Exercise During the First Trimester of Pregnancy and the Risks of Abnormal Screening and Gestational Diabetes Mellitus

The Physical Activity (PA) Guidelines for Americans (1), the American College of Obstetricians and Gynecologists (ACOG) (2,3), and the American College of Sports Medicine (4) recommend that pregnant women participate in at least 150 min of moderate-intensity activity per week. The 2020 ACOG committee opinion on PA and exercise during pregnancy states that in the absence of obstetric or medical complications or contraindications, PA during pregnancy is safe and desirable, and pregnant women should be encouraged to continue or initiate safe PA (2). The committee opinion highlights decreased risk of gestational diabetes mellitus (GDM) as a benefit of exercise during pregnancy.

A 2018 systematic review and meta-analysis found that exercise-only interventions reduced the odds of GDM by 38% compared with no intervention (5), although most of these exercise-only interventions included did not reduce the risk of GDM on their own (5). It is important to supplement data obtained from randomized controlled trials of exercise interventions conducted in motivated volunteers with the findings of population-based cohort studies assessing free-living PA because the later are the basis for PA recommendations. To further improve upon recommendations for exercise in pregnancy, there is the need to better define the minimum threshold for volume of exercise needed to improve health outcomes since volume incorporates the exercise metrics of intensity, duration, and frequency into a single value. Studies examining such thresholds are needed to inform the development of lifestyle interventions to prevent GDM.

This prospective cohort study examined the association of exercise during the first trimester of pregnancy with abnormal glucose values obtained from 50-g, 1-h screening for GDM and for GDM according to Carpenter and Coustan criteria. We additionally sought, a priori, to investigate potential differences by prepregnancy weight status (5–7) and to use contemporary causal inference methods to address this question (8–11).

The study setting was the Kaiser Permanente Northern California (KPNC) health care delivery system, which serves 3.6 million members. The KPNC population is representative of the geographic area served and is racially/ethnically and socioeconomically diverse (12). Data for the current study come from the Pregnancy Environment and Lifestyle Study (PETALS) birth cohort study. Beginning in 2013, PETALS recruited KPNC women initiating prenatal care at <11 gestational weeks; the participation rate was 75% (13).

At 10–13 gestational weeks, participants provided consent and had height and weight measured. The study survey was the Pregnancy Physical Activity Questionnaire (PPAQ), which participants completed on their own. The PETALS PPAQ (14) was slightly modified from the original (15) and updated to reflect technological advances (e.g., watching TV or a video was updated to watching TV, a movie, or video clip). The PPAQ asked women to report time spent in 36 population-appropriate activities during the previous 2 months. Response options included ranges for the amount of time spent in each activity (e.g., none, <1/2 h per day, 1/2 to almost 1 h per day, 1 to almost 2 h per day, 2 to almost 3 h per day, or ≥3 h per day). To provide a conservative estimate, the minimum value of the range selected (duration and frequency of activity) were multiplied by the intensity, expressed in METs, to derive an estimate of the volume of PA (MET-hours per day). Compendium-based MET values were used for activities other than walking and household tasks, which came from field-based measurements of pregnant women (15).

The focus of the current study is the sports and exercise activity domain: PA that is intentional for health and wellness or to increase fitness and results in energy expenditure beyond the demands of everyday living. This domain includes 10 PPAQ activities of moderate intensity (range 3.2–6 METs for walking, swimming, etc.) plus 2 PPAQ activities of vigorous intensity (6.5 and 7 METs for walking quickly up hills and jogging, respectively) performed for fun or exercise (14). It includes an item for weight lifting, resistance exercise (i.e., of moderate intensity), in addition to items assessing for activities that combine resistance and cardiovascular exercises (e.g., swimming). The sum of the volume (MET-hours per day) of all activities in this domain provided an overall estimate of volume of moderate- to vigorous-intensity sports and exercise activity (referred to as exercise hereafter). Cronbach α was 0.75 for the PETALS PPAQ overall (36 items) and 0.70 for the sports and exercise domain (12 items).

To inform public health recommendations pertaining to PA during pregnancy, the exposure was defined by several thresholds. Since PA questionnaire data are self-report and subject to bias, they are most appropriate for ranking individuals with regard to volume of activity. Thus a high volume of exercise was first defined as meeting or exceeding the cohort-specific 75th percentile (i.e., volume of sports and exercise activity ≥13.2 MET-hours per week) (16). Exercise was then defined as meeting or exceeding the lower bound of the current PA guidelines (1,2) recommending 150–300 min per week of moderate exercise, 75–150 min per week of vigorous exercise, or the equivalent volume of moderate/vigorous exercises combined (i.e., ≥7.5 MET-hours per week) (1,16,17). Finally, although vigorous exercise is believed to be safe for most healthy pregnant women, particularly those who participated in vigorous exercise prior to pregnancy, the 2020 ACOG committee opinion expressed the need for more data (2). As such, performing any vigorous exercise was examined.

The study survey also provided information on age, race/ethnicity, marital status, parity, education, and exercise during the year prior to pregnancy (6,7). It included a Block food frequency questionnaire (18,19) assessing dietary intake between conception and the study visit, specifically daily caloric intake in kcal. Data on gestational age were obtained from the electronic health record. Clinic-measured prepregnancy weight (i.e., within 1 year of the last menstrual period date) was available in the electronic health record for the majority of the cohort; otherwise, a clinic-measured weight before 10 weeks gestation from the electronic health record or self-reported prepregnancy weight from the study survey was used. BMI was calculated as weight in kilograms divided by height in meters squared and classified according to standard thresholds (20). Prepregnancy weight was subtracted from weight measured at the study visit (i.e., by PETALS research staff) to isolate gestational weight gain through the study visit (21) since weight gain occurring after the exercise exposure could lie on the causal pathway between exercise and the development of abnormal screening and GDM.

Results of all blood glucose testing were obtained from the KPNC Gestational Diabetes and Pregnancy Glucose Tolerance Registry (22); all plasma glucose measurements in this setting are performed using the hexokinase method at the KPNC regional laboratory. Women with recognized pre-GDM were identified in the KPNC Diabetes Registry (23) and ineligible to participate in PETALS. The KPNC Diabetes Registry identifies patients from four data sources: primary hospital discharge diagnoses of diabetes, two or more outpatient visit diagnoses of diabetes, any prescription for a diabetes-related medication, or any record of an abnormal HbA_1c test (>50 mmol/mol [6.7%] [23]).

The outcomes—abnormal screening and GDM—were defined according to the screening and diagnosis protocol used in this clinical setting. At KPNC, the two-step approach is used to identify GDM: At 24–28 weeks gestation, women are screened with a 50-g, 1-h glucose challenge test (>95% of pregnancies screened), and those with glucose ≥140 mg/dL (7.8 mmol/L) (referred to as abnormal screening hereafter) go on to a diagnostic 100-g, 3-h oral glucose tolerance test (OGTT). For the current study, only women with two or more glucose values on the OGTT meeting or exceeding thresholds proposed by Carpenter and Coustan (24) are considered to have GDM (fasting: 95 mg/dL [5.3 mmol/L]; 1 h: 180 mg/dL [10.1 mmol/L]; 2 h: 155 mg/dL [8.7 mmol/L]; 3 h: 140 mg/dL [7.8 mmol/L]).

PETALS participants with PPAQ data who delivered between October 2013 and October 2017 were eligible for the current study (N = 2,501). Those with a contraindication to PA during pregnancy (3) who were diagnosed before the study visit were excluded (n = 6) as were 72 women with implausible volumes of PA (25,26). Participants missing blood glucose data (n = 177) were then excluded (e.g., not screened or abnormal screening but no follow-up OGTT). The final analytic cohort consisted of 2,246 women.

Causal inference methods were used to obtain estimates of the average difference in risk of an abnormal screening test result and GDM. Under the assumptions discussed below, all estimates can be interpreted as the difference in risk had all women exercised above one of the three activity levels considered minus the risk had none exercised above the same level during the first trimester. Adjusted risk differences were estimated by four strategies: inverse probability of treatment weighting (IPTW), then targeted maximum likelihood estimation (TMLE) (8–11) with user-defined parametric models, and finally, TMLE implemented with two sets of machine learning algorithms (i.e., the defaults and an expanded set) to potentially improve upon the robustness and precision of the estimates.

For the IPTW analyses, the propensity for exercise was estimated by logistic regression and included the following baseline covariates (i.e., adjusted for): age (continuous), prepregnancy BMI (continuous), daily caloric intake (kcal) (continuous), marital status (married [reference], not married), race/ethnicity (Hispanic [largest group, reference], White, Asian/Pacific Islander, African American, and other), parity (0 [reference], 1, 2+), education (high school or less, some college [reference], college graduate, postgraduate), and exercise during the year prior to pregnancy (did not exercise, did exercise [reference]).

Unadjusted (i.e., unweighted) and adjusted IPTW (with stabilized, untruncated weights) estimates were obtained using Proc Genmod with an independence structure for the variance-covariance matrix (SAS 9.4 software). Stabilized IPTW weights were not truncated because all maximums were <13.5, except for one subgroup analysis with a maximum IPTW weight of 36.5 (27) (Supplementary Table 1).

R 4.0.2 software was used to implement TMLE, a doubly robust, locally efficient estimator (8–10). Valid inference from TMLE depends on correctly estimating the model for the propensity score or the outcome regression. To avoid incorrect TMLE inference from misspecified parametric models (e.g., logistic) for propensity score and outcome regression, data-adaptive machine learning methods, such as Super Learning (28), have been proposed to estimate nuisance parameters (10,29–32). Three approaches were used to estimate the propensity score and the outcome regression portions of the likelihood: 1) the same logistic regression modeling approach previously described, 2) adaptively with Super Learning (28) using the default “learners” (8,9) (i.e., prediction algorithms) only and the adjustment variables previously described, and 3) adaptively with Super Learning, with the default learners plus six extra learners and the adjustment variables previously described. The default learners (8,9) included Wrapper for Glm, Choose a Model by AIC in a Stepwise Algorithm, and Wrapper Function for SuperLearner Prediction Algorithm, and the extra learners included SL Wrapper for Biglasso, Elastic Net Regression Including Lasso and Ridge, Wrapper for Kernlab’s SVM Algorithm, Wrapper for Lm, SL Wrapper for Ranger, and Wrapper for Speedglm.

For both adjusted IPTW and TMLE, missing covariate data were addressed using the missingness indicator method (33,34) (i.e., an indicator of missing values was included in the adjustment set, and missing values were replaced with the median of observed values). Baseline categorical covariates with missing values included marital status (n = 4), parity (n = 4), and education (n = 2). Those missing caloric intake (n = 52) or with implausible estimates of daily caloric intake (i.e., <400 cal or >6,000 cal, n = 19) were assigned the cohort median value of 1,448.43 cal; women missing gestational weight gained through study visit (n = 4) were assigned the cohort median 1.6 lb.

Analyses stratified by prepregnancy weight status (underweight/healthy weight vs. overweight/obese) estimated subgroup-specific associations (6). Meeting or exceeding the time point–specific thresholds of Carpenter and Coustan on the 100-g, 3-h OGTT (24) were also examined as outcomes. The threshold for statistical significance was 0.05 for all analyses. The study was approved by the KPNC institutional review board.

Characteristics of the analytic cohort (n = 2,246) are presented in Table 1. The mean age was 30.2 (SD 5.3) years, and 56% (n = 1,272) of the women were overweight or obese as determined by a KPNC clinic-measured prepregnancy weight (i.e., within 1 year of the last menstrual period) for 78% (n = 1,753) of the cohort, a KPNC clinic-measured weight from early pregnancy (i.e., before 10 weeks gestation) for 19.4% (n = 436), and self-reported prepregnancy weight on the study survey for the remaining 2.5% (n = 57). The Pearson correlation coefficient was 0.99 for clinic-measured prepregnancy weight and clinic-measured early pregnancy weight (n = 1,687; P < 0.001), 0.98 for clinic-measured early pregnancy weight and self-reported prepregnancy weight (n = 2,122; P < 0.001), and 0.98 for self-reported prepregnancy weight and clinic-measured prepregnancy weight (n = 1,752; P < 0.001). Women excluded from the current study were less likely to be married, more often multiparous, and attained lower levels of education compared with those included in the analytic cohort (Supplementary Table 2).

Of those included in the current study, 913 (40.7%) women met or exceeded the lower bound of the PA guidelines, and 829 (36.9%) reported any amount of vigorous-intensity exercise; 18.3% reported weight lifting or resistance exercises. Compared with women who were below the cohort-specific 75th percentile for exercise, those who met or exceeded this threshold were more likely to have exercised prior to pregnancy, to be White, and to have attained a college or postgraduate education. Overall, mean gestational age at the PPAQ was 12.8 (SD 2.5) weeks, and mean gestational age at the 50-g, 1-h glucose challenge test was 24.4 (SD 5.7) weeks. The frequency of abnormal screening and GDM were 24.3% and 6.5%, respectively (Table 1).

Table 2 presents the exposures, outcomes, and key confounders (e.g., prepregnancy exercise, gestational weight gain through the study visit) stratified by BMI category. The prevalence of abnormal screening and GDM increased with increasing BMI category, and women with overweight and obesity had lower gestational weight gain than underweight and healthy weight women.

Table 3 presents estimates of the causal risk differences for an abnormal screening test and GDM if all women had exercised at or above the cohort-specific 75th percentile for moderate- to vigorous-intensity exercise, met the PA guidelines, and participated in any vigorous-intensity exercise versus not. All estimates for exercise at or above the cohort-specific 75th percentile indicated a reduction in the risk of abnormal screening. The estimate from data-adaptive TMLE with extra learners revealed 4.8 (95% CI 1.1, 8.5) fewer abnormal screening tests per 100 women had all women exercised at or above the cohort-specific 75th percentile versus not. Exercise at this level was also associated with a reduced risk of GDM in the unadjusted IPTW, data-adaptive TMLE with defaults, and data-adaptive TMLE with extra learners (2.6 [0.2, 4.9], 2.0 [0.2, 4.2], and 2.1 [0.2, 4.1] fewer cases of GDM per 100 women, trespectively). There was the suggestion of a reduced risk of GDM for exercise meeting or exceeding the lower bound of the PA guidelines in the unadjusted IPTW analysis (2.0 [−0.1, 4.1] fewer cases of GDM per 100 women; P = 0.06). Any vigorous-intensity exercise reduced the risk of GDM by 2.5 (0.4, 4.7) per 100 women with unadjusted IPTW.

Table 4 displays estimates of the causal risk differences for abnormal screening and GDM stratified by prepregnancy weight status (i.e., underweight or healthy weight vs. overweight or obesity). Among underweight or healthy weight women (n = 973), 197 (20.3%) had abnormal screening and 42 (4.3%) had GDM, and all risk difference estimates for abnormal screening attained statistical significance in this subgroup. Data-adaptive TMLE with extra learners indicated that 7.1 (95% CI 1.5, 12.7) fewer underweight or healthy weight women per 100 would have an abnormal screening test had all exercised at or above the 75th percentile versus not. In women with overweight or obesity (n = 1,272), 348 (27.3%) had abnormal screening and 105 (8.3%) had GDM, and all adjusted estimates for GDM attained statistical significance. If all the women with overweight or obesity had exercised at or above the 75th percentile, there would be 2.9 (0.3, 5.4) fewer cases of GDM according to data-adaptive TMLE with extra learners.

Estimates of the causal risk differences for meeting or exceeding the time point–specific diagnostic thresholds of Carpenter and Coustan for the 100-g, 3-h OGTT if all women had exercised at or above the 75th percentile (13.2 MET-hours per week), versus not, are presented in Supplementary Table 3. Estimates for the 2-h threshold attained significance in several adjusted analyses; data-adaptive TMLE with extra learners indicated that 2.5 (95% CI 0.3, 4.7) fewer women per 100 would meet the 2-h threshold had all exercised at or above the 75th percentile versus not. There was a decreased risk of meeting the fasting threshold for exercise at or above the 75th percentile using adjusted IPTW (1.6 [−0.3, −3.0] fewer per 100 women). No associations were observed for the 1-h and 3-h thresholds.

The results of this study suggest that meeting current recommendations for exercise during the first trimester of pregnancy does not confer reductions in the risks of abnormal screening and GDM. However, exercise at or above the cohort-specific 75th percentile, a higher minimum threshold, was found to reduce the risks of abnormal screening and GDM by 4.8 and 2.1 per 100 women, respectively. Although a meta-analysis of the results of randomized controlled trials of exercise-only interventions demonstrated a reduced odds of GDM, most of the individual studies of exercise-only interventions did not reduce GDM when evaluated on their own (5). Indeed, randomized controlled trials of pregnancy lifestyle interventions of diet and exercise have largely been unsuccessful in preventing GDM (35). The results of this study suggest that future lifestyle interventions among pregnant women who are free of complications and contraindications to PA (3) and monitored by an obstetric care provider (2) should prescribe a volume of exercise that is higher than the currently recommended minimum in order to reduce GDM risk.

ACOG’s (2,3) recommendation for at least 150 min per week of moderate-intensity PA (i.e., ≥7.5 MET-hours per week [1,16,17]) is based on the PA guideline recommendations for the general American population (1), which prescribe either moderate-intensity exercise, vigorous-intensity exercise, or a comparable combination of moderate- and vigorous-intensity exercises. Moderate-intensity exercise encompasses the 3–6-MET range, and the lowest value (i.e., 3 METs) is typically used to construct the recommendation; moderate-intensity exercise is also recommended for pregnant women (2). The cohort-specific 75th percentile examined in the current study was 13.2 MET-hours per week and would be achieved through ≥264 min per week of moderate-intensity exercise (i.e., at 3 METs) or alternatively stated, ≥38 min per day of moderate-intensity exercise. Since walking is the mostly commonly reported exercise among pregnant women in the U.S. (36,37), future lifestyle interventions should consider prescribing this amount of moderate-intensity walking to reduce the risk of GDM.

Davenport et al. (5) recommended at least 10 MET-hours per week of moderate-intensity exercise to achieve at least a 25% reduction in the odds of GDM, gestational hypertension, and preeclampsia. Their meta-analysis included 6,934 women from 26 randomized controlled trials of exercise-only interventions and found that among motivated trial volunteers, exercise-only interventions reduced the odds of developing GDM by 38% compared with no intervention (5). The current study found that 2.1 cases of GDM per 100 women would be prevented with exercise compared with no exercise. The prevalence of GDM was 6.5 per 100 women; thus, exercise at or above 13.2 MET-hours per week reduced the risk of GDM by 32%, similar to that observed in the meta-analysis. It is worth noting that our study examined free-living or natural exercise, a related but distinct exposure from the exercise interventions examined in randomized controlled trials, which often include elements that providers cannot universally recommend and that are not affordable for all women (e.g., the use of exercise equipment such as a treadmill or stationary bicycle, personalized training sessions).

Valid causal inference with the methods used by the current study requires several assumptions. The first assumption is the positivity or experimental treatment assignment assumption, which posits that all manifestations of the exposure must be possible conditional on the baseline covariates. The few women with contraindications to exercise during early pregnancy were thus excluded from the current study (n = 6), and for women without contraindications, it is reasonable to assume that any level of exercise would be possible. A second, untestable assumption is that of no unmeasured confounding. The current study had high-quality data on several key confounding factors, but the strong possibility for residual confounding remains, as is the case for any observational study of a health behavior. Valid IPTW inference assumes consistent estimation of the propensity score (i.e., the logistic model used in this study being correctly specified). Valid TMLE inference relies on consistent estimation of either the propensity score or the outcome regression; the current study used machine learning (i.e., SuperLearner [31]) in the TMLE analyses to avoid violating this assumption.

Strengths of the current study include the prospective design, the size and racial/ethnic diversity of the cohort, and the availability of measured prepregnancy weight for most women and objective glucoses measurements to define the outcomes. There are several limitations worth noting. Women with implausible exposure data or missing outcome data were excluded from analyses, which could introduce bias. Since excluded women were more often unmarried, multiparous, and attained lower levels of education, results may not be generalizable to these subgroups. Fortunately, in the current study, only 10% of eligible women were excluded. PETALS also had a 75% participation rate overall (13), suggesting that these data are representative of the underlying source population. The PA exposure data were self-reported and subject to bias. As such, the minimum value for the range of duration and frequency of activity reported was used to mitigate the impact of overreporting and to provide a conservative estimate of the volume of PA. The current study used abnormal screening defined as ≥140 mg/dL and GDM by the Carpenter and Coustan criteria only; thus, findings may not be generalizable to practices that use different screening thresholds or testing procedures and diagnostic criteria for GDM, including the International Association of Diabetes and Pregnancy Study Groups criteria (38).

With the proliferation of commercial PA monitors in recent years, there is great potential for future research on the association of objectively measured PA during pregnancy with abnormal screening and GDM. However, device-based measures of PA cannot distinguish between movement that is intentional for health and wellness and that which is occupational in nature, and positive health effects may be limited to the former. It also remains to be determined which specific objective metric of PA (i.e., steps vs. minutes of moderate- to vigorous-intensity PA) should be used to formulate device-based recommendations for PA in the future. Although current recommendations use minutes of moderate- to vigorous-intensity aerobic activity (on the basis of self-report), device-based measures of steps are more accurately estimated than device-based minutes of moderate- to vigorous-intensity PA, and there is less interdevice variability in the estimation of steps (39,40). However, recommendations that are based on minutes of PA do not require the use of a PA monitoring device. Future studies, both randomized controlled trials and observational cohort studies of pregnant women, should include both device-based measures and validated questionnaires.

In conclusion, the results suggest that exercise during the first trimester of pregnancy is an effective tool for reducing the risks of abnormal screening and GDM. However, more exercise than is currently recommended may be needed to achieve reductions in the risks of these outcomes. Pregnancy lifestyle interventions should encourage, among women who are free of complications and contraindications to PA and monitored by an obstetric care provider, at least 38 min per day of moderate-intensity exercise to prevent GDM.

Funding. The study was funded by National Institute of Environmental Health Sciences grant R01-ES-019196 to A.F. and supported by National Institutes of Health (NIH) Environmental Influences on Child Health Outcomes (ECHO) Program contract award UG3-OD-023289. S.F.E. was supported by National Institute of Diabetes and Digestive and Kidney Diseases grant K01-DK-105106. A.F. received support from National Institute of Diabetes and Digestive and Kidney Diseases grant P30-DK-092924.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. S.F.E., A.F., and M.M.H. conceived of the idea and designed the study. S.F.E. and R.N. planned the analyses. S.F.E. and J.F. performed the analyses. R.N. verified the analytic methods. S.F.E. drafted the manuscript. A.F., M.M.H., J.F., and R.N. provided critical feedback and helped to shape the final manuscript. S.F.E. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 79th Scientific Sessions of the American Diabetes Association, San Francisco, CA, 7–11 June 2019.