Introduction
There is growing evidence that some of the most powerful influences on health and disease are related to exposures and influences that occur before and during pregnancy.Reference Misra, Guyer and Allston 1 – Reference Barker 3 For example, it is increasingly recognized that obesity during a woman’s child-bearing years may have antecedents in childhood and adolescence.Reference Ventura, Loken and Birch 4 , Reference Terry, Wei and Esserman 5 Obese women are more likely to enter pregnancy with a chronic disease,Reference Misra and Grason 6 thereby increasing the probability of maternal and infant morbidity and mortality. Obese women also have an increased risk of pregnancy complications, including gestational diabetes and hypertensive disorders of pregnancy, regardless of their health before pregnancy.Reference Yogev and Catalano 7
There is heterogeneity in the development of obesity and several distinct patterns of weight change across childhood and adolescence that are associated with later obesity.Reference Ventura, Loken and Birch 4 , Reference Ong, Ahmed, Emmett, Preece and Dunger 8 – Reference Koontz, Gunzler, Presley and Catalano 10 Most prior studies of obesity in pregnancy focused on variation in body weight at the time of conception and did not consider the change in body weight across the life course.Reference Nohr, Bech and Vaeth 11 – Reference Cnattingius, Villamor and Johansson 13 We postulate that a woman’s exposures and experiences across her life alter her body weight trajectory and can both positively and negatively alter the risk of adverse outcomes. Thus, women who become overweight or obese in adulthood may have different risks from those that have been overweight or obese since childhood. These risks may be due to the duration and severity of obesity. However, differences in body weight trajectories across the life course may also reflect the accumulation of factors that predispose women to obesity in adulthood and also influence the risk of adverse pregnancy outcomes beyond the level of obesity alone.Reference Ong, Ahmed, Emmett, Preece and Dunger 8 , Reference Ong 14
Several studies have reported an increased risk of preterm delivery (PTD) among obese women.Reference Salihu, Mbah and Alio 12 , Reference Cnattingius, Villamor and Johansson 13 , Reference Khatibi, Brantsaeter and Sengpiel 15 However, as suggested by the American College of Obstetricians and Gynecologists’ opinion on obesity in pregnancy, these results have been inconsistent across various studies. 16 We postulate that some of these inconsistencies may arise from differences in the trajectory of body weight over a woman’s life course that lead to obesity in pregnancy.
In this analysis, we examine the association between different body weight trajectories across the life course and PTD in a cohort of black women. Both obesity and PTD are more common among black women,Reference An 17 , Reference Gyamfi-Bannerman and Ananth 18 and differences in their body weight trajectories may reflect variation in the accumulation of risk factors over the life course that contribute to risk of PTD.
Methods
Data collection
Study participants were enrolled as part of a study of PTD from June 2009 through December 2011 at St. John Providence Health System in the metropolitan Detroit area. The study was approved by the St. John Providence Health System and Wayne State University Institutional Review Boards. All participants provided informed consent. Women were approached and consented within 24–48 h after delivery during their postpartum hospitalization. Women were eligible for inclusion in the study if they (1) self-identified as African-American or black; (2) had a live, singleton birth; and (3) were between 18 and 45 years of age. Women were excluded from the study if they (1) did not speak English; (2) had intellectual disabilities, serious cognitive deficits or significant mental illness on the basis of medical history or any prior records; or (3) were currently incarcerated. A total of 1410 (70.6%) women consented to participate in the study.
Data were collected from both interviews and medical records. Structured interviews were conducted during the postpartum hospitalization. Women were asked to retrospectively report weight at three different points in their life course: birth, age 18 and prepregnancy. In addition, information about race/ethnicity, current age, income and education was collected. Medical record abstraction was used to collect clinical information about the woman and her infant, including data to derive the infant’s gestational age at delivery.
A subset of the study participants were asked to provide contact information for their mother (grandmother of index child). We obtained consent from and interviewed 61.6% of the mothers for whom contact information was provided. Only 19 of the mothers (2.0%) refused to participate while the others could not be reached or were not contacted due to limited study resources. The mothers provided retrospective reports about the woman’s childhood, including the woman’s birthweight and weight at age 18.
The women’s birthweight, weight at age 18 and prepregnancy body weight were based on the women’s self-report at the time of the interview. For a minority of women, the mother’s report was used when self-report was unavailable. Birth length was not available; therefore percentiles of body weight were calculated from the sample distribution at birth. For the latter two time points (age 18 and prepregnancy), variance in weight due to height was accounted for by using the residuals from a regression model with weight as the outcome and height as the predictor to calculate the percentiles of body size. Prepregnancy body mass index (BMI) was calculated using height and prepregnancy weight (BMI=weight/height2) and classified as normal (BMI<25 kg/m2), overweight (25 kg/m2⩽BMI<30 kg/m2) or obese (BMI⩾30 kg/m2). 19
PTD was defined as giving birth at <37 completed weeks of gestation for the participant’s completed pregnancy. Gestational age was ascertained from the medical record, with priority given to estimates obtained from an early ultrasound (between 6–20 weeks gestation). Other estimates of gestational age were considered in a hierarchical fashion. When gestational age estimates from early ultrasound were not available or were implausible, the gestational age was estimated using the date of the last menstrual period. If early ultrasound estimates of gestational age and the date of the last menstrual period were both missing or implausible, estimates from a late ultrasound (>20 weeks gestation) were used. Rarely, the provider’s estimate of gestational age at birth was used.
Statistical analysis
Sociodemographic characteristics of women who had term and preterm infants were compared using t-tests and χ2 tests as appropriate using SAS version 9.2 (SAS Institute, Cary, NC, USA). Income was collected as a categorical variable and categories were subsequently collapsed based on the sample distributions to create a dichotomized variable. Maternal age was classified as <35 years of age or ⩾35 because women 35 and older have a well-documented increased risk of several adverse pregnancy outcomes.Reference Kenny, Lavender and McNamee 20 Body weight percentile trajectories were estimated using growth mixture models to classify women into subgroups that have similar body weight trajectories across the life course as implemented in the lcmm package in R v3.3.2. This method models subjects longitudinally and simultaneously estimates a subject’s likelihood of belonging to each of the trajectory groups via multinomial logistic regression. Three linear mixed effects models were tested on the entire cohort. These models included one with only fixed effects, one with a random intercept and one with a random intercept and random slope. The model with random intercept and random slope was not stable and thus could not validly estimate trajectory group membership probabilities. A likelihood ratio test confirmed that the random intercept model was the best fit for the data (P<0.001). The optimal number of groups and the shape of the trajectories were determined using Bayesian information criterion (BIC), Akaike information criterion (AIC) and posterior probabilities of group membership. These probabilities reflect how likely a subject is to belong to each group. The number of groups was allowed to vary from one to six. Participants with partially observed data are included in the model under the assumption that data are missing at random. However, participants who are missing all data for a time-dependent explanatory variable were dropped from the analysis. In our cohort, two women who were missing all three weight measurements, one woman who was missing height and one woman who was missing the outcome of interest (PTD) were excluded from the analysis. Therefore, the final sample size was 1406 women.
The groups in the best fitting model were then used to define weight percentile trajectory membership classes for each participant. Class membership was then used to examine the relationship between maternal body weight trajectory and PTD. Prevalence ratios (PR) and 95% confidence intervals (CI) were calculated using log-Poisson models with a sandwich variance estimator.
Results
Table 1 describes the characteristics of our sample. Of the 1406 pregnancies in our sample, 231 women delivered preterm (16%). Age, education, income, marital status and parity were similar for women who delivered term and preterm. A small percentage of women had chronic hypertension or diabetes before their pregnancy; these conditions were significantly associated with PTD. Prepregnancy BMI was not associated with PTD.
Table 1 Distribution of maternal characteristics by preterm delivery status

GED, general education diploma; BMI, body mass index.
**P<0.01; ****P<0.0001.
Table 2 presents cross-sectional associations between PTD and body weight percentiles at each of the three time points. Women whose own birthweight was less than the 25th percentile had a higher prevalence of PTD (PR=1.52 [95% CI 1.05, 2.22]). Larger birthweights were not associated with PTD. In cross-sectional analyses, the women’s weight percentiles at age 18 and in the prepregnancy period were not significantly associated with PTD.
Table 2 Cross-sectional associations between maternal body weight percentile categories and risk of preterm delivery

PR, prevalence ratio; CI, confidence interval.
Table 3 presents the model fit statistics (AIC and BIC) and posterior probabilities that were used to determine the optimal number of body weight percentile trajectory groups. The model with four body weight percentile trajectory groups was estimated to be the best fit for our sample. These trajectories can be described qualitatively based on the trajectory’s beginning and endpoint (high-high, high-low, low-high and low-low). The median of the membership posterior probabilities for each class were 0.81, 0.84, 0.77 and 0.78. Probabilities closer to one signify higher confidence in a subject’s class assignment. Figure 1 presents the trajectory for each woman in each group as well as the mean trajectory for each group. Figure 2 depicts the mean body weight percentile trajectory for the four groups.

Fig. 1 Individual and mean trajectories for each of the four trajectory groups. The thin colored lines are individual trajectories for each woman and the thick black line is the mean trajectory for that trajectory group.

Fig. 2 Maternal body weight percentile trajectory groups. The colored lines represent the four observed mean body weight percentile trajectory groups across the mother’s life course. These trajectories are described based on beginning and endpoint: high-high (purple vertical bar), high-low (green triangle), low-high (blue square) and low-low (orange circle).
Table 3 Latent class mixed model fit statistics and posterior probabilities

AIC, Akaike information criterion; BIC, Bayesian information criterion.
The proportion of births affected by PTD did not differ significantly across the groups (P=0.079). The trajectory group beginning at a low weight percentile (~25th) and ending near the high (~75th) percentile (low-high group) had the highest prevalence of PTD with about one-fifth of pregnancies ending preterm (21%). In the high-low group as well as the high-high group, 14% of births were preterm. The prevalence of PTD in women who began at a low weight percentile and remained small (low-low group) was 18%.
Table 4 describes the characteristics of women within each trajectory group. The sample was distributed across the trajectory groups with 34% in the low-low group, 16% in the high-low group, 16% in the low-high group and 33% in the high-high group. There were no significant differences in the women’s education, marital status, prevalence of previous PTD or parity between the groups. There were significant differences in prepregnancy BMI between the groups, such that women in the high-high and low-high groups were, on average, obese; those in the high-low and low-low groups were normal weight. There were also significant differences in prevalence of chronic hypertension, diabetes, income, maternal age at delivery and maternal height among the trajectory groups. The highest prevalence of chronic hypertension was among the women in the low-high and high-high groups.
Table 4 Distribution of maternal characteristics among maternal body weight percentile trajectory groups

IQR, interquartile range; BMI, body mass index; GED, general education diploma.
**P<0.01; ****P<0.0001.
Table 5 reports log-Poisson models quantifying the risk of PTD for each group when compared with the high-high trajectory group. In the unadjusted model, the low-high group had a significantly higher prevalence of PTD (PR=1.49, 95% CI=[1.11, 2.00]), but after adjusting for maternal age at delivery, income, chronic hypertension and diabetes the association between was no longer statistically significant. The low-low trajectory group had an increased prevalence of PTD relative to the high-high group (PR=1.35, 95% CI=[1.00, 1.83]) after adjusting for maternal age at delivery, income, chronic hypertension and diabetes. The unadjusted association was not statistically significant (PR=1.30, 95% CI=[0.97, 1.75]).
Table 5 Maternal body weight percentile trajectory groups and risk of preterm delivery

PR, prevalence ratio; CI, confidence interval.
a P-value for differences in preterm rate across the groups=0.079.
b Adjusted for maternal age, income, diabetes and hypertension.
Discussion
We found that PTD was associated with body weight trajectories across the life course in a cohort of black women. Our analysis revealed four distinct body weight trajectories starting at birth in this cohort of women. One group’s average trajectory remained consistently above the 70th percentile (high-high), and one group remained consistently below the 30th percentile (low-low), on average. One trajectory group (low-high group) showed a pattern of increasing weight percentiles, whereas one showed a pattern of decreasing (high-low group) weight percentiles.
The groups with the highest prevalence of PTD were those that began low (low-low and low-high groups), regardless of endpoint. In fact, the prevalence of PTD in the low-high group (21%) was substantially higher than the high-high group (14%), but the difference in PTD prevalence across the four groups did not reach statistical significance. Regardless, this finding suggests life course-dependent heterogeneity of prepregnancy body weight as a risk factor for PTD.
Our results corroborate prior cross-sectional studies examining the association between PTD and body weight across different life stages. Prior studies have suggested an inconsistent relationship between prepregnancy BMI and risk of PTD.Reference Nohr, Bech and Vaeth 11 – Reference Cnattingius, Villamor and Johansson 13 , Reference Khatibi, Brantsaeter and Sengpiel 15 , Reference McDonald, Han, Mulla and Beyene 21 – Reference Zhong, Cahill, Macones, Zhu and Odibo 23 In the results presented here, higher prepregnancy weight was not associated with PTD (Table 2). However, those women in the highest weight percentiles at the time of pregnancy but who were born small had the highest prevalence of PTD, whereas those born large and remained in the high percentiles had a lower prevalence of PTD (Table 5). Prior cross-sectional studies have found that women who were underweight at the time of conceptionReference Cnattingius, Villamor and Johansson 13 , Reference Khatibi, Brantsaeter and Sengpiel 15 , Reference Zhong, Cahill, Macones, Zhu and Odibo 23 , Reference Salihu, Lynch and Alio 24 or born low birthweight had a higher risk of PTD.Reference Sanderson, Emanuel and Holt 25 Consistent with these reports, our cross-sectional analysis also found an increased prevalence of PTD in women who were small at birth.
These results support the hypothesis that a woman’s own birthweight reflects an adaptive state that establishes her health risks over the life course.Reference Godfrey, Gluckman and Hanson 26 If the environments encountered over her life course are outside the anticipated range, she is ‘mismatched’ with a phenotype that may not be optimized for the environmental stressors encountered. These women might be expected to be at the highest risk for adverse health and reproductive outcomes. Accordingly, prior work suggests that girls with increasing weight who cross multiple percentiles during childhood and adolescence accrued higher levels of body fat and had the highest metabolic health risk compared with all other girls.Reference Ventura, Loken and Birch 4 Similarly, we find that large upward changes in weight percentiles from birth to adulthood (the low-high group) is associated with hypertension, as well as PTD.
Several limitations of this study merit mention. We relied on retrospective recall of body weight at each life stage (birth, age 18 and prepregnancy). Previous literature affirms that women can accurately report their own birthweight as well as the birthweight of their offspring.Reference Walton, Murray and Gallagher 27 – Reference Troy, Michels and Hunter 29 We have also previously reported a high level of agreement between the women’s self-reported birthweight and the birthweight reported by their mothers in this cohort.Reference Straughen, Caldwell, Osypuk, Helmkamp and Misra 30 For women of child-bearing age, others have shown that validity of recalled weight at 18 years of age is high.Reference Troy, Hunter and Manson 31 Our own prior work also suggests that women recall their weight at age 18 accurately as there was good agreement between self-reported weight and her mother’s report of the woman’s weight at age 18.Reference Straughen, Caldwell, Osypuk, Helmkamp and Misra 30 In addition, self-reported prepregnancy weight was within 5 kg of the prepregnancy weight recorded in the medical record for 87% of the women included in this cohort. Although some studies suggest that women tend to underreport their weight,Reference Perry, Byers, Mokdad, Serdula and Williamson 32 our use of weight percentiles in determining trajectories minimizes the impact of such misreporting. In addition, we did not use BMI in this analysis. The women in this study, like many other women, were unable to report their length at birth. Instead, variance in weight due to height was still incorporated the determination of the trajectory groups which allowed all time points to be included in the analysis. Although weight does not necessarily correlate with BMI, in this analysis, prepregnancy BMI differed across the trajectory groups such that the groups that ended high tended to have a higher BMI and those that ended low had a lower prepregnancy BMI (e.g. median prepregnancy BMI in the high-low group=23.0 and in the high-high group=32.5). Nonetheless, additional studies are needed.
Our findings may be impacted by the use of an internal reference population for calculating percentiles of body weight. This analysis focused on a cohort of black women, a group disproportionately affected by obesity and PTD. Differences in their weight trajectories may reflect variation in the accumulation of risk factors over the life course that then contributes to their PTD risk. As such, use of an external reference population may in effect dilute the importance of risk factor accumulation. As with all modeling, we caution that the four-class model identified in our sample may not represent ‘true’ underlying subpopulations of change in body weight percentiles. As such, our results may not be generalizable to other populations. Finally, we did not have detailed longitudinal information on covariates that could influence body weight across the life course, nor were we able to evaluate the potential impact of cohort effects. Additional studies that longitudinally measure these time-dependent variables in other demographic groups may help further characterize the association between body weight trajectories and PTD.
In conclusion, our results suggest that a woman’s birthweight, exposures and experiences across her life that alter her body weight trajectory may influence her pregnancy outcomes. Strategies to improve perinatal health have primarily focused on the prenatal, intrapartum and immediate postpartum periods, but such strategies fail to address the impact of child, adolescent and women’s health on maternal and infant outcomes. This study is among the first to examine the association between the trajectory of a woman’s body weight across multiple life stages and PTD in adulthood. Weight trajectories may also influence racial and ethnic disparities in birth outcomes.Reference Misra, Guyer and Allston 1 , Reference Misra and Grason 6 Our results suggest that improving pregnancy outcomes may require strategies that target factors across the life course and not exclusively in the prenatal period. Future analyses should focus on obtaining more measures of body size across the life course so that more detailed and possibly more informative trajectories can be developed and evaluated. Such information will ultimately be useful in improving women’s health and improving their pregnancy outcomes, including the risk of PTD.
Acknowledgements
The authors appreciate the work of their research assistants, particularly Laura Helmkamp and Dr Rhonda Dailey. The authors also acknowledge the statistical advice of Dr Robert Platt and Charlotte Burmeister. Finally, the authors are grateful for critical review and advice from Dr Vinod Misra.
Financial Support
This work was supported by the National Institutes of Health [NIH Grant #1R01HD058510]; and postdoctoral funding support for J.K.S. from Wayne State University.
Conflicts of Interest
None.
Ethical Standards
The authors assert that all procedures contributing to this work comply with the ethical standards described in the United States Federal Policy for the Protection of Human Subjects and with the Helsinki Declaration of 1975, as revised in 2008, and has been approved by the institutional committees at Wayne State University and St. John Providence Health System.