The probability of preterm or early term second live births in the southern U.S. State of Georgia, 2011–2020

Abstract

Background

The extent to which past preterm or early term births are risk factors for recurrent shortened gestational age at delivery in the US is unclear. The underlying causes of shortened gestational age and the role of maternal health can vary over time and across the 50 US states.

Objectives

To estimate differences in the probability of a second live birth being full term (> 38 weeks’ gestation), early term (37–38 weeks’) or preterm (< 37 weeks’) conditional on the gestational age of the first live birth.

Study design

We used linked birth and hospital discharge records from the state of Georgia to construct a retrospective cohort of individuals whose first and second births resulted in a singleton live birth (2011–2020). Multinomial models were used to estimate the difference in the probability of the second live birth gestational age category (< 32, 32–36, 37–38, ≥ 39 weeks) conditional on the first birth gestational age category. Baseline models were only adjusted for year fixed effects. Subsequent models were adjusted for birthing individual characteristics: race, age at first birth, ethnicity, country of birth, and education at first birth; second birth health, behavioral and socioeconomic risk factors; and, for interpregnancy birth interval and change in paternal characteristics between first and second births. All analyses were completed in Stata 17.

Results

Individuals whose first live birth was preterm or early term were significantly less likely to have a second full term live birth, compared to individuals who had a first full term live birth. The probability of a full term second live birth following a first preterm live birth at < 32 weeks or 32–36 weeks decreased by 27.7% points (pp) (95% CI -30.0, -25.2) and 22.1 pp (95% CI-23.3, -21.0) respectively. Similarly, the probability of a full term second birth following a first early term birth decreased by 14.9 pp (95% CI − 15.7, -14.2). Individuals who had early term or preterm first births were more likely to have early term and preterm second births, with higher risk for recurrent preterm birth among birthing individuals with earlier preterm first births. For example, following a first birth that is early term, the probability of a second birth at 32–36 weeks increased by 4.7 pp (95% CI 4.2, 5.1), whereas following a first birth at < 32 weeks the probability of a second birth at 32–36 weeks increased by 13.6 pp (95% CI 11.5, 15.6).

Conclusions

A first live birth that is early term or preterm predicts a higher probability of a second early term or preterm birth. Health information systems that flag these early birth outcomes on the problem list may help clinicians address known risk factors interconceptionally and prenatally in a subsequent pregnancy.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12884-025-07763-1.

Introduction

Preterm birth (< 37 weeks of gestation) is among the leading causes of neonatal death for US infants [1, 2]. While the worst infant outcomes are associated with the earliest preterm births [3], there is growing evidence that early term births (between 37 and 38 weeks of gestation) also increase infant mortality and morbidity relative to full term births [4–6]. Researchers report that early term births exhibit pathologic placental conditions often associated with preterm birth and argue that imposing an upper limit of 36 + 1/6 weeks’ gestation to define preterm birth lacks biological justification [1, 7, 8]. In a 2019 committee opinion, the American College for Obstetrics and Gynecology (ACOG) referred to the early term period as a transitional phase and advised that non-medically indicated early term deliveries should be avoided to optimize the perinatal outcome period [9].

Historically, risk of preterm birth has been linked to past preterm births; accordingly, a past preterm birth is a risk factor for preterm birth in future pregnancies (see Phillips, Velji, Hanly and Metcalfe [10] for a review of the literature) [10–12]. Since the context in which preterm births occur evolves over time (e.g., due to individuals having children later in life or the increased use of fetal surgical interventions), ongoing surveillance and analysis to evaluate preterm births with timely population-level data are necessary to target at-risk populations. These efforts should extend to early-term births which have received little attention from researchers. In addition to targeting individuals at risk for preterm or early term births, analysis of longitudinal patterns in preterm birth outcomes inform growing research aimed at understanding the etiologic complexities of deliveries at a shortened gestational age [8].

Preterm birth rates, including risk of reoccurrence, can also vary across heterogenous populations with varying access to health care [8]. There is limited longitudinal birth outcomes research for US populations in part because the US does not systematically link administrative or medical records which is necessary for longitudinal analysis [10]. Among the few studies focused on longitudinal preterm rates in the US [13, 14], the data used are older even among the most recent studies (e.g., 1999–2012) [12]. Data barriers are similarly reflected in the newer and smaller literature on longitudinal birth outcomes that focuses on early term births [15–17].

In this study, our goal was to inform the use of early term or preterm first birth as risk factors to target populations for interventions to prevent recurrent shortened gestational age at delivery among second pregnancies in the US. We used population data from Georgia, a relatively large and diverse southern state with the tenth highest preterm birth rate in the US in 2021 [18, 19]. We estimated unadjusted and adjusted differences in the probability of full term, early term or preterm second live births conditional on first live birth gestational age category between 2011 and 2020, updating the existing literature with much more recent data.

Methods

Study design, setting, and data sources

This is a retrospective cohort study of live births to Georgia-resident individuals delivering in Georgia 2011–2020 using linked birth and hospital discharge record data made available by the Georgia Department of Public Health (DPH).

Study population/participants and sample size

The unit of observation is at the level of the individual giving birth (the mother). Only individuals whose first pregnancy resulted in a live first birth and whose second birth event also resulted in a live birth in Georgia were included (Appendix 1). Sample selection, missingness and case wise deletion by covariate are reported in Fig. 1. First and second live births were identified by variables present in the vital records that indicate live vs. stillbirth status and birth order. After identifying individuals with two or more sequential live births with gestational ages at delivery between 20 and 44 weeks, we additionally excluded individuals with missing values for covariates and births before 2011, resulting in a total of 99,080 individuals. This study was reviewed by Emory University’s Institutional Review Board. Research consent was waived per 45 CFR Sect. 46 (the Common Rule) [20].

Fig. 1.

Sample construction. Notes: Births to individuals missing one or more covariates for one more birth by covariate: Hispanic (n = 74, 0.04%), education (n = 2,467, 1.21%), born in another country (n = 791, 0.39%). Diagram starts with 1999–2020 births for all birthing individuals who are GA residents with observable, sequential live births resulting in GA hospital discharge, and no missing gestational age data. See Appendix 1 for additional information

Variables of interest

Our primary outcome of interest was gestational age at delivery (completed weeks) as reported on the birth certificate for an individual’s second live birth, which was categorized as < 32, 32–36, 37–38, or ≥ 39 weeks. The primary exposures were first live birth gestational age categories (also as < 32, 32–36, 37–38, or > 39 weeks (reference)).

Statistical analysis

We used multinomial logit models to evaluate statistical differences in the probability of second birth gestational age categories conditional on the gestational age category of an individual’s first live birth. Our baseline model is unadjusted. Subsequently, we explored whether and how the observed baseline association reduces in magnitude with the iterative addition of other risk factors observable to clinicians and that were used in previous US studies [15, 21–23]. Importantly, because unobserved differences in individual or birth specific risk for shortened gestational age at delivery, we cannot isolate the specific causal effect of first birth gestational age on second birth gestational age (e.g., anxiety of recurrent preterm births can itself increase risk of preterm birth outcomes [24–26]). For simplicity, we grouped covariates as birth invariant and birth specific, which were further categorized by first or second birth. The empirical framework guiding our analysis is shown in Fig. 2. The unadjusted empirical association between first and second birth gestational age outcomes will reflect the net aggregate effect of birth invariant and birth specific factors. We refer readers to published conceptual frameworks that offer theoretical hypothesis regarding preterm birth causal mechanisms [23].

Fig. 2.

Empirical framework linking gestational age at first and second live births

Birth-invariant individual-level factors are those that suggest some individuals are at higher risk for shortened gestational age at delivery across both their first and second births [21, 22]. Specifically, we included: race (White (reference), Black or African American, or other), ethnicity (Hispanic, not-Hispanic (reference)), country of birth (Mexico, other, US (reference)), age group at first birth (< 18, 18–34 (reference), ≥ 35), educational attainment at first birth (no high school, high school only (reference), some college or more). This information was available in birth record data. Other race aggregates individuals who identified as Asian, American Indian or Alaska Native, Native Hawaiian or Pacific Islander, and Multiracial because of small sample size. Age and education at first birth accounted for differences in when individuals have their children over the life course.

For birth-specific risk factors for shortened gestational age at delivery we used a set of indicators for health-related, behavioral and socioeconomic preterm risk factors [15, 21–23]. Indicators constructed using ICD 9 and 10 diagnostic or procedural codes in hospital discharge record data included: any hypertension, any diabetes, urinary tract infection (UTI) without indwelling catheter, other infection(s) (syphilis, gonorrhea, other venereal diseases, tuberculosis, malaria, rubella, other viral disease, other infectious and parasitic disease, unspecified infection or infestation), anemia, substance abuse, and mental disorder [27, 28]. Variable construction details are included in Appendix 2.

We also used information from both birth and hospital discharge records to construct a birth specific indicator for maternal prenatal smoking. Length of interpregnancy interval (months) at second birth was included either from the birth records or calculated as the difference between the date of birth for second birth less gestational age for the second birth and the date of birth for first birth. We created an indicator for the same paternity information (father’s year of birth, race, or ethnicity) reported in both first and second births. Because of changes in individuals’ education between births, we also constructed an indicator for educational attainment at second birth. Lastly, in the US, many low-income pregnant individuals access maternal care through insurance coverage from Medicaid, the means-tested public health insurance program. We constructed a Medicaid insurance coverage indicator which included traditional Medicaid, Children’s Health Insurance Program (CHIP) and Georgia’s Medicaid managed care programs.

For a partially adjusted model, we introduced only birth-invariant individual-level factors. For a fully adjusted model, we introduced all birth specific covariates described above only for second births to reflect the birthing individual’s circumstances during their second pregnancy with exception of an indicator for previous cesarian section. Holding these covariates constant, the empirical association between second birth gestational age and first birth gestational age would reflect a mix of potentially causal effects from shortened gestational age at delivery on subsequent birth outcomes and omitted variable bias from unobserved factors that affect both first and second birth gestational age at delivery. Accordingly, we cannot interpret our estimates as causal even adjusting for all covariates in the fully adjusted model.

For supplementary analysis, we additionally controlled for medically indicated births to better reflect the probability that individuals who experience spontaneous or elective first early term or preterm birth will experience a second preterm or recurrent early term birth [29, 30]. We used information from both birth certificates and hospital discharge records to identify early delivery subtype (spontaneous or indicated). Indicated preterm births were those with gestation < 37 weeks, and no premature rupture of membranes (PROM) or premature labor or tocolysis, but with documentation of medical induction of labor, artificial rupture of membranes, or a cesarean delivery without labor (Appendices 2 & 3).

To reduce longitudinal variation from time varying systemic changes that could be reflected in all births across years, we included calendar year-fixed effects for second births in all models. These indicators account for longitudinal differences in gestational age due to temporal variation in clinical practices, data quality and policy such as the Affordable Care Act (ACA) or the 2011 introduction of Georgia’s Medicaid Family Planning for Healthy Babies (P4HB) waiver program. We reported differences in the probability of second birth gestational age category conditional on for first birth gestational age category along with 95% confidence intervals for model results. The reference group for the multinomial logit is gestational age ≥ 39 weeks at second birth. All empirical analyses for this study were completed using Stata17.

Results

At their first birth, most birthing individuals in the sample were between 18 and 34 years of age (89.8%) and had some college education or more (59.4%) (Table 1). Most individuals were born in the USA (90.0%). The population was 63.5% White followed by 31.2% Black or African American individuals. Hispanic individuals make up 7.3% of the sample. One third (30.6%) of first births and 44.5% of second births were cesarean deliveries, the interpregnancy interval between first and second births was 34 months on average, and almost half of births (49.6% of first births and 48.6% of second births) were paid for by Georgia Medicaid.

Table 1.

Birth invariant individual-level and birth-specific characteristics of the retrospective cohort of individuals delivering first and second live births, Georgia, 2011–2020 (N = 99,080)

Note: The analytical sample is limited to individuals who have at least 2 singleton live births in Georgia between 2011–2020

Table 2 shows crude rates of second birth gestational age categories broken down by first birth gestational age category. Individuals whose first birth was preterm (< 37 weeks) or early term (37–38 weeks) are less likely to have a second full term birth (39–44 weeks) compared to those whose first birth was full term. Differences in crude rates are all statistically significant at p < 0.001% and are depicted in the first row of Fig. 3 (and Appendix 4). Relative to an individual with first full term birth, the probability of a second full term birth decreased by 27.7 pp (95% CI -30.0, -25.2) after a first birth before 32 weeks and by 22.1 pp (95% CI-23.3, -21.0) after a first birth at 32–36 weeks. For individuals who had an early term first birth, the probability of a full term second birth decreased by 14.9 pp (95% CI − 15.7, -14.2) compared to those whose first birth was full term.

Table 2.

Crude rates of second live births gestational age category by first live birth gestational age category, Georgia, 2011–2020 (N = 99,080)

Note: The analytical sample is limited to individuals who have at least 2 singleton live births in Georgia between 2011–2020. The percentages reflect the share of individuals who give birth at each gestational age category in their 2nd pregnancy by 1st pregnancy gestational age. Within each panel, the rows sum to 100%

Fig. 3.

Unadjusted, partially and fully adjusted differences in probability of second birth gestational age category conditional on first birth gestational age category. Notes: All estimates are statistically significant at 0.1%. The analytical sample is limited to individuals who have at least 2 singleton live births in Georgia between 2011–2020, and whose first pregnancy resulted in a live birth and subsequent birth event was for a live birth. All models include year fixed effects including unadjusted models (row 1). Partially Adjusted models include covariates for maternal age at first birth (ref: 18–34), educational attainment at first birth (ref: high school diploma), birthplace (ref: USA), ethnicity (ref: not Hispanic), mother race (ref: white). Fully adjusted models include all covariates from partially adjusted models and covariates for: same father between 1st and 2nd pregnancy, Medicaid coverage, interpregnancy interval, previous c-section, smoking/substance use status, maternal hypertension, diabetes and anemia, maternal infection (UTI and non-UTI), maternal mental health. All models include 2nd birth year fixed effect

Individuals whose first birth was preterm, or early term had a higher probability of experiencing a preterm or early term second birth, compared to those whose first birth was full-term. For those who had preterm first birth before 32 weeks, the probability of a second (recurrent) preterm birth before 32 weeks was increased by 8.4 pp (95% CI 7.0, 9.9), the probability of a second preterm birth of 32–36 weeks increased by 13.6 pp (95% CI 11.5, 15.6), and the probability of a second early term birth increased by 5.7 pp (95% CI 3.4, 8.0).

Later preterm first births were linked to relatively lower rates of earlier preterm second births. Among individuals whose first birth was 32–36 weeks, the probability of a second preterm birth before 32 weeks increased by 1.9 pp (95% CI 1.5, 2.2), the probability of a second (recurrent) preterm birth between 32 and 36 weeks increased by 10.7 pp (95% CI 9.8, 11.6), and the probability of a subsequent early term birth increased by 9.5 pp (95% CI 8.4, 10.6). Among individuals who experienced an early term first birth, the probability of a second preterm birth before 32 weeks increased by 0.7 pp (95% CI 0.5, 0.8), the probability of a second preterm birth between 32 and 36 weeks increased by 4.7 pp (95% CI 4.2, 5.1), and the probability of a second (recurrent) early term birth was an increase of 9.6 pp (95% CI 8.9, 10.3).

The second and third rows of Fig. 3 show how the baseline differences in probability of second birth gestational age category conditional on first birth gestational age category varies with addition of covariates for partial adjustment and complete adjustment. Estimates reduced in magnitude but remained largely intact and all were significant at p < 0.001. In our supplemental analysis, we additionally controlled for indicated preterm first births (Appendix 5). The estimates for the difference in probability of a second preterm or early term following a preterm first birth further decreased but remained significant.

Discussion

We find that birthing individuals who experienced first preterm (< 37 weeks’ gestation) or early term (37–38 weeks’) live births have 18.3 to 27.7 pp and 12.7 to 14.9 pp lower probabilities of second full term (≥ 39 weeks) live births respectively, when compared to individuals with first full term births in Georgia between 2011 and 2020. Instead, individuals with shortened gestational age for their first birth were at higher risk of subsequent preterm and early term births between 0.01 and 11.8 pp. Our findings are consistent with the existing literature on recurrent preterm births which documents an association between prior preterm births and subsequent preterm deliveries largely observed in other countries or the state of California [10, 12].

We also find that risk of recurrent preterm birth is greater with earlier preterm first births [31], but note that the increment in the probability of a second early term birth was reduced with earlier preterm first births. Inclusion of covariates to adjust for other observable individual-level and birth-specific risk factors that would signal increased risk of shortened gestational age in clinical settings only subtly reduced increment probabilities of subsequent preterm or early term births. Importantly, these findings suggest that first preterm or early term birth is useful in identifying at-risk populations in Georgia.

Our findings are especially important contributions to the limited literature on early term births, which includes studies with site-specific data (e.g. select hospitals in Europe) rather than population-wide representative data. Researchers reported that previous early term births were associated with the risk of future preterm and recurrent early term births [7, 16, 17]. We are aware of only one US-based study by Yang, et al. (2016) on the association between past early term birth and subsequent preterm or early term births using longitudinal data for the state of California between 2005 and 2011. Using the same gestational age categories, the authors reported (unadjusted) odds ratios between 2 and 3 for subsequent preterm births after a first early term birth (compared to a first full-term birth), and an odds ratio of 2.2 for recurrent early term birth. Estimating equivalent unadjusted odds ratios for individuals in Georgia, we find values between 2.1 and 1.8 for preterm and early term second births respectively after a first early term birth. Interestingly, the odds ratios for recurrent preterm were much smaller in our findings for Georgia. Yang et al. report odds ratios between 10 and 30 for recurrent preterm after a first preterm birth at < 32 weeks, whereas we find odds ratios between 4 and 15. For a first preterm birth between 32 and 36 weeks, Yang et al. report odds ratios between 6 and 8, and we find odds ratios between 3 and 5.

It may be that the experience of birthing individual in Georgia seen in our data is more representative of recent experiences in other southern states that have not yet expanded Medicaid eligibility as part of the Affordable Care Act. Still, our findings using Georgia data show some improvement in preterm recurrence rates relative to older pre-ACA data from California, a larger and richer state that ultimately did expand Medicaid coverage. We note that the expansion of health insurance coverage under the ACA has been found to increase Medicaid enrollment, the share of some pregnant individuals with a prenatal care visit in the first trimester and the receipt of guideline prenatal care, but it has not been associated with pre-pregnancy health, maternal health during pregnancy or birth outcomes [32–34].

Recurrence of shortened gestational age at delivery is linked to the number and order of past preterm births as well as to clinical differences between individuals who experience recurrent preterm births and those who do not [31, 35–38]. As in Yang et al., our supplementary findings show risk of recurrent preterm births does not solely reflect medically indicated first births. Clinical conditions that medically require earlier first deliveries are likely to affect second births. The robustness of our estimates to controls for indicated first birth suggest there remains other risk for recurrent shortened gestational age at births reflected in a previous spontaneous preterm or early term birth. We note however, the growing support for better understanding of underlying preterm causes, clinical presentation, and nutritional status beyond indicated versus spontaneous births to prevent preterm births [8]. We do not observe the extensive longitudinal data necessary to apply new preterm classification frameworks and note that even with the necessary data a third of preterm births still have no clear identifiable cause. Still, our findings of risk for shortened gestational age at delivery for second births following early term births supports calls to include early term births as preterm births in this literature [8]. Further, research on understanding longitudinal patterns in gestational age at delivery across heterogenous populations informs researcher efforts to understand preterm etiology.

Finally, while this study contributes to the limited literature on early term births, future analysis on late term (41 − 0/7 weeks through 41 − 6/7 weeks) and post-term birth (42 − 0/7 weeks and beyond) is needed as both are associated with increased risk for stillbirth, perinatal death and additional fetal risks of post-term births include macrosomia, neonatal seizures, and meconium aspiration [39, 40].

Researchers have made important progress in identifying risk factors for preterm births such as short cervical length, underweight maternal body mass index, short interpregnancy intervals, race and maternal smoking. Our findings underscore the continued importance of monitoring population-level birth outcomes in first and subsequent pregnancies, and the importance of recognizing that a first preterm or early term birth serves as an important risk indicator for shortened gestational duration in a subsequent pregnancy. Population based systemic longitudinal data collection for surveillance of birth outcomes is essential for evaluating management of high-risk pregnancies. Specifically, linked electronic health or vital record data could help inform health care providers of the increased risks for a given individual in their second pregnancy.

An implication for clinical practice is that it will likely require more targeted clinical interventions to mitigate risks of subsequent preterm or recurrent early term births. There are recognized interventions for high-resource clinical settings such as continued efforts to prevent non-medically indicated late preterm birth, prenatal care programs, progesterone therapy, and smoking cessation programs [41]. Expanded access to interpregnancy care that addresses medical issues or complications from previous pregnancies, in addition to screening for and management of chronic and behavioral health conditions, education on health risks, and optimization of health and health risks before a subsequent pregnancy can also reduce preterm birth risk [19, 32, 42–46].

There are limitations to some preterm birth interventions. For instance current Medicaid coverage for progesterone therapy to prevent preterm births varies by state [47]. Further, while some prenatal care programs have reduced very preterm births, not all studies report improved outcomes [48–50]. There are also additional challenges from scaling up programs beyond research settings. Failure to incorporate culturally sensitive program design and to engage community partners can limit success of program outreach in minority populations who experience disproportionately higher preterm birth rates [48, 51, 52].

Strength and limitations

Like all studies using observational retrospective data, we cannot make conclusions about the causal effect of shortened gestational age on future birth outcomes. Individuals who experience shortened gestational age at delivery likely have unobservable differences from individuals who do not experience shortened gestational age at delivery and those differences would be reflected in our findings. Although we acknowledge possible mechanisms and made efforts to explore this variation, our study aimed to evaluate first birth gestational age as a risk factor for subsequent birth outcomes. Still some important observable indicators for preterm risk were not observable in our data such as use of medications (corticosteroids and progesterone), past abortion or stillborn births. Further, case-wise deletion of individuals with missing data (Fig. 1) could have also introduced bias in our estimates.

Despite these limitations, our timely findings from Georgia provide a unique perspective into longitudinal birth outcomes in the US [15, 29]. Gestational age, like birth outcomes more generally, can reflect the environments individuals live in [36, 53, 54]. Georgia is a populous state with a racially and economically diverse population spread across metropolitan and rural regions that has relatively poor maternal and infant health outcomes for a high resource country [55]. Systematic consideration of past birth outcomes can signal individuals at risk for future preterm birth and better target interventions to ultimately improve outcomes in states like Georgia.

Conclusions

A first live birth that is early term or preterm predicts a higher probability of a second early term or preterm birth. Health information systems that flag both of these early birth outcomes on the problem list may help clinicians address known risk factors interconceptionally and prenatally (in a subsequent pregnancy).

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

ALD and PKC developed the study objective and design. PKC, MK and MD prepared the data for analysis. PKC, MD and LZ analyzed data and interpreted outcomes. PKC wrote the first draft. MK, EKA, ALD, and MD reviewed revised the manuscript. EKA was a major contributor in writing the manuscript. All authors read and approved the final manuscript.

Funding

PKC was funded by a KL2 from the Georgia Center for Translational Science Alliance (KL2TR002381) and UL1 (UL1TR002378).

Data availability

The data that support the findings of this study are available from Georgia Department of Public Health (DPH) but restrictions apply to the availability of these data, and so are not publicly available. Statistical code for the analysis for this study are however available from the authors upon reasonable request.

Declarations

Ethical approval and consent to participate

This study was reviewed by Emory University’s Institutional Review Board (#IRB00112252). The research adhered to the WMA Declaration of Helsinki – Ethical Principles for Medical Research Involving Human Participants Research. Emory University IRB determined consent was waived per 45 CFR Sect. 46 (the Common Rule) [20].

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.