Association Between Stillbirth and Risk Factors Known at Pregnancy Confirmation FREE

The number of stillbirths (fetal deaths at ≥20 weeks' gestation) is nearly equal to the number of infant deaths in the United States (25 894 vs 28 384 in 2005).¹ Although rates of stillbirth decreased in the first part of the 20th century, rates have been relatively stable over the past few decades.² Racial/ethnic disparities have been reported, with stillbirth rates among non-Hispanic blacks being twice those among non-Hispanic whites.²

Most studies of risk factors for stillbirth use vital statistics with limited data. The Stillbirth Collaborative Research Network was created to conduct a detailed, population-based study of stillbirth in selected areas of the United States, with one of the a priori objectives to determine risk factors for stillbirth and reasons for racial disparities. Many of the factors associated with stillbirth need to be addressed early in pregnancy. Although other factors may be important later in pregnancy, clinicians providing obstetrical care frequently spend relatively more time at the initial visits counseling patients regarding their risk of adverse pregnancy outcomes. We focused this initial report on factors that could be ascertained at the start of pregnancy to provide the clinician and patient with population-based data applicable to the first prenatal visits. A subsequent report will address pregnancy predictors of stillbirth risk.

The overall design and methods of the study have been reported.³ The study population consisted of residents in 5 geographic catchment areas defined a priori by county lines. Study participants were recruited at delivery from 59 community and academic hospitals, urban and rural, with an average of 80 000 deliveries per year. Based on prestudy vital statistics data, these hospitals were projected to provide access to at least 90% of stillbirths and 90% of live births for residents within each catchment area.

Participants were enrolled between March 2006 and September 2008. Catchment-area residents who were at least 13 years of age and nonincarcerated were potentially eligible for the study. All eligible women whose delivery resulted in 1 or more stillborn fetuses, and a representative sample of eligible women whose delivery resulted in only live-born infants, supplemented by oversampling of women delivering at less than 32 weeks' gestation and those of African descent delivering at 32 weeks' gestation or greater, were approached for enrollment.³ A stillborn fetus was defined by Apgar scores of 0 at 1 and 5 minutes and no signs of life by direct observation. Deliveries resulting from the termination of a live fetus were excluded. Gestational age was determined by the best clinical estimate using multiple sources, including information from assisted reproduction, last menstrual period, and obstetrical sonograms.⁴ Fetal deaths at 18 or 19 weeks without good dating were also included so that stillbirths beyond 20 weeks but with incorrect dating parameters could be enrolled.

The protocol included an in-hospital interview, medical record abstraction, placental pathology examination,⁵ and biospecimen collection for cases and controls. For stillbirths, a standardized postmortem examination was also performed.^6- 7

The study was approved by the institutional review boards of the clinical sites and associated hospitals and the data coordinating center. A limited Health Insurance Portability and Accountability Act waiver was obtained for surveillance, screening, and minimal data collection from potential participants. An advisory board reviewed the progress and safety of the study. All participants provided written informed consent.

Pregnancy, defined as a case if there were any stillbirths delivered and as a control if all live births were delivered, was the unit of analysis. For this analysis, cases and controls were excluded if the maternal interview or prenatal chart abstraction was missing. The main analysis compared all cases with all controls. To examine risk factors among potentially viable cases and controls not known to be at high risk of stillbirth, further analyses were performed in 2 subgroups: all deliveries at 24 weeks' gestation or greater and deliveries at 24 weeks' gestation or greater after exclusion of intrapartum stillbirths, multiple gestations, and major congenital anomalies. The anomalies excluded in subgroup analysis are listed in eTable 1.

A number of demographic and prepregnancy variables, including sociodemographic, medical and physiological, psychosocial, and reproductive history variables, were selected for analysis based on their potential association with stillbirth. A composite pregnancy history variable was constructed, with categories for nulliparous with and without prior spontaneous loss and for multiparous with no previous spontaneous loss, with spontaneous losses at less than 20 weeks' gestation only, and with prior stillbirth. When data were available from multiple sources, we developed decision criteria to prioritize information. In general, we prioritized data from the maternal interview regarding sociodemographic information and substance use, and data from the chart abstraction regarding medical information. Maternal race/ethnicity was self-reported in response to prespecified options provided by the investigators and evaluated because of known associations with stillbirth.

It was anticipated that approximately 1000 stillbirths would be identified across the 5 catchment areas over a 2-year period and that 700 would enroll. Sample-size considerations were directed at the number of live-birth controls needed, given the number of stillbirths to be studied. A sampling strategy was developed to produce a sample size of approximately 2:1 live births at 32 weeks' gestation or greater per stillbirth, with oversampling of live births in certain subgroups to ensure adequate numbers for stratified analyses. The planned sample size for controls was approximately 1400. Minimum detectible odds ratios (ORs) for 700 stillbirths and 1400 controls with 80% power and a 2-sided α = .05 are 1.3 at 50% exposure prevalence in the controls, 1.4 at 20%, 1.5 at 10%, 1.7 at 5%, and 2.7 at 1%.

The analyses were weighted for oversampling and other aspects of the study design as well as for differential consent based on characteristics recorded in the screened population.³ Analyses were performed using SUDAAN version 10.0.⁸ Crude ORs and adjusted ORs (AORs) and 95% confidence intervals were calculated from univariate and multivariable logistic models. All tests were performed at a nominal significance level of α = .05. All tests with a single degree of freedom were 2-sided.

Multivariable models included variables with P ≤ .20 for univariate association with stillbirth (overall or within the subgroups), after excluding variables missing for more than 5% of participants (paternal education, human immunodeficiency virus [HIV], hepatitis B, hepatitis C) or that were sparse in 1 or more response categories (HIV, hepatitis B, hepatitis C, hemoglobinopathy, syphilis). The multivariable models included adjustment for clinical site. A generalized R² statistic was computed for each multivariable logistic model using the Cox and Snell⁹ coefficient of determination for logistic regression, scaled to a range of 0 to 1.

Positive and negative likelihood ratios (LRs) were computed for factors with an OR greater than 2.0 in any of the multivariable analyses. Each of these factors was dichotomized. Likelihood ratios for each single variable and for combinations of 2 were calculated. A positive LR between 2 and 5 (negative LR between 0.5 and 0.2) yields a small increase (decrease) in the posttest probability of disease; between 5 and 10 (0.2 and 0.1), a moderate increase (decrease); and greater than 10 (<0.1), a large increase (decrease).^10- 11

A woman could be enrolled in the study more than once during the 2.5 years of recruitment. For simplicity, the primary analysis treated each pregnancy as an independent observation. However, analyses were conducted to account for correlation between pregnancies in women who were enrolled twice and to confirm conclusions from the primary analysis.

Study surveillance identified 953 deliveries with stillbirth; in 70%, the mother provided consent for enrollment. Of the 3088 eligible live-birth deliveries, the mother provided consent in 63%. This resulted in a total of 663 stillbirth pregnancies (cases) and 1932 live-birth pregnancies (controls) in the study (Figure).

eTable 2 presents characteristics of eligible women by enrollment status. Women with stillbirths who did not enroll in the study (n = 290) did not differ from those enrolled according to maternal age, maternal race/ethnicity, insurance and method of payment, and gestational age at delivery. Women with live births who did not enroll in the study (n = 1156) did not differ from those enrolled according to maternal age and insurance/method of payment. However, nonenrolled women differed from those enrolled by maternal race/ethnicity and gestational age at delivery.

Table 1 (and eTable 2) shows unweighted and weighted distributions for the enrolled women based on whether they were included in the analysis or excluded because of missing data. The weighted total number of enrolled control pregnancies was 1439. This is smaller than the unweighted number (1932) because of the down-weighting of oversampled groups. Of pregnancies enrolled, 49 cases (7.4%) and 116 controls (6.0%) (weighted, 49 cases [7.4%] and 85 controls [6.3%]) were excluded from the analysis because of missing maternal interview or prenatal chart abstraction. Women excluded did not differ significantly from those included by maternal age, insurance status, or gestational age. Maternal race/ethnicity were similar among excluded and included pregnancies with stillbirths but differed among controls.

Several reproductive history factors were strongly associated with stillbirth by univariate and multivariable analysis (Table 2 and Table 3). In multivariable analyses, these factors included previous stillbirth and nulliparity with and without a history of prior spontaneous losses at less than 20 weeks' gestation and plural birth in the current pregnancy.

In addition, the following maternal factors were independently associated with stillbirth: diabetes, 40 years or older, AB blood type, history of drug use with addiction, history of cigarette smoking during the 3 months prior to pregnancy, being overweight or obese, and not living with a partner. Relative to non-Hispanic whites (33.4% of cases, 45.3% of controls), the crude ORs for stillbirth were 2.78 (95% CI, 2.12-3.66) for non-Hispanic blacks (23.1% of cases, 11.2% of controls) and 1.38 (95% CI, 1.10-1.73) for Hispanics (36.3% of cases, 35.7% of controls). After adjusting for variables in the multivariable analysis, the AOR for stillbirth decreased to 2.12 (95% CI, 1.41-3.20) for non-Hispanic blacks and 1.22 (95% CI, 0.89-1.68) for Hispanics. The generalized R² was 0.19.

To quantify the magnitude of the contribution of prepregnancy risk factors to the black-white disparity in stillbirth risk, we compared the crude OR and AOR of those participants in the multivariable model. Five percent of the live births and 7% of the stillbirths included in the univariate analysis were not included in the multivariable analysis because of missing values in 1 or more of the predictor variables. The crude OR for non-Hispanic blacks was 2.43 (95% CI, 1.82-3.24). Therefore, after adjustment for risk factors determined at the start of pregnancy, the disparity between non-Hispanic blacks and whites was reduced by 22% toward the null value of 1, from a crude OR of 2.43 to an AOR of 2.12.

Nine women were enrolled in the study twice. Each pregnancy was a singleton. Five women enrolled with a live birth for each pregnancy; 2 others had a stillbirth for each pregnancy and 2 a stillbirth followed by a live birth. Taking into account the correlations between the pregnancies for these 9 women produced results (data not shown) virtually identical to the results that treated all of the pregnancies as independent observations.

Of the variables associated with stillbirth in the multivariable analysis of the overall study group, race/ethnicity, drug use, body mass index, and blood type were no longer significant in the subgroups at 24 weeks' gestation or greater, with the AORs particularly diminished for non-Hispanic blacks and AB blood type (Table 4 and Table 5). Nulliparity (overall and with prior losses or termination of pregnancy at <20 weeks' gestation), history of prior stillbirth, plurality, neither married nor cohabitating, and maternal age of 40 years or older remained independently associated with stillbirth, as did diabetes, for which the AOR was substantially larger in this subset (3.47 [95% CI, 1.86-6.49]). The generalized R² for the multivariable model was 0.16.

To examine factors associated with stillbirths among low-risk, potentially viable fetuses, we further limited the subset of deliveries at 24 weeks' gestation or greater to the 311 antepartum stillbirths and 1593 live births with nonanomalous singleton pregnancies (Tables 4 and 5). The generalized R² for the multivariable model was 0.14.

The results for non-Hispanic black women and for women with AB blood group were similar to those for the subgroup of deliveries at 24 weeks' gestation or greater. Advanced maternal age and cigarette smoking were no longer significant, although the AORs were not greatly diminished. Nulliparity and history of prior stillbirth remained strongly associated with increased risk of stillbirth. The independent risk associated with diabetes was again high. The adjusted association for women not living with a partner remained significant. In addition, there was an increased risk associated with being overweight or obese.

Variables for which LRs were computed were age group, race/ethnicity, lifetime drug use, diabetes, pregnancy history, and plurality. Most of the positive LRs were in the 2 to 5 range, whereas most of the negative LRs were near 1 and none were less than 0.5. The positive LRs for combinations of factors were at least 5 for prior stillbirth/no prior live birth plus diabetes and for prior stillbirth/no prior live birth plus plural birth, but each applied to less than a half percent of the general population (Table 6).

This analysis focused on risk factors that would have been known at the start of pregnancy, when most women present to their obstetrical care clinicians and when a viable pregnancy is usually confirmed. Prior stillbirth, prior pregnancy loss at less than 20 weeks' gestation, non-Hispanic black race/ethnicity, nulliparity, plurality, diabetes, maternal obesity, maternal age, marital status, smoking, and illicit drug use were all associated with stillbirth by multivariable analyses. However, these risk factors explained little of the burden of stillbirth.

Pregnancy history was the strongest risk factor for stillbirth. When compared with multiparous women without previous losses at less than 20 weeks' gestation or stillbirths, there was a progressive increase in the risk for stillbirth for nulliparas, then nulliparas with previous losses at less than 20 weeks' gestation, and then multiparas with a previous stillbirth. Women with a previous stillbirth are well known to be at 5- to 10-fold increased recurrence risk for stillbirth and therefore receive more intensive prenatal care and antepartum surveillance in the subsequent pregnancy. Our findings add to the literature by showing that even a nonelective previous pregnancy loss at less than 20 weeks' gestation increased the risk for stillbirth. Because all women begin childbearing as nulliparas, findings from further investigations into the causes and mechanisms of stillbirth among nulliparas could be translated into interventions that may affect overall risk for stillbirth.

The increasing prevalence of diabetes in pregnancy¹² underlines the importance of our finding that diabetes is associated with early as well as later stillbirths. A recent meta-analysis of preconception and prenatal care of diabetic women estimated that 10% of diabetes-associated stillbirth might be prevented with early detection and control.¹³ In this study, the association of maternal obesity with risk of stillbirth was not explained by diabetes or hypertension. Rather, the overall risk of stillbirth increased with increasing body mass index, even after taking into account the increased prevalence of preexisting maternal disease. Maternal overweight and obesity may have the greatest population-attributable risk among potentially modifiable maternal characteristics in the United States.¹⁴

Being unmarried and not cohabitating was a consistent risk factor for stillbirth and remained significant in the multivariable analyses that included other factors. The AOR of 1.62 (95% CI, 1.15-2.27) is higher than the adjusted relative risk of 1.24 (95% CI, 1.21-1.26) for being unmarried, as reported in another study for the United States from 1995-2004.¹⁵ In that study, there was no distinction between unmarried cohabitating or not cohabitating. In our study, which also controlled for additional factors not available on vital records, cohabitating unmarried women were not at increased risk of stillbirth, whereas unmarried women not cohabitating were at increased risk. Screening for partner relationship may prove useful in identifying higher-risk women who would benefit from social support during pregnancy.

Prepregnancy risk factors for stillbirth that have been found to be more frequent among African Americans include young maternal age (<20 years),^16- 17 low maternal education,^18- 19 previous stillbirth,^1,18 previous cesarean delivery,^16,20 obesity,^21- 22 chronic hypertension,¹⁹ diabetes,²³ systemic lupus erythematosus,²⁴ and multiple gestation.²⁵ A few prepregnancy risk factors for stillbirth may be less frequent among African Americans, including smoking²⁶ and nulliparity.¹⁶ To our knowledge, no study to date has systematically examined whether these stillbirth risk factors taken together explain the excess risk of stillbirth for African Americans.² In our study, after adjustment for these risk factors, the relative odds of stillbirth for non-Hispanic blacks compared with whites was reduced by 22%.

The elevated risk of stillbirth for non-Hispanic blacks occurred predominantly prior to 24 weeks' gestation. This is consistent with the finding that the greatest black-white disparity in hazard for stillbirth in ongoing pregnancies is in the interval of 20 to 23 weeks' gestation.¹⁸ Also, very early preterm birth (<28 weeks' gestation) is increased more than 3-fold among African Americans,²⁷ and the as-yet unknown reasons for excess preterm births may also be responsible for risks for early preterm stillbirth. In fact, an examination of the causes of death for the stillbirths in this study revealed a higher proportion of early intrapartum stillbirths among non-Hispanic blacks. The causes were similar to those reported for early preterm birth.²⁸

In addition to non-Hispanic black race, AB blood group appeared to be preferentially associated with stillbirth before 24 weeks. We are unaware of a previous report of this association between AB blood type and stillbirth. Non-O blood groups are recognized risk factors for thrombosis.²⁹ An association between blood type and adverse pregnancy outcomes, such as preeclampsia, has been reported in some studies but not confirmed in a meta-analysis.³⁰ Mechanistic studies contrasting these risk factors may lead to better understanding of the etiology of stillbirth.

Overall, pregnancy factors known at the start of pregnancy accounted for little of the stillbirth risk, as shown by the low generalized R² values. Apart from occurrence of previous stillbirth or pregnancy loss, the other risk factors have limited predictive value, with low positive LRs. This supports the findings of another study that noted the “relative lack of clinical utility” for prediction of term stillbirth when certain prepregnancy risk factors are present in isolation.³¹

Limitations of our study include the retrospective collection of some information (eg, body mass index) by maternal interview when prenatal records were missing some variables. A potential source of bias was that 30% of women experiencing stillbirth and 37% of sampled women with live births did not participate in the study. However, weighting was used to account for differential consent on a number of factors recorded in the screened population.³ In addition, the maternal interview or the chart abstraction was missing for 49 (7.3%) of enrolled stillbirth pregnancies and 116 (6.0%) of live-birth pregnancies, but differences between the enrolled and analyzed population in examined patient characteristics were small. Last, we had limited power to assess rare risk factors or those with only weak associations with stillbirth, particularly in the subgroup analyses.

To our knowledge, our study is the largest population-based study to date of stillbirth with extensive evaluation of cases as well as controls. The design of the study makes it more generalizable than prior hospital-based studies or studies with small sample sizes, less complete ascertainment, nonstandardized workup, or convenience sampling. Most other studies of risk factors for stillbirth have used large databases, including vital statistics, but these databases often contain limited information, particularly on reproductive and medical complications, and accuracy can be suboptimal.³² The detailed characterization of the study participants through maternal interview and chart abstraction enabled us to confirm prior risk factors as well as perform further in-depth analyses of risk factors.

In this study, factors known at the start of pregnancy explained little of the overall burden of stillbirth. Further research is needed to identify pregnancies at highest risk overall and for specific causes. Although some of the risk factors may not be modifiable (eg, race/ethnicity), knowledge of interactions between these factors and other modifiable ones may be useful. In addition, association studies aid in the identification of areas of investigation and preventive approaches. For example, non-Hispanic black women have a more than 2-fold increase in risk of stillbirth. However, the disparity in risk largely occurs at less than 24 weeks' gestation. Therefore, focusing on the pathophysiology of early preterm birth may reduce racial disparity in stillbirth.