The Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-Be (nuMoM2b) is a prospective cohort study designed to identify factors that contribute to adverse pregnancy outcomes [15]. This prospective cohort study enrolled 10,038 nulliparous women with singleton pregnancies from 8 clinical centers in the United States (Case Western Reserve University; Columbia University; Indiana University; University of Pittsburgh; Northwestern University; University of California at Irvine; University of Pennsylvania; and University of Utah). In brief, women were eligible for enrollment if they had a viable singleton gestation, had no previous pregnancy that lasted more than 20 weeks of gestation, and were between 6 0/7 weeks of gestation and 13 6/7 weeks of gestation at recruitment. Exclusion criteria were maternal age younger than 13 years, history of three or more spontaneous abortions, current pregnancy complicated by a suspected fatal fetal malformation, known fetal aneuploidy, assisted reproduction with a donor oocyte, multifetal reduction, or plans to terminate the pregnancy. Nulliparity was defined as having had no prior pregnancy lasting 20 weeks or more based on self-report. All local institutional review boards approved the study protocol, and participants provided written informed consent prior to enrollment.
Leisure-time PA during pregnancy was self-reported at a study visit in each trimester (6- < 14, 16- < 22, and 22- < 30 weeks gestation, with the third visit occurring at least 4 weeks after the previous visit) using standardized physical activity questions adapted from the Behavior Risk Factor Surveillance System (BRFSS) [16, 17]. Women were asked whether they participated in any leisure-time PA during the previous four weeks. If yes, they were asked to describe the activity in which they spent the most time and to provide information on the number of times per week they had taken part in this activity over the four weeks, and how many minutes per time. For running, jogging, walking, cycling and swimming, they were also asked about distance. This was repeated for the second and third activity in which they spent the most time. By design, these questions assess structured, physical activities which have been linked to health, including pregnancy health [18, 19]. Each activity was assigned an intensity level [metabolic equivalent (MET)] by trained coders based on the Physical Activity Compendium [20]. This MET value was multiplied by the frequency and duration of each activity to obtain volume for each activity (MET-minutes per week), and then summed across all activities reported by the participant. In addition to analysis of MET-minutes per week, adequacy of a participant’s exercise regimen was assessed as ≥150 min of moderate activity per week, ≥75 min of vigorous activity per week, or an equivalent combination of the two. Moderate activity was defined as leisure-time PA with an intensity 3 ≤ METs< 6 and vigorous activity was defined as METs≥6. A total of 10,022 women provided activity data at one or more visits; 10,016 at Visit 1; 9408 at Visit 2; and 9215 at Visit 3. For all 3 visits, the median (Q1, Q3) was 1 (0,2) for number of activities reported.
Adverse pregnancy outcomes were adjudicated from medical record abstraction, performed by certified research personnel. Quality control checks via re-abstraction were performed by the site principal investigator on a random selection of charts with and without complications. For 82% of the charts reviewed, no discrepancies were found. For those charts with discrepancies between the two abstractions, these differences were generally minor and not related to the primary adverse pregnancy outcomes. Preterm births were those delivered prior to 37 weeks, and further classified as spontaneous if after spontaneous onset of labor or premature rupture of membranes. Hypertensive disease of pregnancy included preeclampsia with and without severe features, super-imposed preeclampsia, eclampsia, and gestational hypertension, as defined according to established criteria [21]. Gestational diabetes mellitus was defined by one of the following glucose tolerance testing (GTT) criteria: fasting 3-h 100 g GTT with two abnormal values [fasting 95 mg/dL or greater, 1-h 180 mg/dL or greater, 2-h 155 mg/dL or greater, 3-h 140 mg/dL or greater]; 2) fasting 2-h 75 g GTT with one abnormal value [fasting 92 mg/dL or greater, 1-h 180 mg/dL or greater, 2-h 153 mg/dL or greater]; or 3) nonfasting 50-g GTT 200 mg/dL or greater if no fasting 3-h or 2-h GTT was performed [22]. If no GTT data were available, the clinical diagnosis from chart abstraction was used for GDM classification. Small-for-gestational-age (SGA) birthweight was defined as <5th percentile for gestational age at delivery based on Alexander fetal growth curves [23]. Analysis was restricted to pregnancies carried 20 or more weeks of gestation. Women with pre-existing diabetes were identified via chart abstraction of the medical record from the delivery hospitalization (n = 151) and were excluded from analysis of gestational diabetes.
Covariates were recorded at the enrollment study visit and included maternal age, race/ethnicity, nausea, vomiting, retching in the 12 h prior to the study visit interview (as derived from the PUQE survey), [24] smoking during the 3 months prior to pregnancy, chronic hypertension or pre-existing diabetes, and education. Early pregnancy body mass index (BMI, kg/m2) was based on measured weight and height, which is highly correlated with pre-pregnancy BMI.
Patterns of leisure-time PA throughout pregnancy were identified using growth mixture modeling [25] conducted in Mplus version 8 with missing values addressed using full information maximum likelihood [26]. Models included leisure-time PA values (METs-minutes per week) from the three study visits. Due to the skewness of the distributions of total weekly MET-minutes, we applied a log transformation. Having determined the most appropriate form of the growth curve (linear vs. quadratic), we compared the fit of growth mixture models with one- to five-class solutions. Model fit was assessed based on standard fit indices (i.e., AIC, BIC, and sample-adjusted BIC); the inclusion of additional classes was evaluated using likelihood ratio tests. In addition, the number of participants in each class was examined, requiring a minimum of 5–10% of participants per class to avoid the inclusion of spurious classes. Final model selection was informed by the statistical results and interpretation of the classes. Specifically, we determined that a 5-group solution most ideally fit the data based on the statistical results and the fact that patterns matched what is generally known about activity across gestation. For example, this 5-group solution included a group with zero or very low leisure-time PA across pregnancy, while other groups, on average, reported reductions as pregnancy advanced.
We summarized maternal characteristics according to leisure-time PA pattern group, and compared these characteristics using chi square tests and analysis of variance F-tests. We compared occurrence of adverse outcomes in each leisure-time PA pattern group using logistic regression, and adjusted differences for age (linear and quadratic terms), race/ethnicity, early pregnancy BMI (linear and quadratic terms), and smoking status 3 months prior to pregnancy which were covariates selected a priori. As sensitivity analyses, we conducted additional adjusted analyses adding the probability of pattern group membership taken from the growth mixture model to the covariates in the logistic regression model. Odds ratios from all logistic regression models were computed with the most active class as the referent group.