Maternal history of childhood maltreatment and pregnancy weight outcomes

Abstract

Background

Childhood maltreatment is associated with elevated adult weight. It is unclear whether this association extends to pregnancy, a critical window for the development of obesity.

Methods

We examined associations of childhood maltreatment histories with pre-pregnancy BMI and gestational weight gain among women who had participated for >20 years in a longitudinal cohort.

At age 26–35 participants reported childhood maltreatment (physical, sexual, and emotional abuse; emotional neglect) and, 5 years later, about pre-pregnancy weight and gestational weight gain for previous pregnancies (n=656). Modified Poisson regression models were used to estimate associations of maltreatment history with pre-pregnancy BMI and gestational weight gain z-scores, adjusting for sociodemographics. We used Multivariate Imputation by Chained Equations to adjust outcome measures for misclassification using data from an internal validation study.

Results

Before misclassification adjustment, results indicated a higher risk of pre-pregnancy BMI ≥30 kg/m² in women with certain types of maltreatment (e.g., emotional abuse RR=2.4; 95% CI: 1.5, 3.7) compared with women without that maltreatment type. After misclassification adjustment, estimates were attenuated but still modestly elevated (e.g., emotional abuse RR=1.7; 95% CI: 1.1, 2.7). Misclassification-adjusted estimates for maltreatment associations with gestational weight gain z-scores were close to the null and imprecise.

Conclusions

Findings suggest an association of maltreatment with pre-pregnancy BMI ≥30 kg/m² but not with high gestational weight gain. Results suggest a potential need for equitable interventions that can support all women, including those with maltreatment histories, as they enter pregnancy.

Introduction

Pre-pregnancy body mass index (BMI) and gestational weight gain have important influences on maternal and infant health (1–5). Clarifying who is vulnerable to high pre-pregnancy BMI and gestational weight gain can inform ways to optimize well-being before, during, and after pregnancy. Known risk factors for higher pre-pregnancy BMI include social determinants such as lower income and marginalized racial identity (6–9) and psychological factors such as depression and disordered overeating (10). Fewer factors have been identified that reliably predict high gestational weight gain; these factors include pre-pregnancy overweight and obesity, younger age, primiparity, and smoking (11–13).

One potential risk factor that has received little study is childhood maltreatment (abuse and neglect), which appears to be an important determinant of adult (non-pregnant) BMI (14) and obesity-related chronic disease (15), and disproportionately impacts on marginalized populations (16–18). Childhood maltreatment is thought to influence weight gain by disrupting emotion regulation (19), increasing disordered overeating (20), changing brain architecture in ways that increase the reward potential of food (21), and/or changing endocrine processes affecting appetite hormones (22,23). Each of these mechanisms may also operate during the perinatal period and lead to high pre-pregnancy BMI and gestational weight gain. However, most studies of risk factors for these conditions have focused narrowly on characteristics within the pre-conception and prenatal periods (12,13,24), limiting understanding of the extent to which earlier life experiences are linked to pregnancy weight measures.

The few prior studies examining associations of maltreatment with pregnancy-related weight have found that individuals with histories of maltreatment and other adverse childhood experiences have a greater risk of pre-pregnancy overweight and obesity, with findings more mixed for high gestational weight gain (25–28). A limitation of these studies is that most have been conducted in pregnancy cohorts with limited access to data on potential childhood confounders (e.g., childhood socioeconomic status). One exception is work in the National Longitudinal Survey of Youth (NLSY), which asked participants to retrospectively recall pre-pregnancy weight and gestational weight gain, finding a modest association of physical abuse with high gestational weight gain; the NLSY did not assess other child maltreatment types.

The nesting of pregnancy-related weight analyses within longitudinal cohorts such as the NLSY has the advantage of linking data from across the pre-pregnancy life course with pregnancy weight outcomes. However, this design generally requires retrospective self-report of pre-pregnancy BMI and gestational weight gain, which introduces measurement error in these outcomes (29). One approach to maintaining the benefits of this design while minimizing the impacts of measurement error is the use of quantitative bias analysis to adjust for misclassification in the self-reported measures (30). To date, quantitative bias analysis has not been used to adjust for misclassification in studies of childhood maltreatment and pregnancy-related weight.

The Life-course Experiences And Pregnancy (LEAP) study was designed to examine the associations of a range of childhood maltreatment types with pre-pregnancy BMI and gestational weight gain (31) while accounting for prospectively-measured childhood confounders. To address measurement error in self-reported pregnancy weight, internal validation data were used to adjust for misclassification using a multiple imputation approach. We anticipated that all maltreatment types would be associated with higher pre-pregnancy BMI and higher gestational weight gain, and, based on prior work (32), that emotional abuse would show the strongest associations.

Methods

Participants and procedures

The LEAP study enrolled women participating since 1998–1999 (age 11–18 years) in Project EAT (Eating and Activity in Teens and Young Adults), a Minnesota-based prospective cohort study of eating, activity, and weight-related health (33). Project EAT included 2,357 people identifying as girls or women at baseline. Details of LEAP are described elsewhere (31). Briefly, in 2019, women from Project EAT were invited to complete the LEAP survey if they had responded to two of the last three Project EAT surveys (n=1,267). Participants who had not consistently identified as women over survey waves (n=10), who reported a mailing address outside the US (n=4), or who had requested not to be contacted (n=1) were excluded, leaving n=1,252 recruited (Figure 1). Retrospective self-report data on LEAP participants’ reproductive histories, including pre-pregnancy weight and gestational weight gain for each live birth, were collected via a survey administered between July 2019 and July 2020. Of the 1,252 women invited to participate in LEAP, 977 (78%) completed the survey. Test–retest reliabilities of measures were assessed in a random sample of 141 respondents who completed the survey a second time within a two-week period. Of the 977 respondents, 656 reported at least one live birth. On average, these participants’ first live birth occurred 8 years prior to the LEAP survey. LEAP data were linked with four prior waves of Project EAT survey data (EAT-I through EAT-IV) on weigh-related risk factors, including retrospective self-reports of childhood maltreatment collected in 2014–2015 when participants were aged 25–36; 79% of parous women had had their first live birth before the time of maltreatment report. The current analysis was restricted to the first live birth for each woman and included singleton births to participants 18 years of age or older at the time of their first birth (n=607; Figure 1).

Figure 1.

Analytic sample and validation subsample selection

For validation of self-reported pregnancy weight measures, participants reporting at least one live birth were invited to sign HIPAA Authorization and medical records release forms authorizing the study team to request their medical records. Of the 656 eligible participants, 379 (58%) provided authorization for records release, and study staff obtained records for 250 (38%; Figure 1). The 250 participants with records were similar to the overall eligible population, except that they were somewhat more likely to be White (31). Records data were abstracted by trained study staff, with double data entry for 61 randomly sampled medical records; overall inter-rater reliability was high (Cohen’s Kappa: 0.97). Details of the validation substudy are presented elsewhere (34).

The University of Minnesota Institutional Review Board Human Subjects Committee approved all protocols implemented in the LEAP Study.

Measures

Details on LEAP pregnancy weight and child maltreatment measures are provided in Tables 1 and 2.

Table 1.

Pregnancy weight measures

Variable	Survey measure	Medical record measure
Pre-pregnancy weight status	BMI (weight in kg / [height in m]2) calculated from: - Recall of weight just prior to pregnancy resulting in first live birth (test-retest r=0.96) - Height from LEAP (test-retest r=0.94); if missing, from EAT-IV (age 25–36) or EAT-III (age 19–28)	BMI (weight in kg / [height in m]2) calculated from: - Pre-pregnancy weight recorded in the prenatal or delivery record; if missing, measured weight from first prenatal visit if ≤ 8 weeks gestation - Height from LEAP (test-retest r=0.94); if missing, from EAT-IV (age 25–36) or EAT-III (age 19–28)
GWG z-score	Recalled total GWG during pregnancy resulting in first live birth; standardized based on recalled gestational age and pre-pregnancy BMI category, using gestational age and pre-pregnancy BMI-standardized z-score charts	Total weight gain calculated from: - Delivery weight recorded in the delivery record; if missing, last measured weight if prenatal visit ≤ 7 days of delivery date - Pre-pregnancy weight recorded in the prenatal or delivery record; if missing, measured weight from first prenatal visit if ≤ 8 weeks gestation Total weight gain standardized based on medical record gestational age and pre-pregnancy BMI category, using gestational age and pre-pregnancy BMI standardized z-score charts; gestational age primarily based on date of last menstrual period (LMP) from first ultrasound (if performed at <14 weeks gestation); if missing, based on self-reported LMP from first prenatal visit or gestational age from delivery record

Table 2.

Childhood maltreatment variables, survey items, response options, cut-points, and psychometrics

Childhood maltreatment variable	Survey question(s) or statement	Response options	Cut-point	Test-retest agreement
Sexual abuse	Nonfamilial sexual abuse: “Did someone not in your family touch you in a sexual way against your wishes or force you to touch them in a sexual way?” Familial sexual abuse: “Did someone in your family touch you in a sexual way against your wishes or force you to touch them in a sexual way?”	“no,” “once,” or “more than once”	No vs. once or more than once	99% (nonfamilial); 99% (familial)
Physical abuse	“An adult in my family hit me so hard it left bruises or marks.”	“never,” “rarely,” “sometimes,” “often,” or “very often”	Never vs. Rarely, sometimes, often, or very often	90%
Emotional abuse	“An adult in my family said hurtful or insulting things to me.”	“never,” “rarely,” “sometimes,” “often,” or “very often”	Never, rarely, or sometimes vs. Often or very often	93%
Emotional neglect	“My family was a source of strength and support.”	“never,” “rarely,” “sometimes,” “often,” or “very often”	Sometimes, often, or very often vs. Rarely or never	82%

Pre-Pregnancy BMI.

The LEAP survey asked women to estimate their weight just before each pregnancy resulting in a live birth. Pre-pregnancy BMI for the first live birth was calculated from this reported pre-pregnancy weight (test–retest Pearson correlation for first birth: 0.96) and women’s height reported on the LEAP survey (test–retest: 0.94). If height was missing on LEAP (n=9, 1%), height reported on either EAT-IV (n=5; age 25–36; test–retest: 0.98) or EAT-III (n=4; age 19–28; test–retest: 0.99) was used instead. We dichotomized pre-pregnancy BMI into two outcomes of interest: ≥30 vs. <30 kg/m² and BMI ≥25 vs. <25 kg/m².

Gestational Weight Gain.

Women reported approximately how much weight they gained in the pregnancy resulting in their first live birth (test–retest: 0.94). We converted total gestational weight gain to z-scores using gestational-age and pre-pregnancy BMI-standardized z-score charts (35,36). Gestational weight gain z-scores provide a measure that is independent of gestational duration and are specific to pre-pregnancy BMI category (underweight, normal weight, overweight, and obesity) (37). We dichotomized gestational weight gain z-scores into two outcomes of interest: high (>+1 SD vs. ≤+1 SD) and low gestational weight gain (<−1 SD vs. ≥−1 SD). We used these z-score cut-offs because the corresponding weight gain is outside of guidelines for all BMI groups.

Medical record pre-pregnancy BMI.

For the validation subsample, medical record-based pre-pregnancy BMI for the first live birth was calculated using pre-pregnancy weight recorded in the prenatal or delivery record or, if missing, a measured weight at a prenatal visit occurring at ≤8 weeks gestation (31). Pre-pregnancy BMI was then calculated from the recorded pre-pregnancy weight and height reported on LEAP (or, if missing, on EAT-IV or EAT-III).

Medical record gestational weight gain.

For the validation subsample, we abstracted weight at delivery from the delivery record or, if missing, from the last measured weight at a prenatal visit if it was within seven days of the delivery date (31,37). We calculated gestational weight gain by subtracting the medical record-based pre-pregnancy weight from this delivery weight. As with survey-based gestational weight gain, we converted the medical record-based measure to z-scores based on pre-pregnancy BMI and gestation length and dichotomized into high vs. non-high and low vs. non-low for analysis.

Childhood maltreatment.

Exposure to physical abuse, sexual abuse, emotional abuse, and emotional neglect in childhood (<18 years old) was captured retrospectively on the EAT-IV survey when participants were ages 25–36 (38). Physical and emotional abuse and neglect were assessed using items from the Childhood Trauma Questionnaire (39); sexual abuse was ascertained using questions developed by Finkelhor et al. (40). Test–retest reliability in the LEAP cohort ranged from 0.82 for emotional neglect to 0.99 for sexual abuse. Cut-points for dichotomizing each childhood maltreatment variable followed the Adverse Childhood Experiences (ACE) scale (41). Table 2 presents the survey questions, cut-points, and test–retest agreement for each maltreatment type. Maltreatment types were examined separately and combined into an overall maltreatment exposure coded as a dichotomous variable (any vs. none) and as an ordinal variable derived from the sum of maltreatment types (range of 0 to 4) and categorized as 0, 1, and 2+.

Covariates.

Main analyses adjusted for childhood race and ethnicity and socioeconomic status (SES), sociodemographic characteristics that are associated with both maltreatment and pregnancy weight and not on the causal pathway. Maternal race and ethnicity were based on self-report on the EAT-I baseline survey and dichotomized as White vs people of color (Black, Hispanic, Asian, or Other race) to maintain adequate cell sizes. Maternal childhood SES was based on highest parental education reported on the EAT-I baseline survey, with missing or implausible values imputed based on family eligibility for public assistance, eligibility for free/reduced-price school meals, and parental employment status (42). Childhood SES was coded as a 5-category variable (low, low-middle, middle, upper-middle, and high, with low as the referent). Because no participants with high childhood SES reported experiencing emotional neglect, analyses of emotional neglect included childhood SES as a 4-category variable, with the upper-middle and high SES categories combined to ensure non-zero strata.

Adolescent BMI z-score, maternal age at delivery, and maternal SES in adulthood may serve as proxies for unmeasured childhood confounders (e.g., social class, cultural context). However, these variables could also be mediators given that they may be downstream from childhood maltreatment. Thus, we ran models with and without adjustment for these variables to assess their influence on estimates. Adolescent BMI was measured at EAT-I (43). Maternal age in years at delivery (continuous) was ascertained through self-report on the LEAP survey. Maternal educational attainment was captured via self-report on the EAT-IV survey and dichotomized (high school or less vs. some college or more).

Statistical analyses

We first used non-imputed LEAP survey data to examine the association of maltreatment with pre-pregnancy BMI and gestational weight gain. For each maltreatment exposure–outcome association, we fit a separate modified Poisson regression model with robust standard errors to estimate risk ratios and 95% confidence intervals. Model 1 adjusted for only race and ethnicity and childhood SES. Model 2 further adjusted for adolescent BMI z-score, maternal age at delivery, and maternal education. These analyses were restricted to complete cases; final analytic sample sizes ranged from 523 to 526 participants for the pre-pregnancy BMI outcomes (86% to 87% of overall sample, respectively) and 504 to 507 participants for the GWG outcomes (83% and 84% of overall sample, respectively). Data preparation and analyses were performed in Stata version 17 (StataCorp, College Station, TX); figures were created using R version 4.1.0 (R Core Team, Vienna, Austria).

Misclassification-adjusted models.

We next fit misclassification-adjusted models to address measurement error in the self-reported outcome variables (BMI and gestational weight gain) using medical records data from the internal validation study. Medical records data were available on pre-pregnancy BMI for n=153 (25% of the analytic sample) and on gestational weight gain z-scores for n=133 (22% of the analytic sample). For the participants with survey data only, we used Multivariate Imputation by Chained Equations with the mi impute package in Stata to impute the medical record-based information needed to calculate our study outcomes (pre-pregnancy weight, delivery weight, and gestational age) (44). Predictor data to inform the imputed values included the exposures, survey-based BMI and gestational weight gain measures, and sociodemographics (eTable 1). We also imputed missing exposures and covariates; imputation models were run separately for each of the exposure variables to improve model fit and stability. We specified 200 burn-in iterations and created 100 imputed datasets. Imputation convergence was assessed visually using trace plots. Finally, analytic models were fit in each of the 100 imputed datasets and combined using Rubin’s rules (45). For each analytic model, we examined imputation performance, including relative increase in variance, fraction of missing information, and relative efficiency.

The calculation of gestational weight gain z-scores involves log-transformation of weight gain; negative weight gain values have a small constant added to ensure nonnegative values to allow for log-transformation (35). In a small number of imputations (n=2–8) the imputed gestational weight gain had a negative value more extreme than the value of this constant; thus, we truncated the value to a minimum negative value to ensure nonnegative weight. This truncation had no effect on participant gestational weight gain z-score classification.

Sensitivity analyses.

To assess whether findings were sensitive to our original definitions of maltreatment we used alternative cut-points for sexual abuse (more than once vs. never or once) and physical abuse (sometimes, often, or very often vs. never or rarely).

Results

Sample Characteristics

Overall, 392 (65%) women reported no maltreatment during childhood, while 86 experienced one type and 71 experienced two or more types (Table 3). Compared to women without a history of childhood maltreatment, those who experienced maltreatment were more likely to be Black, Hispanic, Asian, or Other race (people of color were 46% of those who experienced maltreatment versus 27% of those who did not); report low or low-middle childhood SES (41% versus 23%); and report educational attainment of high school or less (24% versus 16%; Table 4). Women in the maltreatment group were younger at their first birth (mean=26.6 years [SD=5.1] versus 28.3 years [SD=4.6]) and had a slightly higher adolescent BMI z-score (mean=0.54 [SD=0.91] versus mean=0.48 [SD=0.85]).

Table 3.

Reported childhood maltreatment overall and by type in the LEAP survey sample (n=607)

Variables	Total, n (column %)	Any Maltreatment Exposure, n (column %)
Ordinal maltreatment exposure
No maltreatment types	392 (65%)	--
1 maltreatment type	86 (14%)	86 (54%)
2+ maltreatment types	71 (12%)	71 (45%)
Missing	58 (10%)	1 (1%)
Sexual abuse
No	480 (79%)	85 (54%)
Yes	72 (12%)	72 (46%)
Missing	55 (9%)	1 (1%)
Physical abuse
No	468 (77%)	74 (47%)
Yes	84 (14%)	84 (53%)
Missing	55 (9%)	0 (0%)
Emotional abuse
No	498 (82%)	104 (66%)
Yes	54 (9%)	54 (34%)
Missing	55 (9%)	0 (0%)
Emotional neglect
No	496 (82%)	103 (65%)
Yes	55 (9%)	55 (35%)
Missing	56 (9%)	0 (0%)

Note: The above statistics are based on survey data prior to imputation of missing values. The any maltreatment exposure variable was missing for n=57 (9.4%) observations; thus, the any maltreatment subsample counts do not add up to the overall study sample counts. Percentages may not add up to 100 due to rounding error.

Table 4.

Characteristics of the LEAP survey sample overall and by maltreatment exposure (n=607)

Variables	Total	Any Maltreatment Exposure, n (column %) or Mean (SD)
		No	Yes
Total sample, n (%)	607 (100%)	392 (64%)	158 (26%)
Race, n(%)
White	399 (66%)	283 (72%)	85 (54%)
Black, Hispanic, Asian, or Other race	201 (33%)	105 (27%)	72 (46%)
Missing	7 (1%)	4 (1%)	1 (1%)
Childhood socioecomonic status, n(%)
Low	80 (13%)	40 (10%)	33 (21%)
Low-middle	89 (15%)	50 (13%)	31 (20%)
Middle	158 (26%)	99 (25%)	39 (25%)
Upper-middle	181 (30%)	123 (31%)	41 (26%)
High	90 (15%)	76 (19%)	10 (6%)
Missing	9 (2%)	4 (1%)	4 (3%)
Educational attainment, n (%)
High school/GED or less	102 (18%)	63 (16%)	38 (24%)
Some post-high school training or college	452 (75%)	329 (84%)	120 (76%)
Missing	53 (9%)	0 (0%)	0 (0%)
EAT-I (age 11–18) BMI z-score, mean (SD)	0.49 (0.88)	0.48 (0.85)	0.54 (0.91)
Age at 1st delivery (years), mean (SD)	27.6 (4.9)	28.3 (4.6)	26.6 (5.1)
Gestational age at delivery (weeks), mean (SD)	39.1 (2.3)	39.2 (2.4)	38.8 (2.4)

Abbreviations: GED, General education development. BMI, Body mass index.

Note: The above statistics are based on survey data prior to imputation of missing values. Percentages may not add up to 100 due to rounding error.

Based on survey data without outcome misclassification adjustment, 42% (n=256) of women reported pre-pregnancy BMI ≥25 kg/m² and 17% (n=105) reported pre-pregnancy BMI ≥30 kg/m² (Table 5). High gestational weight gain (z-score >+1 SD) was reported for 89 women (15%) and low gestational weight gain (z-score <−1 SD) for 102 (17%). The proportion of women with pre-pregnancy BMI ≥25 kg/m² was similar between women with and without maltreatment (41% versus 43%). Women with maltreatment were more likely to report a pre-pregnancy BMI ≥30 (23% versus 14%). High gestational weight gain was more common among those with a history of maltreatment (18% versus 13%).

Table 5.

Pregnancy weight measures in the LEAP survey sample overall and by maltreatment exposure (n=607)

Variables	Total	Any Maltreatment Exposure, n (column %) or Mean (SD)
		No	Yes
Pre-pregnancy BMI (kg/m2), mean (SD)	25.5 (5.9)	25.3 (5.4)	26.0 (6.7)
Pre-pregnancy BMI ≥25 kg/m2 n (%)
No	334 (55%)	214 (55%)	88 (56%)
Yes	256 (42%)	169 (43%)	65 (41%)
Missing	17 (3%)	9 (2%)	5 (3%)
Pre-pregnancy BMI ≥30 kg/m2, n (%)
No	485 (79%)	328 (84%)	117 (74%)
Yes	105 (17%)	55 (14%)	36 (23%)
Missing	17 (3%)	9 (2%)	5 (3%)
Total absolute gestational weight gain (kg), mean (SD)	15.4 (8.4)	15.5 (8.4)	15.1 (8.4)
High gestational weight gain z-score (>1 SD), n (%)
No	474 (78%)	320 (82%)	119 (75%)
Yes	89 (15%)	50 (13%)	28 (18%)
Missing	44 (7%)	22 (6%)	11 (7%)
Low gestational weight gain z-score (<-1 SD), n (%)
No	461 (76%)	302 (77%)	120 (76%)
Yes	102 (17%)	68 (17%)	27 (17%)

Abbreviations: BMI, Body mass index.

Note: The above statistics are based on survey data prior to imputation of missing values. Percentages may not add up to 100 due to rounding error.

Childhood maltreatment and Pre-pregnancy BMI

Before pre-pregnancy BMI misclassification adjustment, results based on survey data alone suggested an association for all maltreatment types with pre-pregnancy BMI ≥30 kg/m², except for the ordinal maltreatment exposure of 1 vs. 0 types. RRs ranged from 1.4 (95% CI: 0.9, 2.3) for physical abuse to 2.4 (95% CI: 1.5, 3.7) for emotional abuse (Figure 2 and eTable 2). After misclassification adjustment, associations were attenuated and ranged from 1.2 (95% CI: 0.7, 1.8) for physical abuse to 1.7 (95% CI: 1.1, 2.7) for emotional abuse. For pre-pregnancy BMI ≥25 kg/m², estimates with and without misclassification adjustment were close to the null for all maltreatment types, with confidence intervals compatible with both small negative and small positive associations. Misclassification adjustment narrowed the confidence intervals for estimates of pre-pregnancy BMI. We believe this is because the range of available BMI measures in our data allowed us to accurately impute medical record pre-pregnancy BMI; in addition, the imputation added observations dropped from the non-adjusted complete-case analyses. Additional adjustment for maternal age at delivery, baseline BMI z-score, and maternal education did not meaningfully change estimates (eFigures 1–2 and eTable 3).

Figure 2.

Associations of childhood maltreatment with pre-pregnancy BMI and gestational weight gain z-score adjusted for maternal race and ethnicity and childhood socioeconomic status, in LEAP survey-only (not misclassification-adjusted) analyses and misclassification-adjusted analyses

In sensitivity analyses with alternative cut-points for sexual and physical abuse (eTable 4), sexual abuse findings were imprecise and did not show a different pattern than the main results. However, the more stringent cut-point for physical abuse yielded a stronger association for BMI ≥30 kg/m² (misclassification-adjusted RR=1.8; 95% CI: 1.2, 3.0), indicating that the main findings may be diluted by the inclusion of less severe physical abuse in the exposed group.

Childhood maltreatment and gestational weight gain z-score

Before misclassification adjustment, point estimates based on survey data alone suggested a slight positive association of maltreatment with high gestational weight gain, but with 95% CIs consistent with both negative and positive associations. RRs ranged from 1.2 (95% CI: 0.6, 2.1) for emotional abuse to 1.4 (95% CI: 0.8, 2.4) for the ordinal maltreatment exposure of 1 vs. 0 types (Figure 2 and eTable 2). Misclassification adjustment moved the estimates toward, and in some cases beyond, the null. The estimated associations of maltreatment types with low gestational weight gain z-score were generally below 1 before misclassification adjustment; misclassification adjustment moved estimates close to or beyond the null (Figure 2 and eTable 2), although confidence intervals were wide. As with pre-pregnancy BMI, additional adjustment for maternal age at delivery, baseline BMI z-score, and maternal education did not change results (eFigures 1–2 and eTable 3).

In sensitivity analyses with alternative cut-points for sexual and physical abuse (eTable 4), findings for both exposures were highly imprecise and did not show a different pattern than the main results.

Discussion

In this cohort of predominantly White women, we examined the extent to which a history of childhood maltreatment is associated with pre-pregnancy BMI and gestational weight gain, adjusted for outcome misclassification using medical record validation data. Our results suggest that maltreatment may be linked with higher pre-pregnancy BMI. Specifically, we found a modest association between emotional abuse and pre-pregnancy BMI ≥30 kg/m². Although main analysis results provided little evidence that physical abuse was associated with pre-pregnancy BMI ≥30 kg/m², sensitivity analyses with a more stringent definition of physical abuse did suggest an association. For both high and low gestational weight gain z-score, point estimates were generally close to the null, although wide 95% confidence limits suggest that the data are compatible with a range of positive and negative effect estimates. Our findings suggest that tailored preconception supports for women with maltreatment histories could be helpful for improving pregnancy health.

Our results for pre-pregnancy BMI are consistent with a small body of prior work showing positive associations between childhood maltreatment and higher pre-pregnancy BMI (46) or with greater risk of pre-pregnancy overweight or obesity (23–28); one study only found these associations in women with anxiety (47). Our study builds on this literature by examining a wider array of maltreatment types, adjusting for prospectively measured childhood covariates, and adjusting for outcome misclassification using internal validation data.

The existing evidence on associations of maltreatment with higher gestational weight gain is mixed. One study reported a modest association of physical abuse with high gestational weight gain (26), while another found an association for 3+ adverse childhood experiences (including maltreatment) with high gestational weight gain, although this estimate was highly imprecise (48). A third study reported a small association only in women with anxiety (47), and a fourth reported null findings (46). In the current study, we find little evidence to support a link between childhood maltreatment and either high or low gestational weight gain z-score, a contrast to most prior studies; however, our confidence intervals are compatible with a range of estimates, including those found in previous research. Our point estimates before misclassification adjustment were suggestive of a slight (though highly imprecise) positive association between some types of maltreatment and high gestational weight gain, and misclassification adjustment shifted these estimates toward or beyond the null. It is possible that some of the positive findings in the published literature stem in part from errors in self-reported gestational weight gain.

Our study design differed from most prior studies on this topic, which recruited pregnant women from prenatal clinics and collected survey data on recalled childhood maltreatment histories and pre-pregnancy weight during or shortly after pregnancy (27,28,46,47). We instead recruited women from an existing longitudinal cohort study focused on weight-related health, giving us a rich array of relevant risk factor data from across the life course. For the current analysis, this approach allowed us to adjust for prospectively collected childhood covariate data. Ranchod et al. used a similar approach in the NLSY, which collected retrospective self-report data on childhood physical abuse, pre-pregnancy body weight, and gestational weight gain (26). Key differences between our study and the NLSY are validation of pre-pregnancy BMI and gestational weight gain and the inclusion of a range of maltreatment measures (the NLSY includes only physical abuse). Notably, our physical abuse point estimates from the survey data-only main analyses are similar to Ranchod et al.’s findings of RR=1.6 for pre-pregnancy obesity (95% CI: 1.1, 2.2) and RR=1.2 for excessive gestational weight gain (95% CI: 1.1, 1.4) (26); our estimates were stronger in sensitivity analyses that imposed a more stringent definition of physical abuse. Although some of our strongest findings were for emotional abuse, we cannot compare these to the NLSY, as emotional abuse information is not available in that cohort.

Comparison of self-reported pre-pregnancy weight and gestational weight gain with analogous measures from medical records as part of the LEAP validation sub-study indicate that many women retrospectively report their pregnancy weight with reasonable accuracy (34), though women at the extremes of the BMI and gestational weight gain distributions are less accurate. Further, we found that misclassification varied by maltreatment exposure status. In the case of pre-pregnancy BMI, there was greater under-reporting and resultant misclassification among women without maltreatment than among women with maltreatment. In contrast, women with maltreatment histories were more likely to over-report their gestational weight gain, leading to more misclassification into a higher z-score category; women without maltreatment histories were instead more likely to under-report their weight gain, resulting in more misclassification into a lower z-score category. Both mechanisms of misclassification would be expected to bias results away from the null. These findings suggest that research on the downstream consequences of maltreatment and related exposures (e.g., other traumas) should prioritize validating self-reported outcome measures.

This study had several limitations. First, attrition in the cohort over time may have resulted in selection bias. Adjustment for baseline demographics predicting attrition (childhood SES and race) should help mitigate selection impacts, but many attrition-related factors are likely unmeasured. Second, childhood maltreatment was retrospectively self-reported, which may result in recall or response biases. Given that there is no gold standard for maltreatment measurement, and the accuracy of prospective and/or objective measures (e.g., through child welfare records) is low (49), retrospective self-report is often the best, if not the only, option in epidemiologic studies. Third, the study sample is more affluent and more likely to be White than the average population of childbearing women in the US, and is centered in the Midwest, which may limit generalizability; it is possible that this relatively privileged cohort benefits from buffers that may reduce the impact of maltreatment on their pregnancy weight relative to the impacts that would be seen in other cohorts. Fourth, the validation data available were not from a random sample and were, themselves, subject to possible selection bias.

Despite these limitations, several important strengths of this study help to fill gaps in the existing literature. First, we included a comprehensive examination of multiple maltreatment types; notably, emotional abuse has not been consistently measured in prior studies but appears from our findings to be a potentially important exposure. Second, we adjusted survey self-reports of pregnancy weight measures for misclassification relative to medical records. Medical records data are not a true gold standard (which would involve measurement by trained research staff in a controlled setting), but they were collected either prospectively or with short recall times, improving their accuracy over our retrospective survey data. Finally, embedding this study within a larger cohort gives us access to prospectively measured childhood variables. For the current study, this design allowed us to control for important childhood confounders that might otherwise not be available; this design also provides a foundation for future assessments of a range of prospectively measured childhood and adolescent variables as predictors of pregnancy health.

Our findings are consistent with the hypothesis that women with histories of childhood maltreatment, particularly emotional and physical abuse, may be at greater risk for high pre-pregnancy BMI. Additional research is required to determine the extent to which these association are causal, and the potential mechanisms involved. If our results are replicated future research should consider how to support optimal pregnancy health for those with maltreatment histories.