To examine temporal trends in stillbirth and its risk factors in the United States (US), and to assess the contribution of labour induction and caesarean delivery to the stillbirth rate.
To examine temporal trends in stillbirth and its risk factors in the United States (US), and to assess the contribution of labour induction and caesarean delivery to the stillbirth rate.
Population-based cohort study based on linked birth-infant death and fetal death data files from the US National Vital Statistics System.
Complete data were available for 44 states and the District of Columbia.
Singleton births from 1992 to 2004.
We assessed changes in stillbirth rates from 1992–1994 to 2002–2004 before and after adjustment for changes in maternal characteristics including maternal age, education, smoking, and medical risk factors, using Cox regression models. We also carried out an ecological study, using states as the units of analysis, to assess the impact on the stillbirth rate of increasing induction and caesarean delivery. Race-specific subgroup analyses were performed and included non-Hispanic Whites and non-Hispanic Blacks.
The stillbirth rate among non-Hispanic White singleton births decreased 11.5% from 1992–1994 (5.2 per 1000) to 2002–2004 (4.6 per 1000). After adjustment for maternal risk factors, the hazard ratio (HR) for 2002–2004 was 1.01 (0.99, 1.03) for gestational age (GA) ≤39 weeks, but 0.92 (0.86, 0.99) at 40 or more weeks. The ecologic analysis revealed a nonsignificant negative correlation of −0.17 (−0.44, 0.13) between state-level changes in stillbirth at GA ≥40 weeks and labour induction. A nonsignificant positive correlation of 0.23 (−0.07, 0.49) was observed between changes in stillbirth at all GAs and caesarean delivery and did not differ at GA ≤39 versus ≥40 weeks. Results were similar among non-Hispanic Blacks.
Changes in maternal risk factors explained the reduction in stillbirth at GA ≤39 weeks but not at ≥40 weeks. The rise in labour induction and caesarean delivery rates did not explain the reduction in stillbirth ≥40 weeks of gestation.
In the last two decades, the rate of preterm birth (<37 completed weeks of gestation) has increased in the USA and other developed countries.[1-7] In the USA, most of this increase has occurred at late preterm (34–36 weeks) gestation.[8, 9] Moreover, the gestational age distribution of term births (37–41 weeks of gestation) has shifted toward lower gestational ages, with a larger proportion of infants delivered at early term (37–38 weeks) gestation. These temporal trends have been attributed to the increasing use of obstetric intervention (labour induction and prelabour caesarean delivery) in late preterm and early term gestations.[6, 7, 10]
On the other hand, stillbirth rates have decreased only modestly over the past 20 years in the USA.[11-13] Some authors have attributed the decrease in stillbirth to an increase in obstetric intervention,[14-16] yet the evidence supporting this is weak. Randomised controlled trials have consistently shown that routine labour induction (versus expectant management) at >41 weeks reduces the risks of stillbirth and perinatal mortality. We are aware of only two randomised trials of induction versus expectant management in late preterm or early term gestations: one in women with gestational hypertension, the other in women with gestational diabetes.[19, 20] Both trials were small, however, and neither observed any stillbirths. To our knowledge, no randomised trials at any gestational age have compared prelabour cesarean delivery with expectant management.
In this study, we examine the temporal trends in stillbirth in the United States from 1992 to 2004, compare stillbirth rates between 1992–94 and 2002–2004, and assess the impacts on those trends of contemporaneous changes in maternal risk factors and in rates of labour induction and caesarean delivery.
Our study is based on linked birth-infant death and fetal death data files from the National Vital Statistics System of the United States, which are available online from Centers for Disease Control and Prevention (CDC) (www.cdc.gov/nchs/data_access/Vitalstatsonline.htm). The linked birth–infant death data provide demographic and health data for births occurring during a calendar year, based on information abstracted from birth certificates filed in vital statistics offices of each state and the District of Columbia. The fetal death files include information from all reports of fetal deaths. Most states report fetal deaths at 20 weeks or more of gestation and/or ≥350 g in birthweight, but a few states report fetal deaths for all periods of gestation. Demographic data include variables such as date of birth, age and educational attainment of the parents, marital status, live-birth order, race, sex, and geographic area. Health data include items such as birthweight, gestational age and prenatal care. Since 1989, a clinical estimate of gestational age has also been recorded, as has the use of labour induction.[2, 21] Data on caesarean delivery have also become available since the 1989 revision, but no information is recorded as to whether the caesarean was performed before or after the onset of labour. Until 2004, linked live birth and fetal death data files included geographic data, including state of birth occurrence; since 2005, however, those geographic data are no longer publicly available. Because a substantial number of states did not report the new (1989) data items from 1989 to 1991,[10, 22] our study is based on the singleton live birth and fetal death data from 1992 through 2004, the last year when the state identification data were publicly available.
Six states were excluded from our analyses for all time periods, owing to missing data on Hispanic origin in fetal death data files in the 1992–94 period: Louisiana, Maryland, Massachusetts, New Hampshire, Oklahoma and Rhode Island. Moreover, Maryland and Oklahoma had missing LMP estimates of gestational age in >40% of fetal deaths and did not report a clinical estimate of gestational age; Louisiana did not report plurality on fetal death data files. Our overall study sample therefore comprised 6 650 475 births in the 1992–94 period and 6 114 413 births in the 2002–2004 period from 44 states and the District of Columbia, 91.8% of total singleton deliveries among non-Hispanic Whites in the two periods. The following ten states were also excluded for multivariate analysis, owing to missing data on maternal medical risk factors or smoking for >40% of fetal deaths: Alabama, California, Connecticut, District of Columbia, Hawaii, Indiana, New York, South Dakota, Texas and Virginia. This reduced sample for multivariate analysis comprised 4 536 115 births in the 1992–94 period and 4 275 460 births in the 2002–2004 period, 64.5% of total singleton deliveries among non-Hispanic Whites in the two periods.
From the US birth certificate, gestational age (GA) is usually calculated from the first day of the mother's last menstrual period (LMP). Gestational age derived from the LMP estimate is prone to error.[23, 24] A clinical estimate of gestation (CE) has also been available since the 1989 revision, except for the state of California. Although the CE is based on the managing clinician's best estimate, including menstrual history, physical findings, laboratory values, and (if available) sonography,[2, 3] no instructions were provided prior to the 2003 revision of the US birth certificate for specifying the basis of the CE. Several methods for validating the LMP-based gestational age have been proposed to reduce misclassification, including that of Platt et al., Zhang and Bowes, Alexander et al. and Qin et al. In Qin et al.'s ‘LMP/CE’ editing method, LMP is replaced by the CE when the two estimates differ by more than 2 weeks. The LMP/CE method has been shown to reduce misclassification of GA and was the basis for gestational age estimation in this study. California did not report a clinical estimate before 2005, and thus LMP cannot be edited in that state.
Our principal analysis was based on singleton births among non-Hispanic Whites. The effect of increased induction might be less effective in reducing stillbirth than it would be in Blacks or Hispanics because Whites already have high induction rates. The restriction by race was also intended to control for potential confounding due to differences in induction and caesarean rates by race and the changing racial composition over time. We compared stillbirth rates, as well as maternal demographic and medical risk factors, between two time periods 10 years apart: 2002–2004 versus 1992–94. As all surviving fetuses are at risk of subsequent stillbirth, our analytical approach was based on fetuses at risk rather than live births at each gestational week. In analyses based on individual women, Cox regression models were used to adjust for fetal sex, parity, and maternal demographic and medical risk factors including maternal age, education, smoking, diabetes, chronic and gestational hypertension.
The association between stillbirth and labour induction or caesarean delivery cannot be ascertained in analyses based on individual women because it would be highly confounded by the medical indication for the procedure. Although maternal risk factors are reported on the US birth certificate, the specific clinical indications for induction or caesarean are not, and thus they cannot be controlled for adequately at the individual level. To reduce confounding by clinical indication, we carried out an ecological analysis[10, 29-31] to assess the impacts of labour induction and caesarean delivery on stillbirth, based on 45 ecologic units in our overall study sample: 44 states plus the District of Columbia. The ecologic study is based on changes in rates of stillbirth, labour induction, and caesarean delivery between 1992–94 and 2002–2004; ecologic correlations between the change in rates of stillbirth and of labour induction and caesarean delivery were calculated, with weighting for the number of births in each state. Although a few states have implemented the 2003 revision of birth certificates since 2003, coding for the 1989 revision was also available in those states, and the comparison between the two periods was therefore based on the 1989 coding.
As a secondary analysis to further reduce confounding by the clinical indication for labour induction or caesarean delivery, we restricted the study sample to women with low-risk pregnancies, defined as maternal age 20–34 years and the absence of diabetes, chronic hypertension and pregnancy-induced hypertension.[32-34] This classification of low risk has been used consistently in our previous studies[7, 35] and by other investigators,[32-34] both in the USA and Canada. All other pregnancies were considered not to be at low risk. Ecologic analysis was repeated and stratified by low versus nonlow risk. Finally, to assess the generalisability of our findings and examine effect modification by race, we repeated our analysis among non-Hispanic Blacks. All data were analysed using SAS version 9.2 (SAS Institute, Cary, NC, USA).
As shown in Supporting Information Table S1 for the overall study sample, mothers were older among 2002–2004 births than among 1992–94 births; the proportion of mothers ≥35 years increased from 11.4 to 14.3%. Mothers had higher education in the later period, with the proportion of college degree or higher increasing from 24.7% in 1992–94 to 30.8% in 2002–2004, Maternal smoking was reduced from 15.2% in 1992–94 to 12.1% in 2002–2004, while the proportion with medical risk factors increased; the prevalence of diabetes, chronic hypertension, and gestational hypertension increased from 2.6, 0.6, and 3.3%, respectively, to 3.2, 0.9, and 4.1%, respectively. Labour induction increased by 61% between the two time periods, from 16.0% in 1992–94 to 25.7% in 2002–2004, while caesarean delivery increased by 23%, from 20.9% to 25.7%. Maternal characteristics and risk factors in the reduced sample for regression analysis were similar to those in the overall study sample in the two periods (Table S1).
Figure 1 shows the GA-specific stillbirth rates, based on fetuses at risk, for the two time periods. As shown in Table S1, the stillbirth rate among non-Hispanic White singleton births decreased from 5.2 per 1000 total births in 1992–94 to 4.6 per 1000 in 2002–2004, a decrease of 11.5% between the two study periods. The stillbirth rate decreased modestly for GA ≤39 completed weeks over the two periods, from 4.7 per 1000 fetuses at risk in 1992–94 to 4.3 per 1000 fetuses at risk in 2002–2004 respectively, a relative 8.5% decrease; while for GA ≥40 completed weeks, the stillbirth rate decreased from 1.0 per 1000 in 1992–94 to 0.8 per 1000 in 2002–2004, a 20% relative reduction. Unadjusted hazard ratios (HR) for stillbirth (2002–2004 versus 1992–94) and their 95% confidence intervals (CI) were 0.92 (0.91, 0.93) overall, 0.93 (0.92, 0.95) for GA ≤39 completed weeks, and 0.83 (0.78, 0.88) for GA ≥40 completed weeks. Stillbirth rates among the reduced sample were also similar to those among the overall sample (Table S1), with the unadjusted HRs for stillbirth (2002–2004 versus 1992–94) 0.93 (0.91, 0.94) for all GAs, 0.93 (0.91, 0.95) for GA ≤39 completed weeks, and 0.87 (0.81, 0.93) for GA ≥40 completed weeks.
Parameter estimates and their 95% confidence intervals from the Cox regression models are presented in Table 1. Although the unadjusted hazard ratio (95% confidence interval) for total stillbirths for 2002–2004 versus 1992–94 was 0.93 (0.91, 0.94), after adjustment for maternal demographic and medical risk factors and fetal sex and parity, the overall risk of stillbirth did not decrease (HR = 1.00 [0.98, 1.02]). The unadjusted reduction in risk for GA ≤39 completed weeks (HR = 0.93 [0.91, 0.95]) disappeared after adjustment (HR = 1.01 [0.99, 1.03]) (Table 1). The reduced risk of stillbirth for GA ≥40 completed weeks was mitigated after adjustment but remained significant (unadjusted HR = 0.83 [0.78, 0.88]; adjusted HR = 0.92 [0.86, 0.99]).
|All GA||GA ≤39 weeks||GA ≥40 weeks|
|2002–2004 versus 1992–94||1.00 (0.98–1.02)||1.01 (0.99–1.03)||0.92 (0.86–0.99)|
|<20 years||1.10 (1.06–1.14)||1.13 (1.09–1.18)||0.76 (0.66–0.88)|
|≥35 years||2.74 (2.68–2.81)||2.73 (2.66–2.80)||2.95 (2.70–3.21)|
|<12 years versus ≥12 years||1.19 (1.15–1.22)||1.18 (1.14–1.22)||1.28 (1.15–1.43)|
|Maternal smoking||1.61 (1.57–1.65)||1.64 (1.60–1.68)||1.31(1.20–1.44)|
|Medical risk factors|
|Diabetes||1.15 (1.09–1.21)||1.12 (1.06–1.18)||1.81 (1.50–2.19)|
|Chronic hypertension||2.60 (2.42–2.79)||2.59 (2.41–2.78)||2.96 (2.20–3.97)|
|Gestational hypertension||1.19 (1.13–1.25)||1.17 (1.12–1.24)||1.47 (1.21–1.78)|
The trend toward higher rates of labour induction and caesarean delivery from 1992–94 to 2002–2004 was observed in nearly all states. Among the 45 ecological units, only three states (Alaska, Hawaii and Wisconsin) had a decrease in induction rates (−5.5, −0.4 and −4.6%, respectively), and all had an increase in caesarean delivery; most states (except Connecticut, Maine, Missouri, Nevada, New Jersey, New Mexico and Virginia) had a reduction in stillbirth. As shown in Table 2, no association was observed between changes in stillbirth and induction rates. The ecologic correlation between changes in stillbirth at GA ≤39 completed weeks and in induction was nearly zero (r = −0.02 [−0.31, 0.28]), while a nonsignificant negative correlation (r = −0.17 [−0.44, 0.13]) was observed between changes in stillbirth at GA ≥40 weeks and induction. A nonsignificant positive ecologic correlation (r = 0.23 [−0.07, 0.49]) was observed between changes in stillbirth and caesarean delivery (Table 2). The ecologic correlations between changes in stillbirth and changes in caesarean were similar for stillbirth at GA ≤39 and ≥40 weeks.
|Change in stillbirth rates|
|All GAs||GA ≤39 weeks||GA ≥40 weeks|
|Change in induction rates|
|All pregnanciesa||−0.07 (−0.36, 0.23)||−0.02 (−0.31, 0.28)||−0.17 (−0.43, 0.14)|
|Low-risk pregnanciesb||0.00 (−0.33, 0.33)||0.02 (−0.32, 0.35)||−0.06 (−0.39, 0.28)|
|Not-low-risk pregnanciesb||0.02 (−0.32, 0.35)||0.04 (−0.30, 0.37)||−0.13 (−0.44, 0.21)|
|Change in caesarean rates|
|All pregnanciesa||0.23 (−0.07, 0.49)||0.21 (−0.09, 0.47)||0.19 (−0.11, 0.46)|
|Low-risk pregnanciesb||0.04 (−0.30, 0.37)||0.02 (−0.32, 0.35)||0.08 (−0.26, 0.40)|
|Not-low-risk pregnanciesb||0.28 (−0.06, 0.56)||0.29 (−0.05, 0.57)||−0.03 (−0.36, 0.31)|
Among low-risk non-Hispanic White births, stillbirth rates were lower than the overall rates in both periods: 4.0 per 1000 in 1992–94 and 2.5 per 1000 in 2002–2004 (adjusted HR = 0.64 [0.62, 0.66] for the later versus earlier period). Rates of caesarean delivery were also lower in both periods (19.3 and 23.4%, respectively) but rates of labour induction were similar to the overall rates. Ecologic analysis showed no significant association between changes in stillbirth and labour induction or caesarean delivery among low-risk women; all weighted ecologic correlation coefficients were close to zero (Table 2). In contrast to low-risk pregnancies, stillbirth rates among nonlow-risk women were much higher and did not diminish over the 10-year study period: 7.2 per 1000 in 1992–94 and 8.1 per 1000 in 2002–2004 (adjusted HR = 1.19 [1.16, 1.22] for the later versus earlier period). Rates of labour induction and caesarean delivery were also higher in both periods (18.5 and 28.8%; 21.8 and 28.4%). Weighted ecologic correlations between changes in stillbirth and labour induction or caesarean delivery were similar to those among all pregnancies (Table 2).
Stillbirth rates were much higher among non-Hispanic Blacks than among non-Hispanic Whites: 11.5 per 1000 in 1992–94 and 11.0 per 1000 in 2002–2004, respectively. The relative reduction in rates across the two periods was only 4.3% (adjusted HR = 1.01 [0.98, 1.04]). The adjusted HRs were similar for stillbirth at GA ≤39 and ≥40 completed weeks. The ecologic correlation between changes in stillbirth and induction rates among non-Hispanic Blacks was also similar to that among non-Hispanic Whites. However, a strong and significant positive correlation (r = 0.56 [0.30, 0.74]) was observed between changes in rates of stillbirth and caesarean delivery among non-Hispanic Blacks. This highly significant correlation (r = 0.56, [0.30, 0.74]) was observed for stillbirths at GA ≤39 completed weeks but was much weaker and statistically nonsignificant for stillbirths ≥40 completed weeks (r = 0.12 [−0.20, 0.42]).
We observed a substantial reduction in the US stillbirth rate among non-Hispanic Whites from 1992–94 to 2002–2004. Our Cox regression model, based on fetuses at risk at each week of gestation, suggests that the reduction at ≤39 weeks can be explained by changes in maternal risk factors. The reduced rate at ≥40 weeks, however, was only partly attenuated by adjustment for those same risk factors. In our state-level ecologic analysis, we observed no significant association between reduction in stillbirth (in 2002–2004 versus 1992–94) and increasing rates of labour induction or caesarean delivery.
One of the strengths of our study is our use of the fetuses-at-risk approach. At the beginning of each week of gestation, all live fetuses are at risk of subsequent stillbirth and should therefore be used as the denominator in calculating stillbirth risk. Another strength is our state-level ecologic analysis to examine the association between changes in rates of iatrogenic delivery and rates of stillbirth. Ecologic analysis should reduce confounding by medical indication,[10, 29-31] which cannot be adequately measured and adjusted for in individual-level analysis. We used pair-wise analysis (difference in rates within states between the two time periods) to take into account potential dependence of outcomes, maternal risk factors, and practice style within states at two time periods and to reduce bias due to systematic reporting errors by states. As previously reported, the increase in stillbirths at 20–22 weeks of gestation may be attributable to reporting differences. Our pair-wise ecological design should also account for this reporting difference among the states. Moreover, excluding stillbirths with GA<24 weeks in our pair-wise ecological analysis yielded nearly identical results and conclusions (results available on request).
A limitation of our study is our use of a nationwide vital statistics database, in which coding errors are known to occur.[36, 37] In particular, missing data are more frequent in data files for fetal deaths than in those for live births, and for 1992–94 than for 2002–2004. In 1992–94, six states did not report gestational age, Hispanic origin or plurality for a large proportion of fetal deaths. Most of those states did report these data in 2002–2004. Excluding those states (rather than including individuals with missing data) from our principal analysis, however, should reduce systematic state-level reporting errors and thus reduce bias due to missingness. On the other hand, ecologic analysis reduced statistical power relative to the individual-based analysis and results in wide confidence intervals, and exclusion of states further reduced our statistical power.
In the multivariate Cox regression model, ten additional states were excluded owing to missing data on maternal demographic data or risk factors among a large proportion of fetal deaths. Nevertheless, the stillbirth rates and maternal characteristics in this reduced sample were similar to those of the overall study sample. Moreover, missingness was higher among stillbirths than live births, and higher in the earlier versus the latter period. For example, maternal medical risk factors were missing in nearly 5% of stillbirths but only 0.6% of live births. Surprisingly, more data on maternal smoking and maternal education were missing in 2002–2004 than in 1992–94. Therefore, bias due to differential missingness in covariates in our Cox regression models cannot be excluded.
Finally, lack of state-level information on births since 2005 prevented us from including more recent data in our analysis; we are thus unable to assess whether our findings would persist in recent years. Nor can generalisability to other countries and health systems be assumed. Published induction rates in the US are among the highest known internationally.
Based on our findings, most of the reduction in the US stillbirth rate over the two study periods 10 years apart appears attributable to improvement in maternal risk factors. A major argument offered for increasing obstetric intervention is that iatrogenic delivery reduces the number of stillbirths.[14-16] Randomised controlled trials have shown that routine labour induction (versus expectant management) at ≥41 weeks reduces the risks of stillbirth and perinatal mortality. However, any benefit of reducing stillbirth at <39 weeks remains unsupported. Consistent with the results from randomised trials, our ecologic (state-level) analysis showed a modest, nonsignificant negative correlation (−0.17) between changes in stillbirth and induction rates ≥40 weeks. In an earlier study, Yuan et al. showed a reduction in stillbirths at 40 completed weeks or beyond among non-Hispanic Whites in 1997 versus 1991 and a significant ecologic association with increasing use of labour induction over that period. The main methodological difference between the two studies was that the recent study used the repeated measure design in the ecological analysis to take into account the potential dependence of outcomes, maternal risk factors, and practice style within states at two time periods and thus to reduce bias due to systematic reporting errors by states. In contrast, the earlier analysis considered that two time points within a state are independent. As a result, our study produced larger but more robust standard errors. Another methodological difference between the two studies was the basis for gestational age estimate: the previous one was based entirely on LMP, whereas the current one was based on LMP corrected by the clinical estimate. This difference in gestational age estimation leads to different estimates of the postterm stillbirth rate and thus may yield different results and conclusions.
Our secondary ecologic analysis stratified by low- versus nonlow-risk pregnancies showed a significant reduction of 61% in stillbirth among low-risk pregnancies in the latter versus earlier periods, but an increase of 13% among women not at low risk. Ecologic analysis among low-risk pregnancies, however, showed no associations between the reduction in stillbirth and increasing use of induction or caesarean (Table 2). The large reduction in stillbirths among low-risk pregnancies may be attributable partly to improved data reporting on maternal medical risk factors over time. On the other hand, the increase in stillbirth among nonlow-risk pregnancies may well be due to both improved data reporting on maternal medical risk factors and an increase in prevalence of these risk factors (diabetes, hypertension, etc.) over time.
We have no good explanation for the observed positive ecologic correlations (statistically significant for non-Hispanic Blacks) between changes in rates of stillbirth and caesarean delivery. No information on the timing (prelabour versus intrapartum) of caesarean deliveries is available from US birth certificates. As most caesareans are performed intrapartum for such indications as slow progress of labour, perceived cephalo-pelvic disproportion or fetal distress, the positive correlation may reflect an increase in the occurrence of these (unrecorded) indications or, more likely, a greater reduction in antepartum than in intrapartum stillbirths, particularly among non-Hispanic Blacks.
Recent increases in late preterm (34–46 weeks) and early term (37–38 weeks) singleton births in the USA have been largely attributed to contemporaneous increases in iatrogenic delivery at late preterm and early term.[6, 7, 10] Iatrogenic delivery at preterm and early term gestation should be based on specific clinical indications, yet recent studies have consistently shown that in a substantial proportion, those indications are not substantiated.[41-43] It seems likely that the rise in late preterm and early term intervention is due to a lowering of the threshold for intervention, rather than to a higher incidence or severity of obstetric complications.
It has long been debated whether today's obstetric intervention rates are too high.[17, 44] The short- and long-term adverse consequences of preterm delivery have been well documented.[45-48] Even early term delivery has been shown to be associated with higher risks of infant mortality and neonatal morbidity. In the absence of clear evidence that these iatrogenic early deliveries reduce the risk of stillbirth, these consequences should be taken into account when deciding to intervene in the absence of clear clinical indications. Future research should include large randomised trials of iatrogenic delivery versus expectant management in the late preterm period among pregnancies with non-urgent clinical indications.
Neither author has any conflict of interest concerning the topic or contents of this article.
The authors jointly designed the study. XZ carried out the analysis and wrote the first draft of the manuscript. MSK obtained the funding for the study, supervised the interpretation of the analysis, and helped XZ in revising the manuscript.
Our study was funded by a grant from the Canadian Institutes of Health Research. The funding source did not contribute in any way to the design, the collection, analysis and interpretation of data, the writing of the manuscript or the decision to submit the manuscript for publication. Drs Kramer and Zhang are members of the Research Institute of the McGill University Health Centre, which is supported in part by the Fonds de recherche du Québec—Santé.