Study sample
In the prospective cohort study, we recruited 970 women who received antenatal care from obstetricians, family practitioners, and midwives at 12 clinics throughout the five regions of Ontario from 2015 to 2017. The study methods of the cohort were previously described [19]. Women were eligible for the study if they were English-speaking with a live singleton fetus from 8 weeks and 0 days up to 20 weeks and 6 days gestation. There were 1050 women who were recruited early in their pregnancy and completed the baseline data questionnaire. We followed the women to the end of their pregnancies and extracted the outcome data from their antenatal records. We excluded 80 women from the analysis based on the following criteria: 1) pregnancy with twins or higher-order multiples; 2) a fetus with a known lethal anomaly, a fetal demise, or a termination of pregnancy after enrollment; 3) maternal pathological conditions that affect weight gain; or 4) missing antenatal records for study outcome assessments [19]. The Hamilton Integrated Research Ethics Board (REB #13–021) and local REB committees reviewed and approved the study before its initiation. We obtained informed consent from all participants prior to data collection.
Development of the questionnaire
Six content experts (an obstetrician, a clinical psychologist, a research personality psychologist, a perinatal nurse, a midwife, and a family physician) developed the questionnaire on sociodemographic, psychological, and behavioural factors related to weight gain, which has been published previously [19].
From the questionnaires, we obtained data on marital status, education level, and annual household income (Additional file 1). We classified women as never having smoked, currently smoking, or having quit smoking. We defined chronic health conditions, depression, and anxiety as any such condition diagnosed by a physician.
We calculated prepregnancy BMI as weight in kilograms divided by height in meters squared. We classified prepregnancy BMI as underweight (< 18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25–29.9 kg/m2), and obese (≥30 kg/m2) according to the IOM [1] and World Health Organization criteria [20]. We assessed planned weight gain with the question, “How much total weight do you plan to gain during this pregnancy?” and classified the responses as within, below, or above the IOM guidelines, or not reported. We asked participants to report the recommendations of their healthcare providers regarding weight gain in the first trimester. Such recommendations were categorized as none, within the IOM guidelines, and outside the IOM guidelines. Healthcare providers’ recommendations on total weight gain were also classified as within, below, or above the IOM guidelines, none, or not reported/I can’t remember [1]. We obtained total GWG from the antenatal record by subtracting prepregnancy weight from the final pregnancy weight.
The data collection and definitions of variables for health and pregnancy-related behaviours, as well as psychological factors, are detailed in the cohort’s initial publication [19]. In brief, behavioural factors included diet, eating in front of a screen, sleep, physical activity, pregnancy-related nausea, and food cravings, as well as the means to cope with nausea and food cravings. Diet was assessed with a numerous multiple-choice questions to determine the frequency of behaviours such as drinking soda/juice, eating fast food, eating fruits and vegetables, eating snack foods such as cookies and chips, and eating in the middle of the night. The frequency of eating in front of a screen was similarly assessed. Sleep and physical activity were assessed using a picture and validated scale adapted from Aadahl 2003, which was used to obtain a 24-h MET-time (metabolic equivalent of task) score [21]. Nausea and food cravings were assessed using Likert scales and the means to cope with nausea and food cravings were assessed with multiple-choice questions. Guided by our previous systematic review [18] on psychological factors and GWG and a pilot study [22], we selected validated psychological scales or subscales, or items from such scales and subscales, to assess the following psychological domains: 1) cognition (attitudes on body weight, body image, self-efficacy, weight locus of control, dietary restraint, and barriers to healthy eating); 2) affect (depression, anxiety); and 3) personality (impulse control, perfectionism, motivation, emotional suppression, and the Big Five Personality Factors [extraversion, agreeableness, conscientiousness, emotional stability, and openness]) (Additional file 2).
Assessment of study outcomes
Our study outcome was total GWG, which was classified as inadequate, appropriate, or excessive according to the IOM recommendations [1]. Total GWG was calculated by subtracting the prepregnancy weight from the final measured weight during pregnancy. Weight and height measurements were extracted from provincial Ministry of Health Antenatal Records [23].
Statistical analysis
We summarized descriptive statistics stratified by GWG status by calculating frequencies and proportions for categorical variables, and means and standard deviations for continuous variables (Additional file 1). We used chi-square tests and analyses of variance (ANOVA) for categorical and continuous variables, respectively, to test for significant differences between GWG statuses. We examined collinearity between variables using Spearman’s correlation. For variable pairs with bilateral Spearman’s correlation coefficients ≥ | ± 0.70|, we retained the more psychologically- and biologically-relevant variable. We then performed univariable multinomial logistic regression analyses to assess the associations between the exposure variables with the study outcome (i.e., GWG status), using appropriate weight gain as the reference group (Additional file 3). We employed stepwise multinomial logistic regression for the selection of important exposure variables related to inadequate or excess GWG. A p-value cutoff of < 0.10 was used for entry into the variable selection procedure, as defined by the likelihood ratio test statistic [24]. We retained statistically significant variables with a two-sided p-value < 0.05. Missing data were generally low, varying from 0.1 to 3.2% among the 78 exposure variables included in the study. Few variables had greater than 7% missing data, with the highest variable being family income at 9.59%. We used the fully conditional specification method to create 10 imputed data sets [25] with PROC MI in SAS [26, 27]. For variables in the stepwise regression analysis, we calculated the means of the 10 imputed values and rounded them to the nearest integers for categorical variables, and to the nearest decimal values for continuous variables. We used SAS 9.4 software (SAS Institute, Cary, North Carolina) for data management and statistical analysis.