find that the birth centers (CENTER) still have lower ISPMRs than the hospital (HOSPITAL), while the home birth category (HOME) now has slightly but not significantly higher ISPMRs than the hospital (HOSPITAL) (Tables 31, 32).

The next level of our analysis looks at how the overall mortality of a specific category is influenced by its individual components: those births which were planned for the respective birth setting, transfers from home (BIRTHSET=3) or birth center (BIRTHSET=7) respectively to a hospital, or unplanned home birth (BIRTHSET=11) (see Table 15 for birth setting codings45).

1. Study Population after Conservative Exclusions (Table 29): The planned/actual home (BIRTHSET=1) and planned/actual birth center (BIRTHSET=5) cases have not significantly lower ISPMRs than the planned/actual hospital cases (BIRTHSET=9). While transfers from home to hospital (BIRTHSET=3) for the population of this risk level did not have any perinatal deaths, transfers from birth centers (BIRTHSET=7) did have a few deaths and a higher ISPMR than the hospital category, but not significantly. Births originally planned for hospital but actually occurring at home because the women did not make it to the hospital in time (unplanned home birth, BIRTHSET=11) also have a higher ISPMR, but this does not reach significance. For this population segment none of the positive or negative deviations from the regular planned hospital birth is significantly different (at p=0.01). For the population after conservative exclusions we have to conclude that a woman who opts for a natural childbirth approach will have basically the same (slightly though not significantly lower) ISPMR as a woman who opts for a hospital birth. This conclusion also holds for the home birth or birth center subcategory of natural childbirth individually when compared to hospital birth.

2. Study Population after Liberal Exclusions (Table 30): Again the planned/actual home (BIRTHSET=1) and birth center (BIRTHSET=5) cases have lower ISPMRs than the planned/actual hospital cases (BIRTHSET=9), but these do not reach significance. While overall the home birth category (HOME) for this population has a lower ISPMR than the hospital category (HOSPITAL), the transfers from home to hospital (BIRTHSET=3) now have a significantly higher ISPMR. The transfers from birth center to hospital (BIRTHSET=7) and the unplanned home births (BIRTHSET=11) also have somewhat higher ISPMRs, but not at the level of significance. For the population after liberal exclusions, we have to conclude that a woman who opts for a natural childbirth approach will have basically the same (slightly though not significantly lower) ISPMR as a woman who opts for an obstetric

hospital birth, including the unfavorable results from transfers from home to hospital. This conclusion also holds for the home birth subcategory of natural childbirth compared to hospital birth. In this segment of the Study Population we see for the first time that the transfers from home to hospital have a rather high ISPMR.

1. Study Population =2,500 Grams (After Excluding Birthweight <2,500 Grams) (Table 31): Again the planned/actual home (BIRTHSET=1) and birth center (BIRTHSET=5) cases have lower ISPMRs than the planned/actual hospital cases (BIRTHSET=9), still not significant. And again we observe that the transfers from home to hospital (BIRTHSET=3) have a significantly higher ISPMR, now even bringing the overall home birth category (HOME) to a slightly though not significantly higher ISPMR than the hospital category (HOSPITAL). As to the transfers from birth center to hospital (BIRTHSET=7) and the unplanned home births (BIRTHSET=11), they continue to show somewhat higher ISPMRs, but not significantly.

For the Study Population only excluding birthweight <2,5000 grams, we conclude again that a woman who opts for a natural childbirth approach will have basically the same (slightly though not significantly lower) ISPMR as a woman who opts for a hospital birth, including the unfavorable results from transfers from home to hospital. This conclusion also holds for the home birth subcategory of natural childbirth compared to hospital birth, although the ISPMR is now slightly though not significantly higher for the home birth.

4. Entire Study Population (Without Any Exclusions)(Table 32): Even though we are now including all risk profiles in the study population, again the planned/actual home (BIRTHSET=1) and birth center (BIRTHSET=5) cases have lower ISPMRs than the planned/actual hospital cases (BIRTHSET=9), but now approaching significance. The patterns from the previous study population segment repeat themselves: the transfers from home to hospital (BIRTHSET=3) have a significantly higher ISPMR, again bringing the overall home birth category (HOME) to a slightly though not significantly higher ISPMR than the hospital category (HOSPITAL). As to the transfers from birth center to hospital (BIRTHSET=7) they continue to show a somewhat but not significantly higher ISPMR, while the higher ISPMR of the unplanned home births (BIRTHSET=11) is now approaching significance. So even for the entire Study Population including all risk categories we have to conclude that a woman who opts for a natural childbirth approach will have basically the same (slightly though not significantly lower) ISPMR as a woman who opts for a hospital birth, including the unfavorable results from transfers from home to hospital.

Logistic Regression - Results

The indirect standardization controls directly only for the six risk adjusters birthweight, race, sex, age, education/insurance, and controls indirectly through the three sets of exclusions of cases with certain risk factors. In the logistic regression models we enter all potential risk factors (see Table 14 for a complete listing and Table 24 for the dummy variables with their respective reference category) to see whether there is a main effect for the out-of-hospital birth settings home (LOGLOC_H/Home) and birth center (LOGLOC_C/Center) compared to hospital (reference category) with respect to perinatal mortality. We use the same sets of exclusions for lower risk populations as in the indirect standardization. We ran all models twice, once with home (LOGLOC_H/Home) and birth center (LOGLOC_C/Center) as separate variables and once by combining both settings into one variable for natural childbirth (LOGLOC_N/Home+Center). This allows us to separate out the individual effects of the home setting and the birth center setting. The categories used for Home, Center, and Hospital are identical to those used in the indirect standardization (see also Table 15).

The results for the logistic regression models repeat our previous findings: on all four risk levels of our Study Population the hypothesis, that the natural approach to childbirth is as safe as the obstetric approach, cannot be rejected (Table 33). For our three lower risk populations (defined by the three different exclusion sets) as well as for the entire Study Population (without any exclusions) the perinatal mortality of the natural childbirth category (LOGLOC_N/Home+Center) is slightly but not significantly lower than for the hospital (odds ratios below 1). While the home and birth center settings separately consistently show lower mortality than the hospital, these differences are also not significant. While in the indirect standardization the home birth category separately had slightly though not significantly higher perinatal mortality in the higher-risk levels of our Study Population (Study Population =2,500 and entire Study Population without any exclusions), in the logistic regression controlling now for all risk factors we find that the home birth category separately has consistently lower perinatal mortality. We did not run the logistic models with all eleven birth settings separately since we did not expect this to produce any new insights beyond what we already saw in the indirect standardization. As for the indirect standardization we also observe for the logistic regression that the natural approach to childbirth is not only effective for the low-risk population segments but also for the higher risk-level segments of our Study Population.

The above discussed results were based on models which included all risk factors in our database (see Table 14 for a complete listing). To test whether our hypothesis would also hold for a scenario where we could only use risk assessment information available up to the onset of labor, we categorized the variables in our database into risk factors available antepartum and intrapartum (see Table 14). In addition we eliminated those risk factors which did not have significant coefficients in our univariate analyses. We then ran logistic regressions on the antepartum variables alone, and also on the intrapartum variables alone and the sum of the antepartum and intrapartum variables. The results for the antepartum models are shown in Table 34. All main effects for the natural childbirth/out-of-hospital settings (LOGLOC_N/Home+Center) show lower perinatal mortality, though no statistical significance. The same is true for all models with the antepartum and intrapartum variables combined. And in the models with only the intrapartum variables, only the Liberal Exclusions model has a minimally positive main effect (.0018 with p=0.911) for natural childbirth. We are mentioning these analyses here only to provide some insight into the stability of the logistic regression findings when using different sets of risk factors.

We mentioned already that even for the entire Study Population before excluding any high-risk pregnancies we find slightly though not significantly lower perinatal mortality for the natural approach. When we now look at our three different exclusion sets for higher-risk pregnancies (excluding only low birthweight <2,500 grams, liberal exclusions, and conservative exclusions) it is not surprising that logistic regressions on the excluded, presumably higher-risk cases show the same slightly though not significantly lower mortality for the natural approach for all there exclusion sets. This just reinforces our overall conclusion that our findings are not an artefact of our choice of exclusion criteria for higher-risk pregnancies.

THE OBSTETRIC APPROACH IN A LARGER PERSPECTIVE

Examining the findings from the indirect standardization and the logistic regression for the 1989 and 1990 California data, we conclude that low-risk women who opt for a natural childbirth approach in an out-of-hospital setting will experience the same, possibly a somewhat lower perinatal mortality than low-risk women who opt for a hospital birth under the management of an obstetrician, including the unfavorable results for transfers from home to hospital. Both the indirect standardization with only the six risk adjusters (birthweight, sex, race, age, education, insurance) as well as the logistic regressions including all risk factors in the models show generally slightly though not significantly lower perinatal mortality for the natural approach. Our expectation is that this small difference is likely a little larger if one could identify those births that occur today in midwife managed in-hospital birth centers and properly extract these births from the hospital category and allocate them to the natural approach. In our database they are included in the hospital category, which most likely benefits from their superior outcomes (MacDorman and Singh 1998). We also need to keep in mind that the natural approach, while operating today in the United States under suboptimal conditions (see p. 62), still is able to produce these results. We would expect the natural approach when being part of a shared maternity care system to produce even better results. Under no circumstances do the California data for 1989 and 1990 allow the obstetric profession to uphold the claim that for the large majority of low-risk women hospital birth is "safer" with respect to perinatal mortality. Our data also suggest that even for the high-risk levels of our Study Population the natural approach (including transfers) produces the same perinatal mortality outcomes as the obstetric approach.

Having established that the obstetric approach cannot claim to have lower perinatal mortality rates than the natural approach to childbirth, we now need to take the recommendations of the Government's Panel on Cost Effectiveness in Health and Medicine to heart and begin to frame the comparison of the obstetric and natural approach to childbirth in a larger perspective (Gold et al. 1996; Manning 1998). We do not attempt to develop a comprehensive framework for cost-benefit (or cost-effectiveness) analysis of the two approaches to childbirth. Just looking at some of the cost elements and long-term consequences of the unnecessary obstetric interventions for low-risk women suggests that a large cost to society is at stake, and we will develop very crude first estimates for the order of magnitude of these costs. The magnitude of these costs emphasizes the necessity for society to start taking a closer look at the question whether the obstetric approach to childbirth as practiced today is not harmful. After taking a brief look at the question why the obstetric approach is so dominant, we will conclude by suggesting a few steps for further investigation.

A LARGE COST TO SOCIETY IS POTENTIALLY AT STAKE

If we forget for a moment the larger ramifications of birth as a cultural event (Davis-Floyd 1992; Davis-Floyd 1994, we will discuss these aspects later in the chapter on dominance of the obstetric approach) and just look at the individual birth event with its two main decision makers, the pregnant woman and the obstetrician, then it is difficult to see where the benefits of the interventionist obstetric approach are. Let us assume we have neutralized the erroneous beliefs about lower perinatal mortality and the medical necessity and effectiveness of the obstetric interventions. What remains as benefits of the interventionist obstetric approach? For the woman probably what is often called the "convenience factor", including the obstetric pain management, the "birth without pain", and the possibility to time her delivery with an elective cesarean. For the obstetrician the "convenience" of managing his workload through induction of labor and the timing of cesareans, a possible incentive from higher fees for interventions46, and possibly a reduction of his exposure to litigation (malpractice). In a proper cost-benefit or cost-effectiveness analysis (Drummond et al. 1997; Garber and Phelps 1997; Gold et al. 1996; Meltzer 1997; Owens 1998) it will be difficult to assign a value to the "convenience" factors. While the obstetricians' desire for higher fees would not be included in such analyses on the societal level, the malpractice issue could be included47 by either assuming one continues to pay the malpractice premiums as before or by including an amount equaling the annual malpractice awards.

These benefits have to be weighed against the price society pays for subjecting 99% of all births to the interventionist obstetric approach instead of having a shared maternity care system where midwives are the primary caregivers attending to low-risk women (the majority of all pregnancies) and using the natural approach while obstetricians use their interventionist approach only for the remaining cases with complications. This price to society has three major components:

1. Cost savings from unnecessary obstetric interventions.

2. Cost savings from a shared maternity care system.

3. Cost savings from a reduction of the negative long-term consequences of obstetric interventions.

We will limit our discussion of cost-effectiveness here to these three components because they stake out the order of magnitude. A comprehensive cost-benefit analysis would also have to look at the benefits of the natural approach beyond direct cost savings, some of which are the empowering experience for the woman from the natural approach, or the closer bonding with her child and improved relationship with the father after birth.

Cost Savings from Unnecessary Obstetric Interventions

Let us first take a look at the most expensive obstetric intervention, the cesarean section. Comparing the use of cesarean sections between the natural approach and the obstetric approach for our California Study Population (the 71% of all California pregnancies eligible for the natural approach), we find an overall cesarean rate of 6.3% for the natural approach caregivers and 22.1% for the obstetricians (see the OBCSRs in Table 35). This means that natural approach caregivers needed cesarean sections only in 28.4% of the cases where obstetricians deemed them appropriate. After controlling for the six risk adjusters of the indirect standardization (birthweight, sex, race, age, education, insurance) we get the same result (ISCSR=0.2757 for Home+Center category, or 27.6% of the Hospital category, see Table 35). This difference between the two approaches remains the same for the lower-risk populations (Table 35), except for our lowest-risk population (conservative exclusions) where the natural approach had cesareans in 36.0% (34.7% after control for risk factors) of the cases of the obstetric approach. If we look only at primary cesareans for our study populations we find that natural approach caregivers need to use this procedure only in about 40% of the cases compared to obstetricians in the hospital. Already in 1980, before the height of the cesarean epidemic, a study of 4,500 women with risk factors predisposing to cesarean compared to women without these risk factors found virtually identical cesarean section rates in both groups (Albers and Savitz 1991). Both findings, as many other studies, suggest that there is a large amount of unnecessary cesarean sections in today's maternity care system.

After a peak cesarean section rate of 24.7% of all deliveries in 1988, the rate for the United States had declined to 20.8% in 1995. The rate for 1996 was up to 21.8% and the decline appears to be stalling (Curtin and Kozak 1998). While everybody agrees that the cesarean section rate in the United States is too high, prompting the Public Health Service to set a year 2000 goal of 15% (Public Health Service 1990), the discussion is still going on of what the appropriate rate should be. Goer has summarized this discussion quite aptly and concludes that the appropriate rate should lie within a few percentage points on either side of 10%.

What, then, is a reasonable rate? Over a decade ago, the World Health Organization concluded that since countries with some of the lowest perinatal mortality rates in the world had cesarean section rates of less than 10%, there was no justification for any region to have a cesarean rate more than 10- 15% [World Health Organization (WHO), 1985 #160]. Dr. Edward Quilligan, long-time editor of the American Journal of Obstetrics and Gynecology, estimated that based on occurrence rates of various complications of labor, cesarean section rates should range between about 7.8% and 17.5%, depending on the mix of low- to high-risk patients (Quilligan 1995). However, several U.S. hospitals serving primarily high-risk, low-income women have been able to maintain cesarean rates in the 10% to 12% range without any detriment in newborn outcomes, which suggests that Quilligan's maximum is an overestimate (Haire and Elsberry 1991; Myers and Gleicher 1988; Sanchez-Ramos et al. 1990). The authors of A Guide to Effective Care in Pregnancy and Childbirth, which summarizes the conclusions of ongoing analyses of the best medical research, observed that little improvement in outcomes occurs when cesarean rates rise above about 7% (Enkin et al. 1995). A pair of British researchers performed an estimate similar to Dr. Quilligan's, and they also concluded the national cesarean rate should be about 7% (Francome and Savage 1993). Taking these opinions all together, the appropriate cesarean rate should lie within a few percentage points on either side of 10% (Goer ).

Assuming 10% as an appropriate cesarean section rate, and using the below discussed48 average cost for vaginal birth ($6,934) and cesarean section ($11,498), our model calculation would suggest a price of $2.154 billions for unnecessary cesarean sections per year.

In addition to cesarean sections we have shown that a large percentage of the other obstetric interventions (induction or augmentation of labor, electronic fetal monitoring, analgesia including epidurals, episiotomies, forceps, vacuum extractions) must also be considered unnecessary. If we accept Odent's experience at the Pithiviers Hospital, most of them are not necessary at all (see Table 10). Again, using our model calculation, and assuming that 50% of all obstetric interventions (other than cesarean section) are unnecessary, the price for these unnecessary interventions would be $5.108 billions per year.

We should keep in mind that this is just the economic aspect of the unnecessary interventions. Of more concern are the mortality (although probably rather low) and morbidity as immediate consequences of these unnecessary interventions with their associated discomfort and pain. A detailed analysis of mortality and morbidity of unnecessary interventions would go well beyond the scope of this study (Chalmers, Enkin and Keirse 1989; Enkin et al. 1995; Goer 1995). Two examples on the extreme ends of the spectrum illuminate the order of magnitude involved. Maternal deaths as a consequence of unnecessary cesarean sections is estimated at 125 per year (Goer 1995, p. 25). Discontinuing to make the timing of elective cesareans dependent on vacation and weekend schedules and performing elective cesareans no longer before the onset of labor but at term would yield a significant reduction in neonatal respiratory morbidity (Morrison, Rennie and Milton 1995). Klaus, commenting on this paper, calculates that based on the UK figures about 8,000 -10,000 neonatal respiratory distress admissions to NICUs could be prevented annually in the United States (Fanaroff and Klaus 1996, p. 97). To the direct cost savings one would have to add the cost for the associated discomfort and pain as well as lost productivity. A crude guess would be that the figure could easily add up to $1 billion.

Cost Savings from Shared Maternity Care System

The Netherlands has a shared maternity care system where independent professional midwives and to a smaller degree general practitioners provide primary care (Keirse 1982; Wiegers, van der Zee and Keirse 1998) and select pregnant women on risk factors using an official list of medical indications for transferring higher-risk pregnancies to the secondary care of obstetricians. The underlying philosophy is that childbirth is a natural event requiring obstetric interventions only in narrowly defined specific cases of complications (Keirse 1982). For our model calculation we will use the 1991 distribution for natural approach caregivers vs. obstetricians and out-of-hospital births vs. hospital births in this Dutch shared maternity care system (Wiegers, van der Zee and Keirse 1998, p. 194).

To estimate the annual cost of maternity care in the United States for the present system (Model 1) and for a hypothetical model of shared maternity care (Model 2), we are using a cost analysis from the Health Insurance Association of America (HIAA) for 1989 (Minor 1989). We are then updating these 1989 cost figures to 1996 cost by using 1996 data for uncomplicated vaginal birth and cesarean section from the MetraHealth insurance group (Mushinski 1998) which show a 60% cost increase over this period. This leads to the following average cost figures for the United States which are in line with the findings that Rooks reports from several other studies (Rooks 1997, pp. 388-389) :

1989 1996

1. Obstetric approach: Vaginal delivery in hospital49 $4,334 $6,934

2. Obstetric approach: Cesarean section $7,186 $11,498

3. Natural approach: Free-standing birth center $2,111 $3,378

4. Natural approach: Midwife's service at home $994 $1,590

Model 1: Assuming roughly 4,000,000 births in the United States per year, with 99% today occurring in hospitals under obstetricians' care and one half percent each in free-standing birth centers and at home under the care of a natural childbirth caregiver, and taking the 1996 cesarean section rate of 21.8% (Curtin and Kozak 1998) and the above developed 1996 cost figures for maternity care, the total expenditures for maternity care under today's system would then amount to $31.538 billions. Unfortunately there is no national expenditure figure for obstetric care available against which one could validate this estimate. The expenditures for obstetric and neonatal care were estimated at $26.5 billions for 1987 (Gabay and Wolfe 1995, p.1). Using the same 60% increase we used above for the period 1989-1996, that would amount to $42.4 billions in 1996 dollars. It looks like our estimate would fall on the conservative side.

Model 2: If the United States had a shared maternity care system as The Netherlands with 55% of all birth attended by obstetricians in the hospital (vaginal births and cesareans) and 45% by a natural approach caregiver either in the hospital (20%) or at home (25%) (Wiegers, van der Zee and Keirse 1998), and assuming that the natural approach caregivers when delivering in the hospital charge the same as the free-standing birth centers, then - everything else equal - the total expenditures for this shared maternity care system would have been $23.526 billions That is $8.012 billions or 25.4% less than the present obstetric care system (Model 1).

Model 3: Both previous models assume today's prevailing interventionist philosophy with 21.8% cesarean sections and the large number of other unnecessary interventions. We now assume that obstetricians also subscribe to the philosophy of birth as a natural event with interventions only when absolutely necessary (see Odent's philosophy), and reduce our cesarean section rate to 10% and assume that obstetricians will reduce the other interventions for their share of vaginal deliveries by 50%. The total cost for a shared maternity care system with the lower intervention rates would be $18.395 billions. That is $13.143 billions or 41.7% less than the present hospital-based obstetric care system (Model 1).

Long-term Consequences of Perinatal Interventions

While the above developed rough estimates for cost savings from a natural approach to childbirth are based on a body of studies analyzing prevalence of events and costs associated with different forms of maternity care, the research field of long-term consequences of perinatal interventions is still emerging. Although the Committee on Assessing Alternative Birth Settings (Institute of Medicine and National Research Council 1982) suggested already in 1982 in their recommendations for future research design for evaluating alternative birth settings to also include psychological variables and their effects, we could not find studies that would allow us to develop similar crude cost estimates for the long-term consequences. The studies available in this field, however, suggest that there might be a very large cost to society involved.

Odent has recently compiled a database of studies that explore the long-term consequences of interventions women are subjected to in the perinatal period (Odent 1998). We will limit our brief and incomplete review only to the obstetric interventions during labor and delivery, excluding other prenatal or neonatal events and their possible effects. Many of these findings are still controversial but the growing body of research is providing increasingly stronger evidence. We will just present a few studies indicating the possibility of self-destructive adult behavior, severe birth trauma and resulting behavior problems, potential for improved spousal relationships, and potential for deeper bonding and thus a more productive relationship between mother and infant.

1. The work of Jacobson and a group of researchers in Sweden is representative of this new generation of research. They studied teenage and adult drug addiction in relation to the drugs that had been given to the subject's mother when she was in labor. The main finding was that when drugs (such as morphine, Demerol, barbiturates, laughing gas) had been administered to the mother in labor, the child was more likely to become drug addicted in later life than a contemporary whose mother had had a drug-free labor (Jacobson et al. 1987).

2. While the Swedish researchers also found an association between suicide and drugs administered during labor (Jacobson et al. 1987), American researchers showed a significant relationship between suicide and resuscitation and respiratory distress >1 hour (Salk et al. 1985), both frequent consequences of obstetric interventions.

3. A study of male births in Copenhagen found as the main risk factors for being a violent criminal at age 18 the association of birth complications together with early maternal rejection. Early maternal rejection by itself is not a risk factor. But as we discussed earlier, the intervention oriented obstetric approach frequently creates the complications (Raine et al. 1994).

4. In the seventies associations for pre- and perinatal psychology were formed in the United States as well as in Europe starting to assemble and develop further a body of research into, amongst others, the psychological effects of obstetric interventions (Chamberlain 1988a; Emerson 1997; Verny and Kelly 1981). Emerson observed that obstetric interventions (anesthesia, induction and augmentation, forceps and vacuum extraction, and cesarean section), qualifying as severe trauma, together with prenatal trauma cause 45% of babies to experience high levels of birth trauma, leading to long-term detrimental adult behavior problems (Emerson 1997, English translation of this chapter). In addition to creating unhappiness, these behavior problems also result in lost productivity.

5. Children born in a certain hospital in Japan were more at risk of becoming autistic. This hospital used an approach combining sedatives, anaesthetic agents and analgesics with induction of labor by oxytocin or prostaglandins a week before the expected date of delivery (Hattori et al. 1991). The Nobel laureate Tinbergen, looking at autism, identified potentially 'autismogenic' factors, among them induction of labor, breathing stimulation at birth (respiratory distress), and difficult forceps delivery (Tinbergen 1983).

6. An important and well researched finding are the effects of adding one-on-one empowering emotional support from an experienced female companion (doula) back into childbirth (Chalmers and Wolman 1993). Randomized studies found shorter labors, less need for many obstetric interventions including cesarean sections, and more and longer breastfeeding (Kennell et al. 1991; Klaus, Kennell and Klaus 1993). But much more interesting are two findings with long-term consequences. A South African randomized controlled trial of doula care showed that six weeks after the birth 71% of women in the doula group reported that their relationship with their husband had improved versus 30% in the control group (no fathers had been permitted during labor in either group) (Klaus, Kennell and Klaus 1993, p. 45). An American randomized trial reports two months after the birth higher mother-infant affectionate interaction scores for the doula group (Landry, McGrath and Kennell 1998). This could mean that the empowering experience of a doula assisted birth has a positive effect on family stability and the mother's parenting abilities, which would be of major value for today's society. Kennell concludes after several years of studying doulas that "if a doula were a drug, it would be unethical not to use it" (Young 1998).

7. Medicalized childbirth with anesthesia, surgical interventions, and protocols about well-baby nurseries, early mother-infant contact, rooming-in, or formula feeding jeopardize bonding and breastfeeding with major long-term negative consequences. They include abandonment of babies, child abuse, less attachment between mother and infant with likely consequences for the entire child rearing process (Kennell and Klaus 1998). The physiological and immunological benefits of extended breastfeeding, closely related to bonding, are well documented.

Michel Odent, the pioneer in and leading proponent of non-interventive childbirth (Odent 1984a), has made an attempt to put these observations about long-term consequences on a theoretical foundation. Odent argues that the primal period, from conception and including fetal life, the time of childbirth, and the period of breastfeeding up to approximately one year of live, is crucial for our future health. It is during the primal period that the different parts of the primal adaptive system "develop and regulate and adjust themselves. At the end of infancy, the primal adaptive system has reached maturity. I call primal health the balancing of the set point levels which have been reached by the end of infancy" (Odent 1986, p. 15).

The primal adaptive system, in his words (Odent 1986, pp. 115, 163-164; Odent 1998, p. 6), refers to the basic adaptive systems involved in what we commonly call health: the nervous system with the primal brain (the hypothalamus and its associated structures), the immune system, and the endocrine (hormonal) system, which should not be viewed as separate systems but as a whole. "Primal adaptive system" is a simpler term to replace such complex terms as "psychoneuroimmunoendocrinologic system". In this system basic hormonal states are permanently maintained or recovered to their set point levels by feedback mechanisms. For example, when the adrenal gland secretes too much cortisol by the standards of the set point levels, the high rate of cortisol will reduce the secretion of hormones by the hypothalamus, which will reduce the activity of the pituitary gland and indirectly the activity of the adrenal gland. The level of cortisol will then return to its set point level. What is important is the time when these set point levels are established: at the beginning of life. The danger is that when there is an imbalance of the hormones during this critical time, the set point level switches can be set too high or too low and thus the hormonal imbalance persists. It is well known that stresses and pharmacological interventions stimulate hormonal reactions and might well create such hormonal imbalances and thus affect the setting of the set point levels.

For Odent health is how well the primal adaptive system works, it is not the absence of disease. At the end of the primal period we are in a basic state of health which he calls "primal health".

"Thus primal health is built [or reaches maturity] at that time when the baby is closely dependent on its mother, first in the womb, then during childbirth, and then during the period of breastfeeding. Everything which happens during this period of dependence on the mother has an influence on this basic state of health, this primal health" (Odent 1986, p. 16).

The book is full of examples how obstetric interventions might well negatively affect the primal adaptive system. A simple example might illustrate such chain effect.

"If obstetrics had the slightest interest in the physiology of childbirth, some routines would be reconsidered, criticized, discussed and used discerningly. Let us take as a simple example the practice of putting a drip (IV) into the arm of a laboring woman [as required for induction or augmentation of labor]. No one has ever proved that the effects of adrenalin secreted when a needle is inserted are negligible. Moreover, it is uncommon to be interested in the undesirable effects of glucose in the drip. Yet it has been discovered only very recently that glucose received in a drip can actually increase sensitivity to pain. Glucose also lowers the level of sodium chloride in the blood of both mother and baby. A low level of sodium chloride in the baby has the effect of making it breathe too fast. There is also evidence that a high level of glucose blocks the synthesis of prostaglandins, whose role is so important during childbirth. The baby reacts to the excess of sugar by activating its pancreas and secreting more insulin. The result is that after the birth the newborn baby is hypoglycemic. This might help explain the increase in neonatal jaundice in the last ten years" (Odent 1986, p. 134).

Odent considers childbirth a very sensitive period with particularly high levels of hormonal activity.

"Childbirth is an extraordinary event for the primal adaptive system. During delivery it is the primal brain which regulates hormonal secretion. The best way not to disturb the activity of the archaic brain is to reduce the inhibitions which come from the new brain, from the neocortex. Above all, physiological childbirth is a change in the state of consciousness, a reduction in the activity of the upper brain" (Odent 1986, p. 68).

Recognizing that "all mammals have some particular strategy so as to not feel observed when giving birth" (Odent 1994, p. xiii), Odent and his midwives in Pithiviers noticed that women prefer to give birth in a small place, warm and semi-dark, with silence and whispered words, in complete privacy but not being alone (apparently disparate needs which only an experienced midwife can satisfy - not a husband), not subject to being guided or observed, completely free to be in any position and to follow their instincts (ibid. xviii). It is obvious that the obstetric approach to childbirth, with the woman in a bed in the middle of the well lit hospital room, attached to monitor and drip and constrained in her movement, under a variety of drugs and awaiting or afraid of surgical intervention, with ever changing caregivers moving in and out and nervous family members and friends around her and frequently cheering her on, must constitute a major disturbance and stress for the primal adaptive system.

In his book Primal Health Odent suggests that disturbances in the primal adaptive system might not only be correlated to the diseases of civilization, like depression, alcoholism, cardiovascular diseases, schizophrenia, obesity, rheumatism, allergies, auto-immune diseases, viral diseases, cancers, but also to our sexual health and our ability to be in relationship, as well as to our social instincts and spirituality.

New research findings on bonding are shedding some light on these hormonal processes and might support Odent's theory. Breast suckling has been shown to stimulate an outpouring of various hormones in both the mother and the infant. Trigger for this hormonal release are the touch on the mother's nipple and the inside of the infant's cheek. While the release of gastrointestinal hormones stimulates growth of the baby's and mother's intestinal villi and increases the surface area and the absorption of calories with each feeding, the release of oxytocin (the love hormone) in the baby's and mother's brain when suckling might explain the increased attachment between mother and infant from breastfeeding. The period during labor and the first minutes and hour of life seem to form a particular "sensitive period" for laying the foundation for the relationship between mother and baby (Kennell and Klaus 1998). Obstetric interventions or protocol, however, all too frequently limit immediate mother-infant interaction and rooming-in. Instead the baby is deposited in a well-baby nursery and bottle-fed. Future research into long-term physiological and psychological consequences of obstetric interventions and protocols should test whether hormonal phenomena are operating and whether they support Odent's "set-point" theory or Kennell's and Klaus' closely related theory of a "sensitive period".

DOMINANCE OF THE OBSTETRIC APPROACH

While there is, as we have shown, sufficient research indicating that the interventionist obstetric approach has serious drawbacks, the medicalization of childbirth in the United States continues with still increasing rates of obstetric interventions (see Table 10).

Armstrong, a midwife to the Amish and emphasizing non-interventionist childbirth preferably at home, was so puzzled by the pervasive and persistent dominance of the medicalization of childbirth that she decided to research this issue.

[She] ruled out the possibility that it was all the physicians' doing. If one in four women had cesareans, then women must, somehow, be in agreement with them. We felt we must be seeing a social pattern, not the raid of a group of misogynists on unsuspecting females. An explanation of birthplace habits, therefore, had to be complex and, even if unintentionally, collusive: Society in general, including mothers, has legitimate, if not particularly healthy, reasons for participating in and perpetuating the highly medicalized birth (Armstrong and Feldman 1990, p. 11).

There have been many studies of the medicalization of childbirth (Armstrong and Feldman 1990; Jordan and Davis-Floyd 1993; Michaelson 1988; Mitford 1992; Wertz and Wertz 1989). Most of them point out that vested powers and economic interests played an important role in this development. Davis-Floyd's anthropological proposal, however, provides probably the most comprehensive explanation, why not just obstetricians but also women and everybody else in our society subscribes to the medicalization of childbirth (Davis-Floyd 1992; Davis-Floyd 1994). Building on Reynolds (Reynolds 1991), she characterizes our culture as a "technocracy", a society organized around an ideology of technological progress. She argues that obstetrical procedures and interventions are rituals that enact the core values of American society,

communicating messages through the body and the emotions concerning our culture's deepest beliefs about the necessity for cultural control of natural processes, the untrustworthiness of nature and the associated weakness and inferiority of the female body, the validity of patriarchy, the superiority of science and technology, and the importance of institutions and machines...Obstetrical interventions are also transformative in intent. At the same time that they attempt to contain and control the inherently transformative process of birth, they also transform the birthing woman into a mother in the full social sense of the word - that is, into a woman who has internalized the core values of American society: one who believes in science, relies on technology, recognizes her inferiority (either consciously or unconsciously), and so at some level accepts the principles of patriarchy (Davis-Floyd 1992, p. 152).

With this birth ritual in place, Rooks observes that

Many women want to have a full, rich childbirth experience and are fascinated and moved by stories and pictures of women having peaceful, obviously deeply satisfying, completely natural birth. Yet few American women believe they can do this themselves. One explanation is that we have lost the cultural knowledge of the normalcy of birth. When most women gave birth at home, few women arrived at their own first labor without having sat with another woman giving birth. This experience supported a cultural belief that birth is normal and that women can and do go through it in ways that are positive, strong, and joyful. This experience has been lost, replaced by the usual television and movie depiction of childbirth as a clamorous emergency in which someone other than the mother, usually a physician, comes through as a hero. With no experience of normal childbirth and common, widespread evidence of the frailty and failures of childbirth [everybody knows somebody who had a cesarean section], it is not surprising that women lack self-confidence (Rooks 1997, pp. 462-463).

Davis-Floyd's theory explains for the United States why obstetricians never had to provide evidence of the efficacy of obstetric interventions before implementing them, and why, despite clear and in the obstetric research community accepted evidence of the ineffectiveness and frequently harmfulness of obstetric interventions (see the Cochrane Pregnancy and Childbirth Database (Enkin et al. 1995)), these interventions not only persist but are still on the rise (see Table 10). No woman who wants to give birth with a natural childbirth caregiver "can fail to be aware that she is battling almost overwhelming forces that drive her to the hospital" (Davis-Floyd 1994, p. 1139). Any embracing of the natural approach to childbirth will clearly require some changes in society's values and beliefs in technocracy. Meanwhile, concerns about escalating health care cost, not quality considerations, might offer the natural approach to childbirth not just a chance for survival but possibly a window for establishing a more solid foothold and a broader basis for future development.

ISSUES FOR FURTHER INVESTIGATION

Some could argue that all evidence required to move toward a system of maternity care shared between the natural approach as the primary level and the obstetric approach for cases with complications as the secondary level is already available. Others might want to see some of the analyses to be replicated or refined. However, already apparent disadvantages of the obstetric approach have such large order of magnitude, that in any clinical trial it would be considered unethical to continue with the obstetric "treatment". Given the firmly entrenched position of the obstetric approach, further investigation will have to focus on verifying the findings on perinatal mortality and probably expanding them to morbidity outcome measures, creating a sense of urgency for change, and pointing the path to how such change could be institutionalized.

Useful next steps would include:

1. Verify the findings about perinatal mortality on a larger and more recent sample by using the recently merged birth certificate and discharge data for California for 1991-1995, ideally finding a way for allocating the in-hospital birth center correctly to the natural approach. A replication of the findings from this study might be helpful given the highly emotional and central role that the safety argument plays in the controversy about the natural approach.

2. Compare the two approaches to childbirth with respect to maternal and infant outcome measures of morbidity, such as infections, birth injuries, neonatal seizures, or neurological condition of the newborn during the first week and later in life, to examine whether the natural approach - while having the same mortality as the obstetric approach -has also the same or even better morbidity outcomes.

3. Design and intensify systematic research on the long-term consequences of obstetric interventions and protocols to gain a better understanding of the physiological and psychological phenomena surrounding childbirth. If it can be shown that the preliminary research findings in this field point into the right direction, then we can expect research findings in this field to create a strong sense of urgency for needed change because of the high cost to society involved.

4. Identify appropriate institutions for delivering the natural approach to childbirth, a central question being whether in-hospital birth centers can deliver that approach or whether they will always have the tendency to subordinate themselves to the obstetric model. Another central question is how to organize the interface between natural and obstetric approach (the transfers) and how to prevent competition between the two approaches to erode a productive cooperation.

5. Develop concepts and programs for moving from the existing obstetric hospital system to a shared system of midwifery-obstetric maternity care, and doing that expeditiously. The question here is what is going to happen with the obsolete part of the existing obstetric hospital infrastructure, the trained obstetric workforce, and the training institutions in place during the period of transition. Such concepts would also have to include suggestions how to influence society's beliefs in what the appropriate approach to childbirth should be.

CONCLUSIONS

This study set out to examine the obstetric profession's claim and society's belief that the obstetric approach to childbirth is safer than the natural approach to childbirth. Based on newly merged birth certificate and hospital discharge data it was possible to identify that part of all pregnancies that could be considered a candidate for natural childbirth and to identify a comprehensive risk profile for each pregnancy. To compare the perinatal mortality for the two competing approaches to childbirth, they were subjected to two statistical analyses: indirect standardization using only the six risk factors birthweight, sex, race, age, education, and insurance as the risk adjusters, and logistic regression including all risk factors available in the database into the regression models.

Examining the findings from the indirect standardization and the logistic regression for the 1989 and 1990 California data, we conclude that low-risk women who opt for a natural childbirth approach in an out-of-hospital setting will experience the same, possibly a somewhat lower perinatal mortality than low-risk women who opt for a hospital birth under the management of an obstetrician, including the unfavorable results for transfers from home to hospital. Our data also suggest that even for the high risk-levels of our Study Population the natural approach (including transfers) produces the same perinatal mortality outcomes as the obstetric approach. Our analyses of the California data for 1989 and 1990 do not support the claim by the obstetric profession that for the large majority of low-risk women hospital birth is "safer" with respect to perinatal mortality. We need to keep in mind that the natural approach, while operating today in the United States under suboptimal conditions (see p. 62), still is able to produce these results. We would expect the natural approach, when being part of a shared maternity care system and supported by society's beliefs, to produce even better results.

Having established that the obstetric approach cannot claim to have lower perinatal mortality rates than the natural approach to childbirth, we took the first steps in developing a comprehensive cost-benefit framework for comparing the present hospital-based obstetric care system with a shared maternity care system where midwives are the primary caregivers attending to low-risk women (the majority of all pregnancies) and using the natural approach while obstetricians use their interventionist approach only for the remaining cases with complications. We assume that in a shared maternity care system all caregivers would subscribe to childbirth being a natural event and unnecessary cesareans and other obstetric interventions would be eliminated. The first observation we make is that a shared maternity care system would lower the cost for childbirth by roughly 40%, or $13.143 billions (see Model 3, p. 161). The second component of our cost-benefit considerations would be the reduction in mortality and morbidity from eliminating the unnecessary cesareans and other obstetric interventions, for which we do not have a cost estimate but which might well add up to $1 billion if we add the associated discomfort and pain into it. The third component are those factors related to the long-term consequences of the present interventionist obstetric approach, such as the possibility of self-destructive adult behavior, severe birth trauma and resulting behavior problems, potential for improved spousal relationships, and potential for deeper bonding and thus a more productive relationship between mother and infant. Even if the discovered associations should account only for a substantively small proportion of the phenomenon in question, still a very large cost to society would be involved. While this is not a comprehensive list of cost-benefit components, the already apparent disadvantages of the obstetric approach have such large order of magnitude, that in any clinical trial it would be considered unethical to continue with the obstetric "treatment"

The obstetric approach as a cultural ritual, however, is deeply rooted in our society's beliefs in science and technology. While in any normal health care evaluation advantages of the order of magnitude as observed for the natural approach would be considered overwhelming, they have up to now not been able to induce those responsible in the maternity care system or those affected by it to question the appropriateness of the obstetric approach for society. It remains to be seen whether developing a comprehensive cost-benefit framework for a shared maternity care system will be sufficient to start a rational debate about the appropriate maternity care system for the United States, or whether this debate will have to wait until society's beliefs in the superiority of science and technology start to change.

APPENDICES

Page

Appendix 1 California Birth Certificate 178

Appendix 2 Worksheet for Hospital Birth Recorder for 181

Communicating with Mother

Appendix 3 Obstetrical Labor and Delivery Summary 182

Appendix 4 Definitions of XXRISK Variables (Pregnancy), 183

Based on Birth Certificate (BXRISK..) and

ICD-9 Codes from Hospital Discharge Abstracts

Appendix 5 Definitions of XXCOMP Variables (Labor and Delivery), 186

Based on Birth Certificate (BXCOMP..) and

ICD-9 Codes from Hospital Discharge Abstracts

Appendix 6 Suggestions to the NCHS for Improving the 189

Information on the Birth Certificate

GLOSSARY

(Taken from Goer 1995)

Abruptio placentae: Placenta separates prematurely from the wall of the uterus, partial or complete

Amniotomy: Artificial rupture of membranes

Analgesia: Pain relief medication

Antepartum: Before labor

APGAR score: Assessing health status of the newborn by scoring for color, heart rate, muscle tone, reactivity, and respiration efforts. Done 1 and 5 minutes after birth. Score 0-10, score less than 7 at 5 minutes indicates need for additional medical care

Augmentation of labor: Stimulation of labor which began spontaneously, usually by intravenous (IV) oxytocin

Ausculation: Listening to fetal heart beat with stethoscope or hand-held ultrasound device

Breech presentation: Fetus is head up instead of the normal head down

Cord prolapse: Umbilical cord comes down ahead of the baby

Delivery, types:

Spontaneous: The mother gives birth without assistance

Assisted or instrumental: Forceps or vacuum extraction

Cesarean section: Major abdominal surgery

Operative: Forceps, vacuum extraction, cesarean section

Dystocia: Difficult childbirth, often used to mean "slow labor progress"

EFM (electronic fetal monitoring): Monitoring fetal heart rate with a machine that either picks up signals through an external ultrasound device or an internal monitor lead attached to the baby's scalp. Contractions can be monitored by an external pressure sensor or an internal pressure catheter. Data are displayed on a monitor and paper tracing

Episiotomy: Perineal incision to enlarge the vaginal opening for childbirth

Intrapartum: During labor

Laceration, Perineal: perineal tears, categorized from 1st to 4th degree, depending on how far the tear reaches into the muscle, sphincter, rectum

Lothotomy position: Positioning the woman for delivery on her back with legs in stirrups

Macrosomia: Overly large babies commonly defined as birthweight =4000 grams

Meconium: Baby's first bowel movement, which accumulates during pregnancy. Sometimes baby passes meconium into the amniotic fluid, a possible sign of fetal distress. If baby inhales (aspirates) meconium into its lungs upon taking its first breath, the irritation and the particles may cause a type of pneumonia

Multipara: Woman who has given birth to ore than 1 child

Nullipara: Woman who has never given birth (= primipara)

Oxytocin: Hormone which stimulates labor

Parity: Denotes the number of previous births

Perineum: Tissue between the bottom of the vagina and anus

Pitocin: Synthetic oxytocin, hormone which stimulates contractions

Placenta previa: Placenta implants to the uterus low, partially or completely overlaying the cervix. This causes hemorrhage when the cervix begins to dilate

Postdates/Postterm: Generally defined as birth after 42 weeks gestation

Postpartum: After birth

Premature rupture of membranes: Spontaneous rupture of the fetal membranes before the onset of labor

Preterm: Birth prior to 37 weeks gestation

Primary cesarean: A first cesarean section, as opposed to a repeat cesarean

VBAC: Vaginal birth after cesarean

REFERENCES

Ackermann-Liebrich, Ursula, Thomas Voegeli, Kathrin Gunter-Witt, Isabelle Kunz, Maja Zullig, Christian Schindler, and Margrit Maurer. 1996. "Home versus Hospital Deliveries: Follow-up Study of Matched Pairs for Procedures and Outcome." British Medical Journal 313:1313-1318.

Alberman, Eva. 1984. "Statistical Comparison of Home and Hospital Confinements." Pp. 93-104 in Pregnancy Care for the 1980's, edited by Luke Zander and Geoffrey Chamberlain. London: The Royal Society of Medicine & The Macmillan Press Ltd.

Alberman, Eva. 1986. "The Place of Birth." British Journal of Obstetrics and Gynaecology 93:657-658.

Albers, Leah L., and David A. Savitz. 1991. "Hospital Setting for Birth and Use of Medical Procedures in Low-Risk Women." Journal of Nurse-Midwifery 36:327-333.

Anderson, Rondi E., and Patricia Aikins Murphy. 1995. "Outcomes of 11,788 Planned Home Births Attended by Certified Nurse-Midwives: A Retrospective Descriptive Study." Journal of Nurse-Midwifery 40:483-492.

Armstrong, Penny, and Sheryl Feldman. 1986. A Midwife's Story. New York: Arbor House.

Armstrong, Penny, and Sheryl Feldman. 1990. A Wise Birth. New York: William Morrow.

Blumberg, Mark S. 1986. "Risk Adjusting Health Care Outcomes: A Methodological Review." Medical Care Review 43:351-393.

Braveman, Paula, Michelle Pearl, Susan Egerter, Kristen Marchi, and Ronald Williams. 1998. "Validity of Insurance Information in California Birth Certificates." American Journal of Public Health 88:813-816.

Buescher, Paul A., Karen P. Taylor, Mary H. Davis, and J. Michael Bowling. 1993. "The Quality of the New Birth Certificate Data: A Validation Study in North Carolina." American Journal of Public Health 83:1163-1165.

Bulgar, Roger J. 1993. "There are Preventing, Curing and Caring! These Three! and the Greatest of These is Caring!!" HSR Report October 1993:10-13

Butler, N., and D. Bonham. 1963. Perinatal Mortality. Edinburgh: Churchill Livingstone.

California Hospital Outcomes Project. 1996. "Maternal Outcomes Following Delivery: Risk-Adjustment Methodology and Preliminary Findings." . Sacramento: Office of Statewide Health Planning and Development (OSHPD), State of California.

Campbell, Rona, and Alison Macfarlane. 1986. "Place of Delivery: A Review." British Journal of Obstetrics and Gynaecology 93:675-683.

Campbell, Rona, and Alison Macfarlane. 1987. Where to Be Born? The Debate and the Evidence. Oxford: National Perinatal Epidemiology Unit.

Campbell, Rona, and Alison Macfarlane. 1990a. "General Practitioner Maternity Units." British Medical Journal 301:983-984.

Campbell, Rona, and Alison Macfarlane. 1990b. "Obstetrics and the General Practitioner." British Medical Journal 301:1103-1104.

Campbell, Rona, and Alison Macfarlane. 1990c. "Recent Debate on the Place of Birth." Pp. 217- 237 in Recent Debate on the Place of Birth, edited by Garcia, Jo, Kilpatrick, Robert, and Richards. Oxford: Clarendon Press.

Campbell, Rona, and Alison Macfarlane. 1994. Where to Be Born? The Debate and the Evidence. 2nd ed. Oxford: National Perinatal Epidemiology Unit.

Campbell, Rona, Alison Macfarlane, and Sandy Cavenagh. 1991. "Choice and Chance in Low Risk Maternity Care." British Medical Journal 303:1487-1488.

Chalmers, B., and W. Wolman. 1993. "Social Support in Labor - A Selective Review." Journal of Psychosomatic Obstetrics and Gynaecology 14:1-15.

Chalmers, Iain. 1978. "Implications of the Current Debate on Obstetric Practices." Pp. 44-54 in The Place of Birth, edited by Sheila Kitzinger and Jon A. Davis. Oxford: Oxford University Press.

Chalmers, Iain, Murray Enkin, and Marc J. N. C. Keirse (Eds.). 1989. Effective Care in Pregnancy and Childbirth. Oxford: Oxford University Press.

Chamberlain, David B. 1988a. Babies Remember Birth. New York: Ballantine Books.

Chamberlain, Geoffrey. 1984. "Modern Obstetrics and Patient Care." Pp. 153-164 in Pregnancy Care for the 1980's, edited by Luke Zander and Geoffrey Chamberlain. London: The Royal Society of Medicine & The Macmillan Press Ltd.

Chamberlain, Geoffrey. 1988b. "The Place of Birth - Striking a Balance." The Practitioner 232:771-774.

Chamberlain, Geoffrey, E. Philipp, B. Howlett, and K. Masters. 1978. British Births 1970 Vol. 2, Obstetric Care. London: Heinemann.

Chamberlain, Geoffrey, and Luke Zander. 1992. Pregnancy Care in the 1990's: Proceedings of a Symposium at the Royal Society of Medicine, April 1991. Carnforth: The Parthenon Publishing Group.

Chamberlain, R., Geoffrey Chamberlain, B. Howlett, and A. Claireux. 1975. British Births 1970 Vol. 1, The First Week of Life. London: Heinemann.

Chambliss, Linda R., Cornelia Daly, Arnold L. Medearis, Mary Ames, Martha Kayne, and Richard Paul. 1992. "The Role of Selection Bias in Comparing Cesarean Birth Rates Between Physician and Midwifery Management." Obstetrics and Gynecology 80:161-165.

Chapman, M. G., M. Jones, J. E. Spring, M. De Swiet, and G. V. P. Chamberlain. 1986. "The Use of a Birthroom: A Randomized Controlled Trial Comparing Delivery with that in the Labour Ward." British Journal of Obstetrics and Gynaecology 93:182-187.

Clark, Kathleen, Chun-Mei Fu, and Claude Burnett. 1997. "Accuracy of Birth Certificate Data Regarding the Amount, Timing, and Adequacy of Prenatal Care Using Prenatal Clinic Medical Records as Referents." American Journal of Epidemiology 145:68-71.

Creasy, Robert K., Beverly A. Gummer, and Graham C. Liggins. 1980. "System for Predicting Spontaneous Preterm Birth." Obstetrics and Gynecology 55:692-695.

Curtin, Sally C., and Lola Jean Kozak. 1998. "Decline in U.S. Cesarean Delivery Rate Appears to Stall." BIRTH 25:259-262.

David, Richard J. 1980. "The Quality and Completeness of Birthweight and Gestational Age Data in Computerized Birth Files." American Journal of Public Health 70:964-973.

Davies, J., E. Hey, W. Reid, and G. Young. 1996. "Prospective Regional Study of Planned Home Births." British Medical Journal 313:1302-6.

Davis-Floyd, Robbie E. 1992. Birth as an American Rite of Passage. Berkeley: University of California Press.

Davis-Floyd, Robbie E. 1994. "The Technocratic Body: American Childbirth as Cultural Expression." Social Science and Medicine 38:1125-1140.

Daviss, Betty-Anne, and Kenneth C. Johnson. 1998. "Eloquent and Documented Defense of Homebirth." Midwives' Alliance of North America Newsletter 16:1, 9-10.

Declercq, Eugene R. 1992. "The Transformation of American Midwifery: 1975 to 1988." American Journal of Public Health 82:680-684.

Declercq, Eugene R. 1993. "Where Babies Are Born and Who Attends Their Birth: Findings From the Revised 1989 United States Standard Certificate of Live Birth." Obstetrics and Gynecology 81:997-1004.

Drummond, Michael F., Bernie O'Brien, Greg L. Stoddart, and George W. Torrance. 1997. Methods for the Economic Evaluation of Health Care Programs. Oxford: Oxford University Press.

Edwards, Laura E., M. Ismail Barrada, Robert W. Tatreau, and Erick Y. Hakanson. 1978. "A Simplified Antepartum Risk-Scoring System." Obstetrics and Gynecology 54:237-240.

Efron, Bradley. 1982. The Jackknife, Bootstrap, and Other Resampling Plans. Philadelphia: Society for Industrial and Applied Mathematics.

Emerson, William R. 1997. "Birth Trauma: The Psychological Effects of Obstetrical Interventions." Pp. ?? in Seelisches Erleben vor und während der Geburt, edited by L. Janus and S. Haibach. Neu-Isenburg: LinguaMed.

Enkin, Murray, Marc J. N. C. Keirse, Mary Renfrew, and James Neilson. 1995. A Guide to Effective Care in Pregnancy and Childbirth. 2nd ed. Oxford: Oxford University Press.

Ernest, J. M., Robert Michielutte, Mary Lou Moore, Paul J. Meis, and Penny C. Sharp. 1988. "Identification of Women at High Risk for Preterm-Low-Birthweight Births." Preventive Medicine 17:60-72.

Eskes, T. K. A. B. 1992. "Home Deliveries in The Netherlands - Perinatal Mortality and Morbidity." International Journal of Gynecology and Obstetrics 38:161-169.

Fanaroff, Avroy A., and Marshall H. Klaus (Eds.). 1996. Yearbook of Neonatal and Perinatal Medicine. St. Louis, MO: Mosby.

Feldman, George B., and Jennie A. Freiman. 1985. "Prophylactic Cesarean Section at Term? (Occasional Notes)." New England Journal of Medicine 312:1264-1267.

Francome, Colin, and Wendy Savage. 1993. "Caesarean Section in Britain and the United States 12% or 24%: Is either the right Rate?" Social Science & Medicine 37:1199-1218.

Freedman, M. A., G. A. Gay, J. E. Brockert, and et al. 1988. "The 1989 Revision of the U.S. Standard Certificate of Live Birth and Death and the U.S. Standard Report on Fetal Death." American Journal of Public Health 78:168-172.

Gabay, Mary, and Sidney M. Wolfe. 1994. "Unnecessary Cesarean Sections: Curing a National Epidemic." . Washington, DC: Public Citizen's Health Research Group.

Gabay, Mary, and Sidney M. Wolfe. 1995. "Encouraging the Use of Nurse-Midwives." . Washington, DC: Public Citizen's Health Research Group.

Garber, Alan M., and Charles Phelps, E. 1997. "Economic Foundations of Cost-Effectiveness Analysis." Journal of Health Economics 16:1-31.

Gjerdingen, Dwenda K. 1992. "Premature Labor, Part I: Risk Assessment, Etiologic Factors, and Diagnosis." Journal of the American Board of Family Practice 5:495-509.

Goer, Henci. 1995. Obstetric Myths Versus Research Realities: A Guide to the Medical Literature. Westport, CT: Bergin & Garvey.

Goer, Henci. in press. The Thinking Woman's Guide to a Better Birth. New York: Perigee.

Gold, Marthe R., Joanne E. Siegel, Louise B. Russell, and Milton C. Weinstein (Eds.). 1996. Cost-Effectiveness in Health and Medicine. New York: Oxford University Press.

Goodrick, Elizabeth, and Gerald R. Salancik. 1996. "Organizational Discretion in Responding to Institutional Practices: Hospitals and Cesarean Births." Administrative Science Quarterly 41:1-28.

Goodwin, James W. 1997. "Where to be Born Safely. Professional Midwifery and the Case against Home Birth." J Soc Obstet Gynaecol Can 19:1179-1188.

Goodwin, James W., James T. Dunne, and Bruce W. Thomas. 1969. "Antepartum Identification of the Fetus at Risk." Canadian Medical Association Journal 101:57-67.

Gould, J. P., C. M. Heilig, and T. L. Blackwell. 1997. "Perinatal Profiles of California Regions and Hospitals 1989-1993." . Berkeley, CA: Maternal and Child Health Program, School of Public Health, University of California, Berkeley.

Greulich, Betsy, Lisa L. Paine, Cindy McClain, Mary K. Barger, Nancy Edwards, and Richard Paul. 1994. "Twelve Years and more than 30,000 Nurse-Midwife Attended Births: The Los Angeles County + University of Soutern California Women's Hospital Birth Center Experience." Journal of Nurse-Midwifery 39:185-196.

Haire, Doris B., and Charlotte Cram Elsberry. 1991. "Maternity Care and Outcomes in a High-Risk Service: The North Central Bronx Hospital Experience." BIRTH 18:33-37.

Halliday, Henry L., Paul K. Jones, and Susan L. Jones. 1980. "Method of Screening Obstetric Patients to Prevent Reproductive Wastage." Obstetrics and Gynecology 55:656-661.

Hattori, R., g. xx, o. tt, and g. dd. 1991. "Autistic and Developmental Disorders after General Anaesthetic Delivery." Lancet 337:1357-1358.

Herman, Allen A. B., Les M. Irwig, and Hennie Groeneveld. 1988. "Evaluating Obstetric Risk Scores By Receiver Operating Characteristic Curves." American Journal of Epidemiology 127:831-841.

Hobel, Calvin J., Marcia A. Hyvarinen, Donald M. Okada, and William Oh. 1973. "Prenatal and Intrapartum High-risk Screening." American Journal of Obstetrics and Gynecology 117:1-9.

Holbrook, R. Harold, Russell K. Laros, and Robert K. Creasy. 1989. "Evaluation of a Risk-Scoring System for Prediction of Preterm Labor." American Journal of Perinatology 6:62-68.

Hosmer, David W., and Stanley Lemeshow. 1989. Applied Logistic Regression. New York: John Wiley & Sons.

Institute of Medicine. 1985. Preventing Low Birthweight. Washington, DC: National Academy Press.

Institute of Medicine and National Research Council, Committee on Assessing Alternative Birth Settings. 1982. Research Issues in the Assessment of Birth Settings. Washington, DC: National Academy Press.

Jacobson, Bertil, Gunnar Eklund, Lars Hamberger, Dag Linnarsson, Göran Sedvall, and Milan Valverius. 1987. "Perinatal Origin of Adult Self-destructive Behavior." Acta Psychiatrica Scandinavia 76:364-371.

Jordan, Brigitte, and revised and expanded by Robbie Davis-Floyd. 1993. Birth in Four Cultures: A Crosscultural Investigation of Childbirth in Yucatan, Holland, Sweden, and the United States. 4th ed. Prospect Heights, IL: Waveland Press.

Kaminski, Monique, and Emile Papiernik. 1974. "Multifactorial Study of the Risk of Prematurity at 32 Weeks of Gestation." Journal of Perinatal Medicine 2:37-44.

Keeler, Emmett, Ed Park, Robert Bell, and Joan Keesey. 1995. "Management and Outcomes of Childbirth Study: Secondary Data Final Report." . Santa Monica, CA: RAND.

Keirse, Marc J. N. C. 1982. "Interaction Between Primary and Secondary Antenatal Care, with particular Reference to The Netherlands." Pp. 222- 233 in Effectiveness and Satisfaction in Antenatal Care, edited by Murray Enkin and Iain Chalmers. London: Spastics International Medical Publications, William Heinemann Medical Books.

Kennell, John, Marshall H. Klaus, Susan McGrath, Steven Robertson, and Clark Hinkley. 1991. "Continuous Emotional Support During Labor in a U. S. Hospital." Journal of American Medical Association 265:2197-2201.

Kennell, John H., and Mashall H. Klaus. 1998. "Bonding: Recent Observations that Alter Perinatal Care." Pediatrics in Review 19:4-12.

Kitzinger, Sheila, and John A. Davis. 1978. The Place of Birth. Oxford: Oxford University Press.

Klaus, Marshall H., John H. Kennell, and Phyllis H. Klaus. 1993. Mothering the Mother: How a Doula Can Help You Have a Shorter, Easier, and Healthier Birth. Reading, MA: Addison-Wesley.

Kloosterman, G. J. 1978. "The Dutch System of Home Birth." Pp. 85-92 in The Place of Birth, edited by Sheila Kitzinger and John A. Davis. Oxford: Oxford University Press.

Kloosterman, G. J. 1984. "The Dutch Experience of Domiciliary Confinements." Pp. 115-125 in Pregnancy Care for the 1980's, edited by Geoffrey Chamberlain and Luke Zander. London: The Royal Society of Medicine & The Macmillan Press Ltd.

Knox, A. J., L. Sadler, N. S. Pattison, C. D. Mantell, and P. Mullins. 1993. "An Obstetric Scoring System: Its Development and Application in Obstetric Management." Obstetrics & Gynecology 81:195-199.

Kotelchuck, Milton. 1994. "An Evaluation of the Kessner Adequacy of Prenatal Care Index and a Proposed Adequacy of Prenatal Care Utilization Index." American Journal of Public Health 84:1414-1420.

Kraemer, Helena C., and Sue Thieman. 1987. How Many Subjects: Statistical Power Analysis in Research. Newbury Park, CA: Sage Publications.

Landry, S. H., S. K. McGrath, and J. H. Kennell. 1998. "The Effects of Doula Support During Labor on Mother-Infant Interaction at 2 Months." Pediatric Research 43:13A.

Levy, Barry S., Frederick S. Wilkinson, and William M. Marine. 1971. "Reducing Neonatal Mortality Rate with Nurse-Midwives." American Journal of Obstetrics and Gynecology 109:50-58.

Luke, Barbara, and Louis G. Keith. 1991. "The United States Certificate of Live Birth: A Critical Commentary." Journal of Reproductive Medicine 36:587-591.

MacDorman, Marian F., and Gopal K. Singh. 1998. "Midwifery Care, Social and Medical Risk Factors, and Birth Outcomes in the USA." Journal of Epidemiology and Community Health 52:310-317.

Macfarlane, Alison. 1989. "Commentary: Revised U.S. Certificate of Birth: A View from England." BIRTH 16:193-195.

Macfarlane, Alison. 1997. "Commentary: The Safest Place of Birth - Is There a Better Analysis than Meta-analysis?" BIRTH 24:14-16.

Magee, Lonnie. 1990. "R2 Measures Based on Wald and Likelihood Ratio Joint Significance Tests." The American Statistician 44:250-253.

Manning, Willard G. 1998. "Improving the Methodology Behind Cost-Effectiveness Analysis." in Frontlines. .November 1998. Washington, DC: Association for Health Services Research.

Maternal and Child Health Branch. 1993. "Tape Documentation of Perinatal Record Format, California Birth Cohort, 1990." . Sacramento, CA: Department of Health Services, State of California.

Mayes, Frances, Deborah Oakley, Barbara Wranesh, Ninfa Springer, Jane Krumlauf, and Roberta Crosby. 1987. "A Retrospective Comparison of Certified Nurse-Midwife and Physician Management of Low-risk Births: A Pilot Study." Journal of Nurse-Midwifery 32:216-221.

McCarthy, Brian J., Kenneth F. Schulz, and Jules S. Terry. 1982. "Identifying Neonatal Risk Factors and Predicting Neonatal Deaths in Georgia." American Journal of Obstetrics and Gynecology 142:557-562.

McDermott, Jeanne, Carolyn Drews, Diane Green, and Cynthia Berg. 1997. "Evaluation of Prenatal Care Information on Birth Certificates." Paediatric and Perinatal Epidemiology 11:105-121.

McKeown, Tom. 1976. The Role of Medicine: Dream, Mirage or Nemesis. London: Nuffield Provincial Hospitals Trust.

McKinlay, John B., and Sonja M. McKinlay. 1990. "Medical Measures and the Decline of Mortality." Pp. 10-23 in The Sociology of Health and Illness: Critical Perspectives, edited by Peter Conrad and Rochelle Kern. Third Edition ed. New York: St. Martin's Press.

Mehl, Lewis E. 1978. "The Outcome of Home Delivery Research in the United States." Pp. 93-117 in The Place of Birth, edited by John A. Davis and Sheila Kitzinger. Oxford: Oxford University Press.

Mehl, L. F., L. A. Leavitt, G. H. Peterson, and D. C. Creevy. 1976. "Home versus Hospital Delivery: Comparison of Matched Populations." in Annual Meeting of the American Public Health Association. Miami Beach, FL.

Mehl-Madrona, Lewis E., and Morgaine Mehl Madrona. 1997. "Physician- and Midwife-Attended Home Birth: Effects of Breech, Twin, and Post-Dates Outcome Data on Mortality Rates." Journal of Nurse-Midwifery 42:91-98.

Meltzer, David. 1997. "Accounting for Future Costs in Medical Cost-Effectiveness Analysis." Journal of Health Economics 16:33-64.

Menard, Scott W. 1995. Applied Logistic Regression Analysis. Thousand Oaks, CA: Sage Publications.

Michaelson, Karen L. 1988. Childbirth in America. South Hadley, MA: Bergin & Garvey.

Michielutte, Robert, J. M. Ernest, Mary Lou Moore, Paul J. Meis, Penny C. Sharp, H. Bradley Wells, and Paul A. Buescher. 1992. "A Comparison of Risk Assessment Models for Term and Preterm Low Birthweight." Preventive Medicine 21:98-109.

Minor, A. F. 1989. "The Cost of Maternity Care and Childbirth in the United States, 1989." . Washington, D.C.: Health Insurance Association of America.

Mitford, Jessica. 1992. The American Way of Birth. New York, NY: Dutton.

Morrison, J. J., J. M. Rennie, and P. J. Milton. 1995. "Neonatal Respiratory Morbidity and Mode of Delivery at Term: Influence of Timing of Elective Caesarean Section." British Journal of Obstetrics and Gynaecology 102:101-106.

Mosteller, Fredrick. 1993. "The Prospect of Data-based Medicine in the Light of ECPC." The Milbank Quarterly 71:523-532.

Mushinski, Margaret H. 1998. "Average Charges for Uncomplicated Vaginal, Cesarean and VBAC Deliveries: Regional Variations, United States, 1996." Statistical Bulletin, Metropolitan Life Insurance Company 79:17-28.

Myers, Stephen A., and Norbert Gleicher. 1988. "A Successful Program to Lower Cesarean-Section Rates." New England Journal of Medicine 319:1511-1516.

Northern Region Perinatal Mortality Survey Coordinating Group. 1996. "Collaborative Survey of Perinatal Loss in Planned and Unplanned Home Births." British Medical Journal 313:1306-9.

O'Reilly, Marina P. 1991. "A Birth Certificate Audit Program in Pennsylvania." Pp. 57-62 in Proceedings of the 1991 Public Health Conference on Records and Statistics. Washington, DC: National Center for Health Statistics.

Oakley, Ann, Lynda Rajan, and Penrose Robertson. 1990. "A Comparison of different Sources of Information about Pregnancy and Childbirth." Journal of Biosocial Science 22:477-487.

Odent, Michel. 1984a. Birth Reborn. New York: Pantheon.

Odent, Michel. 1984b. "How to Help Women in Labour." Pp. 77-90 in Pregnancy Care for the 1980's, edited by Geoffrey Chamberlain and Luke Zander. London: The Royal Society of Medicine & The Macmillan Press Ltd.

Odent, Michel. 1986. Primal Health: A Blueprint for Our Survival. London: Century Hutchinson.

Odent, Michel. 1994. Birth Reborn. 2nd ed. Medford, NJ: Birth Works Press.

Odent, Michel. 1998. "Primal Health Research Database." in Primal Health Research Newsletter. 5.(4).Spring 1998. London: Primal Health Research Centre.

Olsen, Ole. 1997. "Meta-Analysis of the Safety of Home Birth." BIRTH 24:4-13.

Owens, Douglas K. 1998. "Interpretation of Cost-Effectiveness Analysis." Journal of General Internal Medicine 13:716-717.

Phibbs, Ciaran S., Janet M. Bronstein, Eric Buxton, and Roderic H. Phibbs. 1996. "The Effects of Patient Volume and Level of Care at the Hospital of Birth on Neonatal Mortality." The Journal of the American Medical Association 276:1054-1059.

Piper, Joyce M., Edward F. Mitchel, Mary Snowden, Conni Hall, Margaret Adams, and Paula Taylor. 1993. "Validation of 1989 Tennessee Birth Certificates using Maternal and Newborn Hospital Records." American Journal of EpidemiologyAmerican Journal of Epidemiology 137:758-768.

Powell-Griner, Eve. 1986. "Perinatal Mortality in the United States: 1950-81." Monthly Vital Statistics Report (NCHS) 34:1-14.

Public Health Service, Department of Health & Human Services. 1990. Healthy People 2000: National Health Promotion and Disease Prevention Objectives. Washington, D.C.: U. S. Government Printing Office.

Quilligan, Edward J. 1995. "Preface: Defining the "Appropriate" Cesarean Section Rate." Pp. v-vii in Cesaren Section: Guidelines for Appropriate Utilization, edited by Bruce L. Flamm and Edward J. Quilligan. New York: Springer.

Raine, A., a. aa, a. aa, and a. aa. 1994. "Birth Complications Combined with Early Maternal rejection at Age 1 Year Predispose to Violent Crime at Age 18 Years." Archives of General Psychiatry 51:1984-1988.

Reynolds, Peter C. 1991. Stealing Fire: The Mythology of the Technocracy. Palo Alto, CA: Iconic Anthropology Press.

Rice, Alison. 1996. "Midwifery Soon to Be an Option in B.C. for Low-Risk Births." Nursing B C 28:24-26.

Rooks, Judith Pence. 1997. Midwifery and Childbirth in America. Philadelphia: Temple University Press.

Rooks, Judith P., Norman L. Weatherby, and Eunice K. M. Ernst. 1992a. "The National Birth Center Study: Part I - Methodology and Prenatal Care and Referrals." Journal of Nurse-Midwifery 37:222-253.

Rooks, Judith P., Norman L. Weatherby, and Eunice K. M. Ernst. 1992b. "The National Birth Center Study: Part II - Intrapartum and Immediate Postpartum and Neonatal Care." Journal of Nurse-Midwifery 37:301-330.

Rooks, Judith P., Norman L. Weatherby, and Eunice K. M. Ernst. 1992c. "The National Birth Center Study: Part III - Intrapartum and Immediate Postpartum and Neonatal Complications and Transfers, Postpartum and Neonatal Care, Outcomes, and Client Satisfaction." Journal of Nurse-Midwifery 37:361-397.

Rooks, Judith P., Norman L. Weatherby, Eunice K. M. Ernst, Susan Stapleton, David Rosen, and Allan Rosenfield. 1989. "Outcomes of Care in Birth Centers: The National Birth Center Study." The New England Journal of Medicine 321:1804-1811.

Rosenblatt, Roger A., Sharon A. Dobie, L. Gary Hart, Ronald Schneeweiss, Debra Gould, Rina R. Raine, Thomas J. Benedetti, Michael J. Pirani, and Edward B. Perrin. 1997. "Interspecialty Differences in the Obstetric Care of Low-Risk Women." American Journal of Public Health 87:344-351.

Rumeau-Rouquette, Claude, M. Kaminski, and J. Goujard. 1974. "Prediction of Perinatal Mortality in Early Pregnancy." Journal of Perinatal Medicine 2:196-207.

Salk, Lee, Lewis P. Lipsitt, William Q. Sturner, Bernice M. Reilly, and Robin H. Levat. 1985. "Relationship of Maternal and Perinatal Conditions to Eventual Adolescent Suicide." The Lancet :624-627.

Sanchez-Ramos, Luis, Andrew M. Kaunitz, Herbert B. Peterson, Beverly Martinez-Schnell, and Robert J. Thompson. 1990. "Reducing Cesarean Sections at a Teaching Hospital." American Journal of Obstetrics and Gynecology 163:1081-1088.

SAS Institute Inc. 1990. SAS/STAT User's Guide, Version 6. 4th ed. Cary, NC: SAS Institute.

SAS Institute Inc. 1995. Logistic Regression Examples Using the SAS System, Version 6, First Edition. Cary, NC: SAS Institute.

Scherjon, Sicco. 1986. "A Comparison Between the Organization of Obstetrics in Denmark and The Netherlands." British Journal of Obstetrics and Gynaecology 93:684-689.

Schoendorf, Kenneth C., Jennifer D. Parker, Leonid Z. Batkhan, and John L. Kiely. 1993. "Comparability of the Birth Certificate and 1988 Maternal and Infant Health Survey." . Hyattsville, MD: National Center for Health Statistics.

Selwyn, Beatrice J. 1990. "The Accuracy of Obstetric Risk Assessment Instruments for Predicting Mortality, Low Birth Weight, and Preterm Birth." Pp. 39-65 in New Perspectives on Prenatal Care, edited by Irwin R. Merkatz and Joyce E. Thompson. New York: Elsevier.

Simpson, John Andrew, and Edmund S. C. Weiner (Eds.). 1989. The Oxford English Dictionary. 2nd ed. Oxford: Clarendon Press.

Smith, Robin. Health Information Services, Stanford Health Services. July 1997. Personal Communication about: Completeness and Reliability of Specific Information Items on Birth Certificate

Sokol, Robert J., Mortimer G. Rosen, Jean Stojkov, and Lawrence Chik. 1977. "Clinical Application of High-Risk Scoring of an Obstetric Service." American Journal of Obstetrics and Gynecology 128:652-661.

Springer, Nachiel P., and Chris van Weel. 1996. "Home Birth: Safe in Selected Women, and With Adequate Infrastructure and Support." British Medical Journal 313:1276-1277.

Steiger, Carolyn. 1987. Becoming a Midwife. Portland, OR: Hoogan House.

Sullivan, Deborah A., and Ruth Beeman. 1983. "Four Years' Experience with Home Birth by Licensed Midwives in Arizona." American Journal of Public Health 73:641-645.

Taffel, Selma M., Stephanie J. Ventura, and George A. Gay. 1989. "Revised U.S. Certificate of Birth-New Opportunities for Research on Birth Outcome." BIRTH 16:188-193.

Tew, Marjorie. 1977a. "Obstetric Hospitals and General-Practitioner Units: The Statistical Record." Journal of the Royal College of General Practitioners 27:689-694.

Tew, Marjorie. 1977b. "Where to Be Born." New Society 27:120-121.

Tew, Marjorie. 1978. "The Case against Hospital Deliveries: The Statistical Evidence." Pp. 55-65 in The Place of Birth, edited by Sheile Kitzinger and John A. Davis. Oxford: Oxford University Press.

Tew, Marjorie. 1984. "Understanding Intranatal Care through Mortality Statistics." Pp. 105-114 in Pregnancy Care for the 1980's, edited by Luke Zander and Geoffrey Chamberlain. London: The Royal Society of Medicine & The Macmillan Press Ltd.

Tew, Marjorie. 1985a. "Place of Birth and Perinatal Mortality." Journal of the Royal College of General Practitioners 35:390-394.

Tew, Marjorie. 1985b. "Safety in Intranatal Care: The Statistics." Pp. 203-223 in Modern Obstetrics in General Practice, edited by G. N. Marsh. New York: Oxford University Press.

Tew, Marjorie. 1985c. "We Have the Technology." Nursing Times 81:22-24.

Tew, Marjorie. 1986a. "Do Obstetric Intranatal Interventions Make Birth Safer?" British Journal of Obstetrics and Gynaecology 93:659-674.

Tew, Marjorie. 1986b. "Home, Hospital, or Birthroom." The Lancet ii:749-749.

Tew, Marjorie. 1986c. "The Practices of Birth Attendants and the Safety of Birth." Midwifery 2:3-10.

Tew, Marjorie. 1988. "General Practitioner Obstetrics: Does Risk Prediction Work." Journal of the Royal College of General Practitioners 38:521-521.

Tew, Marjorie. 1990. Safer Childbirth? A Critical History of Maternity Care. London: Chapman and Hall.

Tew, Marjorie, and S. M. I. Damstra-Wijmenga. 1991. "Safest Birth Attendants: Recent Dutch Evidence." Midwifery 7:55-63.

Tinbergen, N. 1983. Autistic Children: New Hope for a Cure. London: Allen & Unwin.

Treffers, Pieter E., Martine Eskes, Gunilla Kleiverda, and Dik Van Alten. 1990. "Home Birth and Minimal Medical Interventions." The Journal of the American Medical Association 264:2203-2208.

Treffers, Pieter E., and R. Laan. 1986. "Regional Perinatal Mortality and Regional Hospitalization at Delivery in The Nederlands." British Journal of Obstetrics and Gynaecology 93:690-693.

Turnbull, Deborah, Anne Holmes, Noreen Shields, Helen Cheyne, Sara Twaddle, W. Harper Gilmour, Mary McGinley, Margaret Reid, Irene Johnstone, Geer Ian, Gillian McIlwaine, and C. Burnett Lunan. 1996. "Randomised, Controlled Trial of Efficacy of Midwife-Managed Care." Lancet 348.

Tyson, H. 1991. "Outcomes of 1001 Midwife-Attended Home Births in Toronto." BIRTH 18:14-19.

University of California at Santa Barbara, Community and Organization Research Institute. 1989. "California Maternity Hospital Codes, 1960-1990." . Santa Barbara, CA.

Van Alten, Dik, Martine Eskes, and Pieter E. Treffers. 1989. "Midwifery in the Netherlnds. The Wormerveer Study: Selection, Mode of Delivery, Perinatal Mortality and Infant Morbidity." British Journal of Obstetrics and Gynaecology 96:656-662.

Vedam, Saraswathi, and Yelena Kolodji. 1995. "Guidelines for Client Selection in the Home Birth Midwifery Practice." Journal of Nurse-Midwifery 40:508-521.

Verny, Thomas, and John Kelly. 1981. The Secret Life of the Unborn Child. New York: Dell.

Wagner, Marsden. 1994. Pursuing the Birth Machine: The Search for Appropriate Birth Technology. Camperdown, Australia: ACE Graphics.

Watkins, Margaret L., Larry Edmonds, Anne McClearn, Lynda Mullins, Joseph Mulinare, and Muin Khoury. 1996. "The surveillance of birth defects: the usefulness of the revised US standard birth certificate." American Journal of Public Health 86:731-734.

Wertz, Rose W., and Dorothy C. Wertz. 1989. Lying-In: A History of Childbirth in America. Rev. ed. New Haven, CT: Yale University Press.

Wiegers, T. A., M J N C Keirse, J. van der Zee, and G. A H. Berghs. 1996. "Outcome of Planned Home and Planned Hospital Births in Low Risk Pregnancies: Prospective Study in Midwifery Practices in the Netherlands." British Medical Journal 313:1309-1313.

Wiegers, Therese A., Jouke van der Zee, and Marc J. N. C. Keirse. 1998. "Maternity Care in The Netherlands: The Changing Home Birth Rate." BIRTH 25:190-197.

Williams, Ronald L. 1979. "Measuring the Effectiveness of Perinatal Care." Medical Care 17:95-110.

Williams, Ronald L., Robert K. Creasy, George C. Cunningham, Warren E. Hawes, Frank D. Norris, and Michiko Tashiro. 1982. "Fetal Growth and Perinatal Viability in California." Obstetrics & Gynecology 59:624-632.

Williams, Ronald L., and Roger Wroblewski. 1991. "1984-1988 Maternal and Child Health Data Base: Descriptive Narrative." . Santa Barbara, CA: Community and Organization Research Institute, University of California Santa Barbara.

Woolbright, Louie Albert, and Dorothy S. Harshbarger. 1995. "The Revised Standard Certificate of Live Birth: Analysis of Medical Risk Data from Birth Certificates in Alabama, 1988-92." Public Health Reports 110:59-63.

World Health Organization. 1985a. "Appropriate Technology for Birth." The Lancet II/85:436-437.

World Health Organization. 1985b. Having a Baby in Europe. Copenhagen: WHO Regional Office for Europe.

Young, Diony. 1998. "Doulas: Into the Mainstream of Maternity Care - Editorial." BIRTH 25:213-214.

Zander, Luke. 1984. "The Significance of the Home Delivery Issue." Pp. 126- 132 in Pregnancy Care for the 1980's, edited by Geoffrey Chamberlain and Luke Zander. London: The Royal Society of Medicine & The Macmillan Press.

Zander, Luke, and Geoffrey Chamberlain. 1984. Pregnancy Care for the 1980's. London: The Royal Society of Medicine & The Macmillan Press Ltd.

1 No statistics available.

2 For the effectiveness of this approach and its side effects (see Enkin et al. 1995, Chapter 35; Goer 1995, Chapter 5)

3 Dr. Michel Odent came to Pithiviers, a small town south of Paris, in 1962 to take charge of general surgery at the public hospital, and found that he was also expected to oversee the small maternity clinic. With his midwives he developed an approach based in non-interference with the woman's instinctive knowledge. He found evidence that the instinctive state, that enables women to labor spontaneously, is connected with a particular hormonal balance (Odent 1984a, p. 14) and refused any hormonal intervention. He argues that their "experiences have clearly shown that an approach which "demedicalizes" birth, restores dignity and humanity to the process of childbirth, and returns control to the mother is also the safest approach " (Odent 1984a, p. 16). The excellent results for the Pithiviers population, which was not screened for risk, support this claim:

- Cesarean sections 6.6%

- Vacuum extraction (never forceps) 5.2%

- Episiotomies 6.0%

- Manual removals of placenta 0.9%

- Perinatal death 0.7%

- Infants separated from mother for transfer

to pediatric or intensive care 1.5%

(Odent 1984a, p. 117)

4 Already in 1985, because of evidence that prenatal care by midwives seemed to result in a decrease in low birthweight, prematurity, and neonatal mortality, the Institute of Medicine recommended that "more reliance should be placed on nurse-midwives ... to serve socioeconomically high-risk mothers" (Institute of Medicine 1985, p. 160-161).

5 (Tew 1977a; Tew 1977b; Tew 1978; Tew 1984; Tew 1985a; Tew 1985b; Tew 1985c; Tew 1986a; Tew 1986b; Tew 1986c; Tew 1988; Tew 1990; Tew and Damstra-Wijmenga 1991)

6 A major new British study looking at perinatal mortality of home births, the Northern Region Perinatal Mortality Survey (Davies et al. 1996; Northern Region Perinatal Mortality Survey Coordinating Group 1996), neither includes a proper control group with hospital births nor contains the information about transfers from home to hospital (it makes assumptions) and, therefore, does not really make a contribution to the safety discussion.

7 The study was designed to test the hypothesis that the low cesarean birth rate on the midwifery service was the result of patient selection bias (a selected low-risk population).

8 Of the women who qualified only 59% agreed to participate in the study. The major reason for declining was previous positive experience with epidural analgesia and fetal monitoring and the desire to have those available again.

9 The study actually compared obstetricians, family physicians, and certified nurse-midwives who practiced in urban areas in Washington State. Since there was little difference between the practice patterns of obstetricians and family physicians and since the involvement of family physicians in obstetrics seems to be declining, we will limit our comparison to obstetricians and certified nurse-midwives.

10 While the APGAR scores (see Glossary) were slightly higher for the home delivery group, the umbilical cord pH was lower for this group. The lower pH value might have been due to later clamping in the home group and longer transportation time to the measuring instrument, both factors that can lower pH.

11 Originally the Oxord Database of Perinatal Trials.

12 The Milbank Quarterly (1993;71(3)) devoted recently the major part of an issue to this research synthesis in Effective Care in Pregnancy and Childbirth (ECPC). The term follows the title of Chalmers' et al. book (Chalmers, Enkin and Keirse 1989).

13 Risk assessment instruments published in the English language literature as of 1986 (Herman) and 1989 (Selwyn).

14 The ROC curve is the plot of sensitivity (true positive) against 1 minus specificity (false positive). The area under the curve is a measure of the predictive accuracy of the model: the higher the number (between 0 and 1), the better the predictive power (Herman, Irwig and Groeneveld 1988, pp. 833-836; SAS Institute Inc. 1995, pp. 87-92).

15 When they applied the Goodwin system to their patient population, the Goodwin system had a predictive accuracy of only 0.80 because the original Goodwin population was "heavily biased by abnormal outcome, thus artificially elevating the predictive accuracy" (Knox et al. 1993, p. 197).

16 Wormerveer is a small village north of Amsterdam.

17 See also the rebuttel by Betty-Anne Daviss, The Canadian Midwives' Statistics Collaboration, and Kenneth Johnson (Daviss and Johnson 1998).

18 For a detailed presentation of the data for this research see the chapter: Data Sources. We are using the old definition of perinatal mortality, which was also used for the California Maternal Child Health (MCH) Database (fetal deaths >20 weeks and infant deaths <28 days) and not the now internationally more commonly used WHO definition (fetal death >28 weeks and infant death <7 days). The WHO definition reduces perinatal mortality by roughly 20%. For our calculation we are applying the perinatal mortality rates from our California data (see Table 18) to the entire U.S. population.

19For example, RAND, having provided some milestone studies to the health care debate, has embarked on a major study of the practice of cesarean sections.

20 On the other hand we are observing a large segment of women wanting a medicated, controlled birth (Armstrong and Feldman 1990; Davis-Floyd 1992).

21 RAND did not attempt to link multiple births. The exclusion of multiple births does not affect our research because multiple birth is an exclusion criterion for birth in birthing centers or at home.

22 See Blumberg, who coined this term (Blumberg 1986)

23 Only 4.8% were attended by MDs, and some of those births were probably administered by lay midwives with back-up physicians signing the birth certificate to ensure insurance coverage.

24 The vast majority (77.9%) of these births in free-standing birth centers during 1989 and 1990 were certified by physicians, only 21.5% by Certified Nurse Midwives.

25 Our data show transfer rates from birth centers to hospitals between 20-25%, significantly higher than the rate of 15.8% reported in the National Birth Center Study (Rooks et al. 1989, p. 1807).

26 It should be noted that this category also includes the midwife administered births in in-hospital birth centers, a number that is steadily increasing but which was relatively small during 1989 and 1990 (see also remarks on pp. 64-65).

27 Table 17 takes all the cases that were reported for a specific condition, either from the birth certificate or from the hospital discharge abstract, and calculates the percentages of this total for the three possible sources: condition only reported on the birth certificate (column 1), only reported on the hospital discharge abstract (column 2), reported on both (column 3). For some conditions the three columns do not add up to 100% because these variables use information from similar variables in the XXRISK and XXCOMP category.

28 Since we want to compare the home birth setting to other birth settings, home birth must at least be an option for every case in the study population. The research is thus limited to California residents giving birth in California. Births within California to non-residents and births to Californians at out-of-State locations are excluded.

29 Logistic regressions were run on the models with all risk variables.

30 Classified as non-preventable by R. Harold Holbrook, Jr., MD, Associate Professor of Gynecology & Obstetrics, Stanford University School of Medicine.

31 Missing values for the medical risk factors (XXRISK.., XXCOMP..) could not be identified. If the risk factor was not identified as such in the birth certificate nor in the hospital discharge abstract, we assumed that the risk factor was not existing in this case.

32 The "conservative" and "liberal" set of exclusion criteria was developed with Linda V. Walsh, CNM, Ph.D., Associate Professor, University of San Francisco School of Nursing, Director of Midwifery Services at Stanford, and Chair of the Division of Research, American College of Nurse Midwives (ACNM)

33 The MCH Data Base has been succeeded by the Perinatal Profiles which were published first in fall of 1997 (Gould, Heilig and Blackwell 1997). Data sources, variables, and methods of analysis have basically remained the same.

34 Our research is limited to singleton births only, 97.8% of all births.

35 We also find high predictive power (R2=88.4%) for our data using these same risk adjusters and as "treatments" the different birth settings (while Williams is using individual hospitals as "treatments"). For the individual pregnancy, however, we get a much smaller R2 of 25.6% (comparable information not reported in Williams' analysis).

36 The MCH Data Base uses the average of all California hospitals as the reference base.

37 (Creasy, Gummer and Liggins 1980; Gjerdingen 1992; Goodwin, Dunne and Thomas 1969; Herman, Irwig and Groeneveld 1988; Hobel et al. 1973; Holbrook, Laros and Creasy 1989; Knox et al. 1993; McCarthy, Schulz and Terry 1982; Michielutte et al. 1992; Selwyn 1990)

38 We use standardized rather than raw coefficients because these allow us better to compare the relative effects of independent variables in influencing the prediction of the dependent variable, since they take into account the degree of variation in the independent variables. An independent variable with a very small variance (relative to that of the dependent variable) can have only a very small influence in the aggregate prediction of the dependent variable, no matter how large its raw coefficient. A standardized coefficient, which is a multiplicative function of the independent variable's raw coefficient and its standard deviation, provides an assessment which makes the contributions of the independent variables more directly comparable.

39 When R2adj =0.001 was met on the level of the study population (SP), the criterion was considered to be met, even when the R2adj was below 0.001 on the level of the liberal or conservative exclusions.

40 To substitute ridging for gradient descent convergence, one just adds "ridging" to the list of options. An example of a model specification:

proc logistic descending data=DATA1;

model peridth = logloc_n xxcomp01 xxcomp04 xxcomp05 etc.

/ risklimits ctable lackfit rsq ridging;

41 While we used R2adj =0.001 as inclusion criterion for the univariates and the higher-order interaction terms, we had to settle on the more exclusive R2adj =0.01 as the inclusion criterion for the two-way interactions. The number of interaction terms would have rendered the models unmanageable, given that the larger models are already running up against the limitations of SAS. From the analysis of the two-way interactions in the model we see, that the vast majority of interactions includes low birthweight or short gestation, both exclusion criteria for our low-risk study populations. This difference for the inclusion criterion eliminates 384 two-way interactions, only 47 of which had a R2adj between 0.005 and 0.01 while the others all had a R2adj <0.005. We felt that the low R2adj justified the exclusion.

42 We do keep those interaction terms with low birthweight and low gestation in our regression models for the high-risk populations.

43 The two-way interactions contained in these models had already been analyzed in our analysis of all possible two-way interactions.

44 These are the effects on the regression models including all variables, not just the antepartum and/or intrapartum variables.

45 For all our four levels of analyses the results for BIRTHSET=10 (planned for hospital but actually occurred in free-standing birth center) as a separate category are meaningless because we consider this coding to be incorrect. The cases of BIRTHSET=10 need to be included in BIRTHSET=5 (planned/actual birth center), which would not change the results for BIRTHSET=5 in any of the four analysis levels.

46 Goodrick and Salancik (1996) showed that financial incentives play a significant role for determining the use of cesarean when the woman is at intermediate risk, i. e. when there are no clear institutional standards for the use of cesarean and when the resulting uncertainty allows discretion. Financial incentives do not affect the use of cesarean for high risk or low risk pregnancies where there are clear institutional standards.

47 If, however, one considers malpractice insurance a transfer cost/transfer payment, then it should not be included in CEA (Drummond et al. 1997, pp. 52-53; Gold et al. 1996, p. 184).

48 See section below on Cost Savings from Shared Maternity Care System

49 Figures include hospital and physician charges.

1

- _ -

1

Смотреть полностью


Скачать документ

Похожие документы:

  1. Universal design and accessibility in education literature annotated reference list sorted reverse chronological by author

    Документ
    ... the potential for and accessibility to undergraduate studies, student characteristics and enrollment, the progression of undergraduate studies in universities, recipients of degrees, staff, and ...
  2. University of bedfordshire

    Документ
    ... on ethical considerations in research projects and on requirements for submissions for ethical approval to the ethics committee. ... and Sport in June 2009, joining the Department of Physical Education and Sport Studies from Leeds Metropolitan University ...
  3. Contents (active hyperlinks CTRL + click to follow link)

    Документ
    ... and over, regardless of discipline. In courses on literature, philosophy, and history; sociology, anthropology, and religious studies; women’s studies, American studies, [...] the ...
  4. 1 of 1258 DOCUMENTS The New York Times December 31

    Документ
    ... to the Department of Energy in Washington. ''Radiological surveys show that the site now meets applicable requirements for ... 280 in the Bay Area on the way to Stanford University. ''There was a traffic slowdown for no reason, and ...
  5. 501 of 1231 DOCUMENTS The New York Times August 8

    Документ
    ... merely because of limits on human ingenuity. In agriculture, safety requirements can trump the need for productivity ... , and went to Michigan State University on the G.I. Bill. He studied business, played football and graduated in 1949. Working in ...

Другие похожие документы..