Study participants and setting
Data is from the Study of Women, Infant Feeding, and Type 2 Diabetes After GDM Pregnancy (SWIFT), which is a prospective, observational cohort of 1035 women with GDM pregnancy who delivered a singleton, live birth ≥35 weeks of gestation, 20–45 years, and no history of diabetes [14]. SWIFT women were recruited from 12 Kaiser Permanente medical centers and medical office facilities from 2008 to 2011 throughout the 5000 mile2 northern California region. Women were diagnosed with GDM between 24 and 34 weeks gestation using the Carpenter and Coustan critera [15]. The complete description of the methods and main outcomes of the study have been previously reported [14, 16]. The Kaiser Permanente Institutional Review Board approved this study protocol. This current analysis only uses data collected during pregnancy, and at study baseline (6–9 weeks postpartum) and 1-year postpartum.
Sample selection
Of 1035 women in the SWIFT Study, 21 were excluded due to elevated fasting glucose ≥126 mg/dL, or 2-h glucose ≥200 mg/dL at 6–9 weeks postpartum diagnostic of diabetes, two were ineligible with an infant born <35 weeks gestation, and two women dropped out at baseline. We also excluded 28 women missing beverage intake at baseline, 65 women missing 1-year postpartum body weight measurements, and 29 women missing dietary intake at baseline or glucose tolerance status at 1-year follow up. Therefore, this analysis included 888 women.
Overview of data collection
SWIFT participants provided written informed consent to participate in three in-person visits at 6–9 weeks (baseline), 1-year and 2-years postpartum [14]. At each study visit, trained research staff administered the 2-h 75 g oral glucose tolerance tests (OGTT), measured weight and height, and administered surveys to collect information on dietary intake and other behaviors, socio-demographics, infant feeding practices and other attributes. Both interviewer and self-administered questionnaires collected information on socio-demographics, age, race/ethnicity, family history of diabetes, previous GDM diagnosis, enrollment in special supplemental nutrition program for Women, Infants, and Children (WIC), medical history, and previous contraception.
Clinical perinatal maternal and infant data
Prenatal and delivery characteristics were obtained from electronic medical records (EMR) that included prenatal laboratory results and dates of GDM diagnosis from 3-h 100 g oral OGTTs, pre-pregnancy weight, weight at delivery, pregnancy complications, and type of GDM treatment. Gestational weight gain (kg) was defined as the last weight before delivery minus reported pre-pregnancy weight [14]. The EMR also provided data on newborn birth weight, length, and gestational age.
Anthropometry
Pre-pregnancy weight was self-reported and validated with measurements within one year prior to pregnancy, and first trimester measured weights. At 6–9 weeks and 1-year postpartum women were weighed on a calibrated digital scale (Tanita, Model WB110A, 100A) to the nearest 0.1 lb. in light clothing. Height was measured in bare feet to the nearest centimeter using a stadiometer (Seca, Model 69,072) [14]. Gestational weight gain (GWG) categories (within or below vs. excessive) for pre-pregnancy BMI were determined by the Institute of Medicine (IOM) [17]. Postpartum weight retention was calculated by subtracting the pre-pregnancy weight from weight at 1-year postpartum. Participants were categorized by SPPWR defined as ≥5 kg retained, or not SPPWR defined as <5 kg retained at 1-year postpartum [18].
Infant feeding
Infant feeding practices, including breastfeeding intensity and duration, and formula feeding were assessed prospectively from delivery via feeding diaries, telephone calls at 1 month post-delivery, and monthly mailed questionnaires. Women were asked to report number of breast-milk and/or formula feeding episodes per 24-h, quantity of formula per feeding, and total number of feedings per day from birth to one month, and the average of these same frequencies within the past week and return the surveys each month. At study visits, research staff also administered questionnaires to assess infant feeding practices during baseline (6–9 weeks) and 1-year postpartum in-person visits [14]. For these analyses, breastfeeding duration was used as a covariate in the models.
Oral glucose tolerance test
Two-hour 75 g OGTTs were performed at the baseline exam (6–9 weeks postpartum) and at follow-up exams at 1-year postpartum. Fasting and 2-h plasma samples were stored at −70 °C until assayed. Plasma glucose was analyzed enzymatically with the Hitachi 917 Autoanalyzer by the University of Washington, Northwest Lipid Metabolism and Diabetes Research Laboratory. The Diabetes Endocrinology Research Center Immunoassay Core Laboratory performed the assay of total immunoreactive insulin (microunits/mL) using a double-antibody radioimmunoassay with high precision.
Glucose tolerance classification
Glucose tolerance was defined as normal, glucose intolerant [e.g., prediabetes (Pre-DM) defined as impaired fasting glucose (IFG) between 100 and 125 mg/dL, and impaired glucose tolerance (IGT) for 2-h 75 g post glucose between 140 and 199 mg/dL, or diabetes (DM) based on the American Diabetes Association diagnostic criteria for the 2-h 75 g OGTT (fasting ≥126 mg/dL and/or 2-h ≥ 200 mg/dL) and a repeat OGTT for women with elevated values [19]. The transition in glucose tolerance from baseline to 1-year postpartum was classified into one of three groups: Normal at both baseline and 1-year follow-up (n = 376); Pre-diabetes Pre-DM at baseline and no diabetes (DM at 1-year (n = 261); and progression from normal tolerance to prediabetes (Pre-DM) or DM at 1-year (n = 257).
Maternal dietary intake and physical activity
At 6–9 weeks, women completed the 18-item PrimeScreen to assess dietary intake of total energy (kcal/d), total and animal fat (% of kcals), dietary fiber (% of kcals), dark leafy green vegetables, cruciferous vegetables (broccoli), carrots, other vegetables, citrus fruits, fruits, sweet grains, whole grains, whole and low-fat dairy products, pasta/rice/noodles, whole eggs, margarine, high fat and lean meats, and fried foods. The PrimeScreen was reduced from 135 items to 18 items, and includes the most frequently consumed foods based on data from the Nurses Health Study [20] and the Health Professional study [21]. The 18-item PrimeScreen was validated against the 131-item semi-quantitative food frequency questionnaire (SSFQ) and plasma levels of selected nutrients in a sample of 160 multi-ethnic men and women [22]. Participants completed the PrimeScreen twice, two to four weeks apart, and reproducibility correlation coefficients were moderate to strong, ranging from 0.50 to 0.87. The correlation coefficients between the PrimeScreen and the SFFQ were weak for dark leafy vegetables (0.44), sweet grains (0.48), moderate for dietary fiber (0.58), saturated fat (0.59), animal fat (0.55), cruciferous vegetables (0.64), fruits (0.58), citrus fruits (0.61), whole milk dairy (0.71), margarine (0.67), whole grains (0.51), pasta, rice, and noodles (0.51), beef, pork or lamb 0.63), and fried foods (0.69), and strong for carrots (0.70), low-fat dairy (0.71), eggs (0.82), and processed meats (0.74). The PrimeScreen did not include questions on SSB intake, thus the 13-item Caffeine Questionnaire, developed by Fred Hutchinson Cancer Center was also administered. This questionnaire identified frequency of regular soda (with and without caffeine) and diet soda (with and without caffeine) intake at 6–9 weeks postpartum. This questionnaire has been used to assess beverage consumption in other adult populations [23]. The frequency of consumption was reported in nine categories ranging from “never or less than once per month” to “six or more per day”. A medium serving size was defined as a 12-oz can of soda, and based on these guidelines, women reported their usual serving size per day as small, medium, or large. However, given the rather small percentage of women reporting daily intake of SSB intake, the responses were collapsed into the following categories: never, 1–3 servings per month, 1 serving per week, and 2–4+ servings per week. Dietary intake from the PrimeScreen and the Caffeine questionnaire were analyzed separately.
Physical activity was measured at baseline and 1-year postpartum with the Pregnancy Physical Activity Questionnaire, which is a 32-item, semi-quantitative, validated questionnaire (reproducibility measures of 0.78 to 0.93) that was adapted for the postpartum period [14, 24].
Statistical analyses
Differences in participant characteristics, dietary intake, and physical activity variables between postpartum weight retention categories were assessed using chi-square tests for categorical variables (maternal race, GWG groups, glucose tolerance groups at baseline, breastfeeding intensity and duration categories, glucose tolerance transition groups, and all food groups) and analysis of variance (ANOVA) with F-tests for continuous variables (maternal age, pre-pregnancy BMI, GWG, breastfeeding duration, infant gestation age, infant birth weight, energy intake, dietary fat, fiber and glycemic index).
Variables associated with postpartum weight retention groups, were evaluated via logistic regression models to estimate odds of substantial postpartum weight retention including variables for food and beverage groups. The referent category used was the “healthy” dietary profile (e.g., never ate eggs, processed meats, fried foods, or never drank soda at baseline). Covariates evaluated as potential confounders based on a priori hypotheses included BF duration, maternal education level, pre-pregnancy BMI, gestational weight gain, birth weight, energy intake, glucose tolerance categories at baseline and transition of glucose tolerance categories (prediabetes or diabetes) from baseline to 1-year postpartum. Next, all food and beverage variables significantly associated with SPPWR in the bivariate chi-square/ANOVA analysis (P < 0.05) were included as covariates in the logistic regression models. SAS for Windows 9.1.3 (SAS Institute Inc., Cary, NC) was used, with significance level set at P < 0.05.