Please cite this paper as: Zhang X, Kramer M. The rise in singleton preterm births in the USA: the impact of labour induction. BJOG 2012;119:1309–1315.
Objective To assess the extent to which increased rates of labour induction and caesarean section have contributed to the recent rise in preterm birth.
Design National birth cohort study.
Population and sample Singleton live births, with primary analysis based on non-Hispanic white women.
Methods Ecological study based on the 50 states and the District of Columbia during two time periods 10 years apart: 1992–94 and 2002–04.
Main outcome measure Preterm birth (live birth <37 completed weeks of gestation), based on an algorithm combining menstrual and clinical estimates of gestational age.
Results The state-level ecological analysis among non-Hispanic white women showed that the change in preterm birth rate from 1992–94 to 2002–04 was significantly associated with the change in rate of labour induction (r = 0.50, 95% CI 0.26–0.68), but not with the change in rate of caesarean delivery (r = −0.06, 95% CI −0.33 to 0.22). Weaker but otherwise similar associations with labour induction were observed in Hispanic women and in non-Hispanic black women.
Conclusions Increasing use of labour induction is probably an important cause of the observed increased rate in preterm birth.
Preterm birth is defined as the birth of an infant at <37 weeks (i.e. <259 days) of gestation.1 Preterm birth is the leading cause of infant mortality and is associated with substantial risks of serious neonatal and life-long morbidity.2–7 In the last two decades, the preterm birth rate has increased substantially in the USA,8–10 other developed countries,7,11,12 and some developing countries.7,13
Meanwhile, the rates of obstetric interventions have also been rising. In the USA, the labour induction rate doubled between 1992 and 2006, from11.4% to 22.5%, and the caesarean delivery rate increased by nearly 50%, from 22.3% in 1992 to 31.1% in 2006.14,15 As reported by Ananth et al.,16 the increase in preterm birth observed in the USA during the 1990s appeared to be entirely driven by an increase in medically indicated preterm birth. But that study did not separate labour induction from caesarean deliveries in its classification of births as ‘medically indicated’, nor was it able to distinguish a temporal increase in the proportion of women with clinical indications for intervention from a lowered threshold for intervention in the presence of those indications.
In a recent study,15 we used an ecological (state-based) design to minimise the potential for the confounding by clinical indication that inevitably occurs when analysing individual-level associations between obstetric intervention and birth outcome.17–19 That study demonstrated a strong association between US state-level changes in rates of labour induction and a decline in both gestational age and birthweight among term and post-term singleton births in non-Hispanic white women. In the current study, we extend that approach to examine associations with rates of preterm birth.
Our study is based on data from birth data files, compiled by the US National Vital Statistics System of the National Center for Health Statistics (NCHS), which are available online from the Centers for Disease Control and Prevention (CDC) (http://www.cdc.gov/nchs/data_access/Vitalstatsonline.htm). These files provide demographic and health data for births occurring during the calendar year based on information abstracted from birth certificates filed in vital statistics offices of each state and the District of Columbia. Because the files are public and anonymised, no approval is required by our Institutional Review Board.
We used singleton live birth cohorts for the years 1992–94 and 2002–04. The US birth certificate was revised in 1989 to include, among other data items, the clinical estimate of gestational age and the use of induction.14 A substantial number of states did not report these data items in 1989–1991, however.14,15 We therefore began our analysis with the year 1992. The most recently available birth data file is for 2008. Since 2005, however, geographic information (state of birth) on newborns is no longer available,14 and we were therefore unable to include the most recent data in our ecological analysis. The study is therefore based on a comparison of 1992–94 with 2002–04.
Gestational age estimation
In the USA, gestational age is usually calculated from the first day of the mother’s last menstrual period (LMP). A clinical estimate of gestation (CE) has also been available since the 1989 revision, except for the state of California. It has been shown that gestational age derived from the LMP estimate is prone to error20,21 and that the preterm birth rate is overestimated by use of the LMP estimate.12 Recent evidence suggests that the clinical estimate provides rates of preterm birth, post-term birth, and gestational-age-specific rates and relative risks of adverse pregnancy outcomes that are more consistent with those reported in other countries.22 Although the CE is based on the managing clinician’s best estimate, including menstrual history, physical findings, laboratory values and (if available) sonography,8,9 no instructions were provided before the 2003 revision of the US birth certificate for specifying the basis of the CE.9 The CE on birth registrations may therefore have changed over the study period, especially with the increasing use of routine dating by early ultrasound.9
Several methods for editing the LMP-based gestational age have been proposed to reduce the proportion of records misclassified, including Platt et al.,23 Zhang and Bowes,24 Alexander et al.25 and Qin et al.26 In the ‘LMP/CE’ method, the CE is substituted for the LMP-based estimate when the two estimates differ by more than 2 weeks. The latter method has been shown to eliminate the second mode of the bimodal birthweight distribution among preterm births observed with the LMP estimate alone, and therefore to reduce misclassification of gestational age.26 We therefore used the LMP/CE method in this study. California did not report the CE before 2005 so the LMP was used for that state.
Our principal analysis was based on singleton live births in non-Hispanic white women. Race-specific analysis was intended to control for potential confounding by differences in induction and caesarean rates by race and the changing racial composition over time.27,28 To assess the generalisability of our findings, however, we repeated the analysis among non-Hispanic black women and among Hispanic women. We restricted all analyses to live births to avoid reverse causality due to the use of induction to deliver dead fetuses.
The association between preterm birth and labour induction or caesarean delivery is likely to be confounded by the medical indication for labour induction or caesarean delivery. Although some maternal risk factors are reported on the US birth certificate, specific indications for induction or caesarean are not and so cannot be controlled for adequately at the individual level. To reduce confounding by clinical indication, we carried out an ecological analysis15–18 based on 51 ecological units: 50 states plus the District of Columbia (D.C.).
The rate of induction in US vital statistics may overestimate the true rate.29 However, differences in induction rates between two time points are less likely to be affected by any systematic under-reporting or over-reporting within states. We therefore based our ecological analysis on changes between two time periods. The variation in rates of induction among the ecological units is largely driven by local practice standards. Because the rates at two time points are not likely to be independent, we used differences in induction rates between two time points to account for this nonindependence.
We calculated rates of preterm birth, labour induction and caesarean delivery for each of the 51 ecological units in the periods 1992–94 and 2002–04, respectively, and then the changes in those rates between the two time periods, i.e. the differences between 1992–94 and 2002–04. Using these state-level differences for each of the 51 units, we assessed the association between the change in rate of preterm birth and the changes in rates of labour induction and caesarean delivery. We calculated partial correlation coefficients between the change in rate of preterm birth and the change in rates of labour induction and caesarean delivery, weighted for the total number of births in each state and adjusted for changes in maternal age and maternal education. Ecological multiple linear regression analysis was then used to estimate the independent effect of change in labour induction or caesarean delivery on the change in rate of preterm birth after adjustment for state-level changes in maternal age and education. Data on maternal smoking in 1992–94 were not available in three states: California, Illinois and South Carolina. We controlled further for the state-level change in maternal smoking in the regression analysis with these three states excluded.
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. Restriction to low-risk delivery has been used in previous studies to control for confounding by clinical indication, both in the USA and Canada,30–33 although definitions have varied slightly depending on available information. We also conducted separate sensitivity analyses based on the LMP (only) and the CE (only) estimates of gestation (excluding California when using the CE).
As shown in Figure 1, the preterm birth rate based on the LMP/CE method among singleton births in non-Hispanic white women increased from 6.4% in 1992 to 8.5% in 2004, an increase of more than 30% over the 13-year study period. The preterm birth rate based on the LMP estimate was at least 1% higher every year than that based on the LMP/CE method, whereas the rate based on the CE was about 0.5% lower than the LMP/CE method in each year (Figure 1). Changes between 1992–94 and 2002–04 in maternal age, education, smoking during pregnancy and medical risk factors are shown in Table 1.
Table 1. Changes (%) and relative risk (RR) with 95% CI in preterm birth, labour induction, caesarean delivery and selected maternal characteristics for singleton births in US non-Hispanic white women comparing 2002–04 with 1992–94
1992–94 (n = 7 238 545)
2002–04 (n = 6 653 579)
RR (95% CI)
*Data not available in California, Illinois and South Carolina in 1992–94.
Maternal medical risk factors
Rates of labour induction among non-Hispanic white women with singleton births doubled during the study period, rising from 13.7% to 26.0%, whereas the caesarean delivery rate rose by 25%, from 22.0% in 1992 to 27.6% in 2004, with a decline from 1992 to 1997. The trend toward higher rates of preterm birth, labour induction and caesarean delivery from 1992–94 to 2002–04 was observed in nearly all states (Figures 2 and 3). Among the 51 ecological units, only the District of Columbia had a decrease in preterm birth rate (−1.8%), and three states (Alaska, Hawaii and Wisconsin) had a decrease in induction rates (−5.5%, −0.4% and −4.6%, respectively).
A strong and statistically significant positive correlation was observed in the state-level ecological analysis between the changes in the rates of preterm birth and labour induction (Figure 2). With adjustment for changes in maternal age and education in each state weighted by its total number of births, the partial correlation and its 95% CI were 0.50 (0.26–0.68) (Table 2). Based on multiple linear regression analysis with adjustment for changes in maternal age and education (Table 2), a 1% increase in the rate of labour induction was associated with an increase of 0.06% (95% CI 0.03–0.09%) in the rate of preterm birth. Excluding the three states without smoking data from the ecological analysis and further adjusting for change in maternal smoking in the model did not alter the results (data not shown). In contrast, no significant association was observed between changes in the rates of preterm birth and caesarean delivery (Figure 3); the weighted partial correlation was −0.06 (−0.33 to 0.22).
Table 2. Ecological correlation (95% CI) and multiple regression coefficient (95% CI) showing the association between the change in preterm birth rates and the change in induction rates by race, USA, from 1992–94 to 2002–04
Model adjusted R2
*Weighted for the total number of births of each state and adjusted for changes in maternal age and maternal education.
**Maternal age was not included in the model.
Non-Hispanic white women
0.50 (0.26 to 0.68)
0.06 (0.03 to 0.09)
0.50 (0.26 to 0.68)
0.06 (0.03 to 0.10)
Non-Hispanic black women
0.17 (−0.11 to 0.43)
0.05 (−0.03 to 0.13)
0.21 (−0.07 to 0.48)
0.07 (−0.02 to 0.18)
0.30 (0.03 to 0.53)
0.04 (0.01 to 0.08)
0.28 (0.01 to 0.52)
0.06 (0.00 to 0.12)
Among low-risk pregnancies, temporal trends in preterm birth, labour induction and caesarean delivery were similar, but with slightly lower rates of all three events in every year. The ecological partial correlations were also similar: 0.50 (0.26–0.68) for the changes in rates of preterm birth and labour induction, and 0.01 (−0.27 to 0.28) for the changes in rates of preterm birth and caesarean delivery. Sensitivity analyses based on the LMP and the CE estimates, respectively, yielded similar results (data not shown).
Temporal trends in induction and preterm birth among non-Hispanic black women and Hispanic women were similar to, but weaker than, those observed in non-Hispanic white women. The preterm birth rate among singleton live births to Hispanic women increased nearly 10% from 8.4% in 1992 to 9.2% in 2004. A statistically significant partial ecological correlation of 0.30 (0.03–0.53) was observed in Hispanic women between changes in the rates of preterm birth and labour induction from 1992–94 to 2002–04 (Table 2), but not with caesarean delivery (partial correlation = −0.17; 95% CI −0.43 to 0.11). Although the rate of preterm birth was much higher among non-Hispanic black women than among white women, the temporal trend was different, with rates fluctuating between 13.1% and 13.8% from 1992 to 2004. The ecological analysis in non-Hispanic black women showed a positive but not significant association between changes in the rates of preterm birth and labour induction (partial correlation = 0.17; 95% CI −0.11 to 0.43) (Table 2) but not with caesarean delivery (0.02; 95% CI −0.26 to 0.29). Results from low-risk analyses among non-Hispanic black women and Hispanic women, respectively, were also similar (Table 2).
Discussion and conclusion
Preterm birth is a major health concern, with serious social and economic consequences.3,34,35 The causes of the rise in preterm birth are complex. Preterm birth has been associated with maternal age, smoking during pregnancy, socio-economic status and racial/ethnic background.3,34 As demonstrated in our and other10,36 studies, changes in maternal age, smoking, socio-economic status and race distribution do not explain the large recent rise in preterm birth. Previous studies have attributed the rise to increasing obstetric intervention, including both labour induction and caesarean delivery.10,36 However, observational analyses conducted at the individual level are likely to be confounded by the clinical indication for labour induction or caesarean delivery. Ananth et al.16 reported that the increase in preterm birth in the 1990s among US non-Hispanic white women was entirely the result of an increase in medically indicated (defined as accompanied by induction or caesarean delivery) preterm birth, but that study was unable to distinguish a temporal increase in the proportion of women with clinical indications for intervention from a lowered threshold for intervention in the presence of those indications.
Using a state-level ecological analysis, we show that the rise in preterm birth in the USA from 1992–94 to 2002–04 was associated with the increasing rate of labour induction, but not with the increasing rate of caesarean delivery. Most of the rise in caesarean delivery has probably resulted from caesarean sections during, rather than before, labour. Caesarean deliveries performed after the onset of labour make a difference of only a few hours in gestational age and should therefore not have a large impact on gestational age and the preterm birth rate. The design of our ecological analysis (i.e. based on changes over time rather than during a single year or period) has a further advantage. Instead of using cross-sectional state-level proportions,16 we based our analysis on the changes between two time periods 10 years apart. Our approach (paired analysis) therefore takes into account the tendency of states to have (or report) consistently high or low rates of labour induction or caesarean delivery.
The relatively smaller increase in preterm birth among Hispanic women and the somewhat different temporal trend among non-Hispanic black women merit comment. We observed a similar but weaker association between the changes in the rates of labour induction and preterm birth among Hispanic women and non-Hispanic black women than among non-Hispanic white women. As has been observed previously, these racial differences may reflect the racial differences in access to prenatal care, and therefore in obstetric intervention.16 As reported by Martin,9 LMP and clinical estimates of gestational age were unknown in 16.5% and 3.0%, respectively, of non-Hispanic black women in 1990, and in 6.9% and 0.4%, respectively, in 2002. Despite these declines, non-Hispanic black mothers continued to have the highest level of incomplete reporting of gestational age among the three racial/ethnic groups.9 As a result, LMP-based and CE-based trends in preterm birth differed slightly for non-Hispanic black women, and the true trend is difficult to gauge.9,22 Misclassification in gestational age can bias (toward the null) the association between preterm birth and labour induction.
The strength of our study is its use of ecological analysis, which should reduce confounding by clinical indication.15–19 The latter cannot be adjusted for in individual-level analyses without valid and reproducible data on medical indication. When studying the association between obstetric intervention and preterm birth, ecological analysis is less likely to be biased than individual-level analysis, because the large variations in labour induction and caesarean delivery rates observed across states are likely to be driven by regional/local practice style, rather than by state-level differences in clinical indication. Our analysis being restricted to low-risk women further reduces confounding by the indication for obstetric intervention and therefore strengthens our findings. Moreover, both our multivariate ecological regression analysis and analysis restricted to low-risk women demonstrate again that the observed trends in preterm birth cannot simply be attributed to increases in maternal risk factors. Finally, use of the LMP/CE editing method should improve the accuracy of the estimate of the preterm birth rate and so reduce bias (towards the null) because of nondifferential measurement error.26
A limitation of our study is our use of a nationwide vital statistics database, in which coding errors are known to occur.37,38 In particular, inaccurate estimation of gestational age and misreporting of labour induction have been widely reported and discussed.9,27,29,39,40 It is unlikely, however, that such misreporting of labour induction would be different with respect to gestational age, and so the associations we observed are likely to be conservative (biased toward the null). In addition, the use of changes in rates of labour induction between two time points reduces the influence of systematic under-reporting or over-reporting within states. Basing our analysis on changes in rates over time should also reduce systematic differences among states in true underlying state-wide differences in maternal risks. Another limitation of our study is the lack of information on state of births since 2005. We were therefore unable to include more recent data in our analysis.
All ecological analyses are subject to the so-called ecological fallacy. As noted above, however, variations in labour induction and caesarean delivery rates observed across states are probably driven by local practice standards, rather than by state-level differences in clinical indication. While practice style is difficult to detect at the individual level, its impact can be measured ecologically. Differences in practice standards create a quasi-experiment across states in use of labour induction and caesarean delivery, and therefore permit an assessment of the potential effect of increasing labour induction or caesarean delivery on preterm birth.16,17 Finally, with only 51 ecological units, we do not have sufficient statistical power to include more available variables in our ecological regression analysis. Residual confounding cannot be excluded. However, our low-risk analysis should reduce confounding to a minimum.
Stillbirth rates decreased from 1990 to 2005 (from 7.5 to 6.2 per 1000).41 Although some evidence suggests that increased labour induction is associated with decreased risk of stillbirth at 40–43 weeks of gestation,31,42,43 we are aware of no evidence that induction reduces fetal deaths among preterm births. More and more labour inductions have been performed at 34–36 weeks of gestation among women at high risk for adverse pregnancy outcomes.44 It is likely that the rise in preterm labour induction is the result of a lowering of the threshold for intervention, and not of a higher incidence or severity of obstetric complications.
The decision to induce a preterm delivery must balance the benefits of early delivery against the risks associated with preterm birth.45 Very few obstetricians would knowingly induce labour before 37 weeks of gestation in the absence of clinical indications such as pre-eclampsia, poor fetal growth, nonreassuring fetal heart tracing, or poorly controlled diabetes. But weighing the risks of stillbirth or serious maternal morbidity with expectant management (watchful waiting) versus those of immaturity-related neonatal morbidity and mortality is difficult and leads to practices that vary considerably by place and over time. An improved evidence base is required to help clinicians to weigh the risks and benefits of alternative management strategies, particularly at late preterm gestational ages.46
Disclosure of interests
Neither author has any conflict of interest concerning the topic or contents of this article.
Contribution to authorship
The authors jointly designed the study. XZ carried out the analysis and wrote the first draft of the manuscript. MK obtained the funding for the study, supervised the interpretation of the analysis, and helped XZ in revising the manuscript.
Details of ethics approval
This study was supported by a grant from the Canadian Institutes of Health Research. SK and XZ are members of the Research Institute of the McGill University Health Centre, which is supported in part by the Fonds de la recherche en santé du Québec.