Are detailed behavioral, psychosocial, and environmental variables necessary to control for confounding in pregnancy weight gain research?

Abstract

Background:

Associations between pregnancy weight gain and adverse outcomes may be spurious due to confounding by factors not typically measured in cohort studies. We determined the extent to which the addition of detailed behavioral, psychosocial, and environmental measurements to commonly available covariates improved control of confounding.

Methods:

We used data from a prospective U.S. pregnancy cohort study (2010‒2013, n=8978). We calculated two propensity scores for low and high pregnancy weight gain (vs. adequate gain) using 11 standard confounders (e.g., age, education). We examined the balance of characteristics between weight gain groups before and after propensity score matching. We used negative binomial regression to estimate the association between weight gain and small- and large-for-gestational-age birth, preterm birth, and unplanned Cesarean delivery, controlling for propensity score. To this model we then added 17 detailed behavioral, psychosocial, and environmental measurements (‘fully adjusted’). We calculated the risk ratio due to confounding as the ratio of the standard confounder-adjusted risk ratio to the fully adjusted risk ratio.

Results:

There were minimal imbalances between weight gain groups in detailed measures after matching for a propensity score of standard covariates. Accordingly, the inclusion of detailed covariates had minimal impact on estimated associations between low or high pregnancy weight gain and adverse pregnancy outcomes: risk ratios due to confounding were null for all outcomes (e.g., 1.1 [95% CI: 1.0, 1.1] for low weight gain and preterm birth).

Conclusions:

Adjustment for detailed behavioral, psychosocial, and environmental measurements had minimal impact on estimated associations between pregnancy weight gain and adverse perinatal outcomes.

Introduction

National guidelines in the U.S. on optimal weight gain during pregnancy have existed for over sixty years. These recommendations are primarily based on observational research linking pregnancy weight gain with a broad range of adverse maternal and child health outcomes, such as excess postpartum weight retention, childhood obesity, preterm birth, and small- and large-for-gestational-age birth.¹ The studies that have been most influential in guideline development were conducted using large administrative databases or perinatal registries.¹ However, these data sources typically contain limited covariate data and lack detailed information on key behavioral, psychosocial, and environmental factors that may be common causes of both pregnancy weight gain and adverse pregnancy outcomes. Thus, there are concerns that observed associations may reflect bias due to confounding rather than a causal effect of weight gain.^1–5

Collecting more comprehensive covariate information on diet, physical activity, and other behaviors, psychosocial factors, and the built environment through detailed, prospective studies could be one strategy to reduce concerns about confounding of pregnancy weight gain studies. However, it is not known whether this gain of information is worth the added burden. The objective of this study was to determine the extent to which the addition of detailed behavioral, psychosocial, and environmental measurements to commonly available covariates improved control of confounding in the study of pregnancy weight gain and adverse health outcomes.

Methods:

Study population

We used data from the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-be (nuMoM2b), a large, prospective U.S. pregnancy cohort (2010‒2013).⁶ A total of 10,038 women from eight medical centers across the U.S. were enrolled if they had a viable singleton pregnancy, were 6‒13 completed weeks of gestation, and had no previous pregnancy that lasted ≥20 weeks’ gestation. At enrollment, trained and credentialed study personnel conducted detailed interviews to ascertain data on demographics, medical history, and behaviors. Two other study visits occurred in the second and third trimesters. After delivery, a trained, certified chart abstractor recorded birth outcomes. Each site’s local institutional review board approved the study and all women gave written, informed consent. This secondary analysis of the nuMoM2b data was deemed by our office of human protections as not human subjects research because it used deidentified data.

The present study was limited to pregnancies that progressed past 20 weeks’ gestation. For simplicity, we excluded the small proportion of participants with missing data on pregnancy weight gain, prepregnancy BMI, perinatal outcomes, and standard covariate values. Participants with missing data for detailed confounders were retained, and missing data was addressed using multiple imputation (described below).

Pregnancy weight gain

At the initial visit, pregnant people had their height measured and self-reported their prepregnancy weight. BMI was calculated as prepregnancy weight (kg)/height (m)².

Total gestational weight gain was calculated by subtracting prepregnancy weight from the last measured weight (abstracted from the medical record at a median [IQR] of 0.14 [0.43] weeks before delivery). Longer pregnancies have more time to gain weight. Ignoring the correlation between total pregnancy weight gain and gestational duration will lead to spurious associations between low weight gain and any gestational age‐dependent outcome (e.g., preterm birth).⁷ This remains a problem when total pregnancy weight gain is evaluated relative to current IOM pregnancy weight gain guidelines (ie, below, within, or above the recommended ranges for term deliveries).¹ We overcame this problem by standardizing total gestational weight gain for gestational age at delivery using prepregnancy BMI category-specific z-score charts for singleton pregnancies.^8,9 Converting total weight gain to gestational age-standardized z-scores produces a measure that is independent of gestational duration.¹⁰ The pregnancy weight gain z-score charts that we used were developed in a Pennsylvania-based cohort. We then defined weight gain as above, within, or below recommendations by estimating the z-scores that correspond to the upper and lower bounds of the guidelines at 40 weeks : underweight: 12.5‒18 kg = −0.59 to 0.54 SD , normal weight: 11.5‒16 kg= −0.96 to −0.09 SD , overweight: 7‒11.5 kg =−1.31 to −0.62 SD , obese: 5‒9 kg =−1.0 to 0.23 SD ) ¹.

Confounders

We identified potential confounders using the 2009 IOM Committee’s conceptual framework that includes determinants of pregnancy weight gain.¹ From this list, we selected all predictors on which we had a measure, and hypothesized that these covariates may have a causal association with adverse pregnancy outcomes. All were upstream from weight gain and adverse pregnancy outcomes. These variables therefore met the definition of potential confounders.¹¹ Parameterization of variables was guided by considerations that were both statistical (grouping to preventing sparse cells, categorizing to account for highly skewed data) and substantive (creating categories with clinical-relevant interpretation) considerations.

Standard confounders

We identified confounders that are routinely available in large clinical and population health datasets (referred to hereafter as ‘standard confounders’). Standard confounders derived from self-report at enrollment were the participant’s age, education, race/ethnicity, marital status, pre-pregnancy smoking, medical insurance, and gravidity. Preexisting diabetes, chronic hypertension, and use of assisted reproductive technologies were collected via medical record abstraction.

Detailed confounders

The study collected data on detailed behavioral, psychosocial, and environmental confounders, which are shown in Table 1 along with their assessment tools. All were taken directly or derived from self-reported information collected in the first trimester. Standardized assessment tools were used to assess diet quality,^12,13 nausea and vomiting,¹⁴ health literacy,¹⁵ depressive symptoms,¹⁶ perceived stress,¹⁷ anxiety,¹⁸ and resilience.¹⁹ US-Census-based neighborhood and built environment variables, which were based on the participant’s census tract and block group at enrollment, were neighborhood walkability,²⁰ neighborhood deprivation,²¹ and percent of neighborhood with income below the poverty line.

Table 1.

Detailed lifestyle, psychosocial, and environmental confounders collected in the Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be cohort (2010‒2013), n=8978.

Domain	Measurement a	Completeness (n (%) of participants with available data)
Periconceptional diet quality	Healthy Eating Index 2015, calculated based on data from a validated semi-quantitative food frequency questionnaire	7608 (85)
First-trimester physical activity	Total physical activity  <450 MET hours/week  450‒899 MET hours/week  ≥900 MET hours/week	8974 (99)
Severity of first-trimester nausea and vomiting	Pregnancy-Unique Quantification of Emesis and Nausea (PUQE) index  0 to 3  4  5  6 or more	8978 (100)
Preconception binge drinking	Any single occasion of drinking 4 or more drinks during the 3 months before conception  None  Any	8901 (99)
Pregnancy planning	Pregnancy was planned  No  Yes	8975 (100)
Weight gain goal	First trimester pregnancy weight gain goal  Had a personal goal and a provider goal  Had a personal goal but no provider goal  Had a provider goal only	8067 (90)
Health Literacy	Rapid Estimate of Adult Literacy in Medicine short form  ≤ 6th grade reading level  7th to 8th grade reading level  9th grade or higher reading level	8655 (96)
Acculturation	Years residing in the United States and birthplace of parent(s)  Born in US and parent(s) born in the US  Born in US to one or more immigrant parents  Born outside US, immigrated to US aged 17 or younger with parents  Born outside US, immigrated to US aged 18 or older	8978 (100)
First-trimester depressive symptoms	Edinburgh Postnatal Depression Scale  Low depressive symptoms (0 to 10)  High depressive symptoms (11 or higher)	8735 (97)
First-trimester perceived stress	Perceived Stress Scale  Low stress (0 to 13)  Moderate stress (14 to 26)  High stress (27 to 40)	8967 (99)
First-trimester anxiety	State-Trait Anxiety Inventory  Low anxiety (20 to 39)  Moderate or high anxiety (40 to 80)	8020 (89)
First-trimester sleep satisfaction	Restless or average Restful or very restful	7960 (89)
Resilience	Connor-Davidson Resilience Score	8629 (96)
Residence in a food desert	Number of grocery stores within 3 km (tertiles)  0 to 8  9 to 44  45 or more	8697 (97)
Neighborhood walkability	US Walkability Index Tool	8697 (97)
Neighborhood deprivation	Area Deprivation Index	8697 (97)
Neighborhood poverty	Percent of neighborhood households living below the US federal poverty line	8697 (97)

^aReferences for standardized tools are: semi-quantitative food frequency questionnaire ¹², Healthy Eating Index – 2015 ¹³, Pregnancy-Unique Quantification of Emesis and Nausea (PUQE) index ¹⁴, Rapid Estimate of Adult Literacy in Medicine short form (REALM-SF) ¹⁵, Edinburgh Postnatal Depression Scale ¹⁶, Perceived Stress Scale ¹⁷, State-Trait Anxiety Inventory ¹⁸, Connor-Davidson Resilience Score ¹⁹, US Walkability Index Tool ²⁰, Area Deprivation Index ²¹.

Perinatal health outcomes

Medical record abstraction was used to ascertain outcome data. Gestational age was determined by applying a published algorithm.⁶ Preterm birth was defined as delivery of a liveborn or stillborn infant from 20+0 to 36+6 weeks’ gestation. Small- and large-for-gestational age births were classified using ultrasound-based intrauterine fetal weight standards at <10th percentile or >90th percentile, respectively ²². Unplanned Cesarean delivery was defined as a Cesarean delivery occurring after the onset of labor.

Statistical analysis

Balance of confounders across weight gain groups

We first examined the extent to which the distribution of confounders differed between pregnancy weight gain groups. We summarized the crude distribution of confounders within each weight gain group using means with standard deviations and frequencies with percentages. We used logistic regression to create a propensity score for our exposure of ‘weight gain below IOM recommendations’ (vs ‘weight gain within IOM recommendations’) using standard confounders only.²³ We re-examined the balance of confounders between weight gain groups after matching on propensity score (using kernel matching), and compared distributions using standardized differences. This enabled us to assess the extent to which imbalances in detailed confounders remained after balancing groups for standardly-measured confounders. We repeated this for the exposure of ‘weight gain above IOM recommendations’ (vs. weight gain within IOM recommendations).

Confounding of weight gain–perinatal outcome associations

We next examined the extent to which inclusion of detailed confounders altered the estimated association between pregnancy weight gain categories and adverse perinatal outcomes. We first built negative binomial regression to estimate the relationship between weight gain adequacy (high or low vs. adequate), controlling for standard confounders using our propensity score as a covariate (Model 1). For comparative purposes, we also estimated these association using a negative binomial regression model controlling for standard confounders as covariates rather than propensity scores. To Model 1, which controlled for standard confounders through propensity-score adjustment, we then added the detailed confounders as additional covariates, and re-estimated risk ratios (which we term the ‘fully adjusted’ risk ratio; Model 2). We defined the risk ratio due to confounding as the ratio of the standard confounder-adjusted risk ratio to the fully adjusted risk ratio and used bootstrapping to calculate 95% confidence intervals for this measure. We used multiple imputation (multiple imputation by chained equations, imputing 10 datasets using all available covariates and perinatal health outcomes) to address missing data in detailed confounders. For preterm birth models, we used a Cox proportional hazards model rather than negative binomial regression, and used the hazard ratio to approximate the risk ratio in the calculation of the risk ratio due to confounding.

Results

Among 9557 participants whose pregnancy progressed to at least 20 weeks, we excluded those with missing data on weight or BMI (n=535, 5.6%), standard covariates (n=6 <0.1%), or outcomes (n=38, 0.4%), leaving 8978 for analysis. Data were available for ≥85% of each of the enhanced covariates, but only 67% of women (n=6062) had complete information for all covariates (Table 1). Patterns of missing data did not vary by weight gain status (eTable 1).

There were more individuals who gained above the IOM guidelines (62%) than who gained within (25%) or below (13%). The prevalence of low for gestational age (LGA) was 5.6%, small for gestational age (SGA) 12%, preterm birth 8.4%, and unplanned cesarean delivery 21%.

Balance of confounders across pregnancy weight gain groups

Before propensity score adjustment, participants with inadequate pregnancy weight gain were more likely to be younger, unmarried, smoke during pregnancy, have higher BMI, be of Hispanic or Non-Hispanic Black race/ethnicity, and be covered by Medicaid insurance (Table 2). There were other important differences in the detailed confounders among people with low weight gain, including having lower diet quality, more nausea and vomiting, preconception binge drinking, lower literacy, and anxiety, and living in neighborhoods with poorer socioeconomic conditions.

Table 2.

Balance in standard and detailed confounders before and after matching on propensity score for low pregnancy weight gain in the Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be cohort (2010‒2013), n=8978.

	Before propensity score matching			After propensity score matching
Variable	Mean		% bias (standardized differences)	Mean		% bias (standardized differences)
	Low weight gain a	Adequate weight gain		Low weight gain	Adequate weight gain
Standard confounders
Maternal age, years	26	27	−24b	26	26	−1.2
Prepregnancy body mass index, kg/m2	25	24	19b	25	25	2.8
Smoker, %	16	13	8.4	16	15	−2.3
Married, %	56	67	−25b	56	57	−2.3
Public insurance, %	41	26	33b	41	40	2.4
Some college education, %	30	26	9.6	30	29	2.0
Bachelor’s degree, %	23	31	−18b	23	24	−1.4
Graduate degree, %	21	27	−14b	21	22	−2.3
Gravidity 1, %	16	19	−7.3	16	16	1.7
Gravidity 2 or more, %	6.1	5.5	−7.3	6.1	6.0	−0.5
Non-Hispanic Black, %	18	12	15b	18	18	−0.1
Hispanic, %	22	16	15b	22	21	−0.1
Other race, %	10	10	−0.6	10	10	0
Used assisted reproductive technology, %	3.0	4.2	−6.4	3.0	3.1	−0.2
Prepregnancy diabetes, %	1.4	0.9	4.5	1.4	1.4	0.7
Chronic hypertension, %	2.3	2.2	0.5	2.3	2.4	−0.9
Detailed confounders
Healthy Eating Index – 2010 score	64	66	−14b	64	64	1.9
Physical activity 450–899 MET hr/wk, %	18	19	−3.0	18	18	0.1
Physical activity ≥ 900 MET hr/wk, %	27	30	−6.0	27	27	0.7
PUQE nausea and vomiting score = 4, %	14	16	−5.5	14	16	−4.6
PUQE nausea and vomiting score = 5, %	14	13	3.5	14	13	3.9
PUQE nausea and vomiting score ≥ 6, %	31	24	16b	31	26	13b
Any preconception binge drinking, %	30	36	−13b	30	33	−7.0
Pregnancy was planned, %	56	64	−18b	56	56	0.1
Had a personal weight gain goal only, %	28	31	−7.9	28	30	−3.9
Had a provider weight gain goal only, %	8.8	8.5	1.3	8.8	8.8	0.1
Had no weight gain goal, %	42	34	17b	42	37	10b
Health Literacy 7th to 8th grade, %	17	12	15b	17	16	3.4
Health Literacy 9th grade or higher, %	79	86	−17b	79	80	−2.8
Born in US/ immigrant parents, %	14	14	0.2	14	15	−2.5
Immigrated to US age ≤17 years, %	10	8.7	6.0	10	10	2.3
Immigrated to US age ≥18 years, %	8.6	7.4	4.5	8.6	7.2	5.2
High depressive symptoms, %	15	11	12b	15	13	6.0
Moderate stress, %	39	35	9.7	39	38	2.7
High stress, %	4.1	2.8	7.1	4.1	3.9	1.0
Moderate or high anxiety, %	26	20	14b	26	23	8.2
Restful or very restful sleep, %	38	43	12b	38	42	−10b
Resilience score	78	80	−10	78	79	−5.9
9 to 44 grocery stores within 3 km, %	37	34	7.3	37	36	3.7
45 or more grocery stores within 3 km, %	35	36	−6.0	35	36	−3.7
Neighborhood walkability score	14	13	2.7	14	13	3.8
Area deprivation index	50	42	25b	50	48	5.1
Percent households below poverty line	19	17	18b	19	19	1.5

PUQE, Pregnancy-Unique Quantification of Emesis and Nausea

^aLow weight gain, n=1194; Adequate weight gain, n=2238.

^bIndicates an absolute standardized difference of at least 10%.

As expected, propensity score matching reduced the imbalances between groups in the standard confounders (all standardized differences <10%). Imbalances between groups in the detailed confounders (which were not included in the creation of the propensity score) were also greatly diminished: all standardized differences decreased to <10%, with the exception of high nausea and vomiting (Pregnancy-Unique Quantification of Emesis and Nausea [PUQE] Score) ≥6 in 31% of low weight gain group vs 26% in adequate weight gain group), having no weight gain goal (in 42% of low vs 37% in adequate weight gain group), and experiencing restful or very restful sleep (in 38% of low vs 42% in adequate weight gain group). Of note, propensity score adjustment for standard covariates produced groups that were also balanced with respect to diet quality and physical activity. There was good overlap in propensity scores across exposure groups; no participants were excluded due to lack of common support.

Crude differences between women with excessive vs adequate weight gain were generally similar in direction but smaller in magnitude compared with results for inadequate weight gain (eTable 2). After balancing groups through propensity score matching, all differences were <10%.

As expected, participants with pregnancy weight gain below recommendations were more likely to have a SGA birth, and less likely to have a LGA birth or an unplanned Cesarean delivery (Table 3). Adjustment for standard confounders (through either propensity scores or covariate adjustment) had only a small effect on estimated associations (e.g., an attenuation of the point estimate for SGA birth from 1.6 to 1.5). Further adjustment for detailed confounders had minimal impact on the estimated associations, as evidence by null risk ratios of confounding for all outcomes (e.g., 1.1 [95% CI: 1.0, 1.1] for low weight gain and preterm birth). We observed similar patterns when examining risks associated with weight gain above vs within recommendations (Table 4).

Table 3.

Influence of adjustment for detailed lifestyle, psychosocial, and environmental factors on estimated associations between low pregnancy weight gain and adverse perinatal outcomes in the Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be cohort (2010‒2013), n=8978.

	Risk ratios (95% CI) for low vs. adequate pregnancy weight gain
Outcome	Unadjusted	Covariate adjustment for standard confounders a	Propensity score adjustment for standard confounders (Model 1)	Model 1 + detailed confounders b as additional covariates (Model 2)	Risk ratio due to confounding c
Large-for-gestational age birth	0.62 (0.38, 0.99)	0.53 (0.33, 0.86)	0.55 (0.34, 0.89)	0.56 (0.34, 0.92)	0.98 (0.89,1.1)
Small-for-gestational age birth	1.6 (1.4, 1.8)	1.5 (1.3, 1.7)	1.5 (1.3, 1.7)	1.5 (1.2, 1.8)	1.1 (1.0, 1.1)
Preterm birthd	1.1 (1.0, 1.2)	1.1 (1.0, 1.2)	1.1 (1.0, 1.2)	1.1 (1.0, 1.2)	1.1 (1.0, 1.1)
Unplanned cesarean	0.85 (0.72, 1.0)	0.79 (0.67, 0.94)	0.79 (0.67, 0.94)	0.80 (0.65, 0.98)	0.99 (0.95, 1.0)

^aStandard confounders are participant’s age, education, race–ethnicity, marital status, pre-pregnancy smoking, medical insurance, gravidity, preexisting diabetes, chronic hypertension, and use of assisted reproductive technologies.

^bDetailed confounders are diet quality, nausea and vomiting, health literacy, depressive symptoms, perceived stress, anxiety, resilience, neighborhood walkability, neighborhood deprivation, proximity to grocery stores, and percent of neighborhood with income below the poverty line.

^cRisk ratio due to confounding is defined as the ratio of risk ratio from the model adjusting for propensity score for standard confounders to the risk ratio from the model adjusting for detailed confounders and propensity score for standard confounders.

^dValues presented are hazard ratios rather than risk ratios.

Table 4.

Influence of adjustment for detailed lifestyle, psychosocial, and environmental factors on estimated associations between high pregnancy weight gain and adverse perinatal outcomes in the Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be cohort (2010‒2013), n=8978.

	Risk ratios (95% CI) for high vs. adequate pregnancy weight gain
Outcome	Unadjusted	Covariate adjustment for standard confounders a	Propensity score adjustment for standard confounders (Model 1)	Model 1 + detailed confounders b as additional covariates (Model 2)	Risk ratio due to confounding c
Large-for-gestational age birth	2.5 (1.9, 3.2)	2.2 (1.7, 2.8)	2.1 (1.6, 2.7)	2.1 (1.6, 2.7)	1.0 (0.97, 1.0)
Small-for-gestational age birth	0.65 (0.57, 0.75)	0.66 (0.57, 0.75)	0.66 (0.58, 0.76)	0.65 (0.57, 0.75)	1.0 (0.98, 1.0)
Preterm birth d	0.98 (0.93, 1.0)	0.97 (0.92, 1.0)	0.97 (0.93, 1.0)	0.97 (0.92, 1.0)	1.0 (0.99, 1.0)
Unplanned Cesarean delivery	1.5 (1.3, 1.6)	1.3 (1.2, 1.5)	1.3 (1.2, 1.4)	1.3 (1.2, 1.5)	0.98 (0.96, 1.0)

^aStandard confounders are participant’s age, education, race–ethnicity, marital status, pre-pregnancy smoking, medical insurance, gravidity, preexisting diabetes, chronic hypertension, and use of assisted reproductive technologies.

^bDetailed confounders are diet quality, nausea and vomiting, health literacy, depressive symptoms, perceived stress, anxiety, resilience, neighborhood walkability, neighborhood deprivation, proximity to grocery stores, and percent of neighborhood with income below the poverty line.

^cRisk ratio due to confounding is defined as the ratio of risk ratio from the model adjusting for propensity score for standard confounders to the risk ratio from the model adjusting for detailed confounders and propensity score for standard confounders.

^dValues presented are hazard ratios rather than risk ratios.

Discussion

In this large, diverse cohort of US nulliparous women, we found that the adjustment of detailed information on behavioral, psychosocial, and environmental confounders along with routinely available confounders had minimal impact on estimated associations between pregnancy weight gain and adverse pregnancy outcomes. Initial imbalances between weight gain groups in factors such as diet, exercise, sleep, and mental health were largely eliminated by matching on a propensity score built only from routinely available confounders. Not surprisingly, the estimated associations between pregnancy weight gain and adverse outcomes did not meaningfully change when these more detailed confounders were then added to regression models. As a result, the relative risk due to confounding was null for all outcomes studied, suggesting minimal confounding bias as a result of unmeasured data on covariates that we had access to in our cohort.

Despite concern raised about the potential for unmeasured confounding,^1–5 few previous studies of pregnancy weight gain and adverse perinatal outcomes, to our knowledge, have adjusted for detailed confounders. Investigators have reported little to no confounding by physical activity or diet,^24–27 but confounding by other detailed covariates has not been well explored. Our finding that a propensity score with standard confounders also reduced standardized differences in the distribution of detailed confounders between groups suggests that there is some correlation among these variables. Such correlation would lessen the impact of additional control for detailed confounders. Somewhat surprisingly, we found that controlling for standard confounders also had little impact on unadjusted effect estimates. This finding nevertheless aligns with our experience with weight gain research in other cohorts,^5,28–32 and is consistent with other published work.^33,34

In large epidemiologic studies where collection of a broad range of covariates is essential to the research objectives, constructs such as diet, physical activity, psychosocial factors, and the built environment can be challenging to measure with complete accuracy. Gold standard assessment methods may not exist or may be impractical to include in this setting. In our study, there is likely residual confounding in our estimates due to imperfect measurement of complex constructs. Nevertheless, this study’s assessment methods represent what can be feasibly measured and collected by epidemiologists in large cohort studies. We had at least one measure from a broad range of domains identified by the 2009 IOM Committee as predictors of weight gain (diet, exercise, sleep, mental health, socioeconomic status, health literacy, resilience, acculturation, and built environment),¹ but we lacked data on other predictors such as natural or human-created disasters, genetics, or media messaging. Thus, we cannot rule out the impact of these and other unmeasured variables on confounding associations between pregnancy weight gain and adverse health outcomes.

The nuMoM2b study enrolled a racially and ethnically diverse sample from eight US centers, but we sacrificed some generalizability in using this group because it included only nulliparous women. Associations among pregnancy weight gain, adverse perinatal outcomes, and their common causes as well as accuracy in their measurement may differ in other cohorts. Exploration is warranted to evaluate whether our results generalize to other settings.

The pregnancy weight gain z-score charts we applied to our cohort were derived in a Pennsylvania sample. While the extent to which a Pennsylvania-based chart is generalizable to pregnant individuals from eight sites across the US is unknown, previous research has found that the estimated associations between weight gain z-scores and preterm birth in a Pennsylvania-based cohort were virtually identical to those when the chart was applied to a California-based cohort with a different racial–ethnic composition.³⁵ This supports the generalizability of this chart in other, diverse US populations.

Self-reported pre-pregnancy weight may be misreported.³⁶ However, we have shown in previous studies that misreporting did not explain associations between pregnancy weight gain and adverse perinatal outcomes.^3,4,37 Complete data on detailed confounders were only available for 67% of participants. We used multiple imputation to retain individuals with missing values and found no differences in the degree of completeness in data according to pregnancy weight gain. However, we cannot rule out the potential for selection bias. The missing values highlight an important design consideration: the collection of detailed confounder information may come with higher missingness rates due to their increased participant burden. Appropriate strategies to understand the nature of the missingness and account for it analytically are needed.

In this study, we found that the inclusion of detailed information on behavioral, psychosocial, and environmental factors had minimal impact on our understanding of the link between pregnancy weight gain and adverse pregnancy outcomes. Our findings are compatible with the conclusion that information on these detailed confounders may not be critical for obtaining unbiased estimates of the consequences of pregnancy weight gain for perinatal health. However, many of these factors are modifiable and warrant exploration as risk factors and targets for future interventions to reduce adverse perinatal health outcomes. Investigators designing large pregnancy cohort studies should carefully consider the goals of their research in balancing the yield from collecting these detailed data with the participant and researcher burden.