Overview and design
The overall goal of the project of which this study is a part was to expand the understanding of how to slow the accumulation of weight in childbearing women. This paper reports on the pregnancy portion of a longer study that followed women until 12 months postpartum in order to evaluate the effectiveness of pregnancy intervention and combined pregnancy and postpartum intervention on weight retention at 12 months postpartum. Thus during pregnancy, two arms received the same intervention (one of these arms becomes a control postpartum) and the third arm was the placebo control (Fig. 1). The primary hypothesis for the pregnancy portion of the study was that the intervention would lead to a 10 percentage point reduction in the proportion of women with excessive total GWG in the intervention arms (45%) compared to the placebo control arm (55%).
This trial was a double-blind, three-arm randomized trial with a parallel group design with the individual as the unit of randomization and analysis [12]. Pregnant women were screened by research staff in prenatal clinics, private obstetric practices, ultra-sound offices, and over the phone and online in a large Northeastern US city from May 2011 through July 2012. Inclusion criteria were age 18-35 years and gestational age ≤ 20 weeks at time of enrollment. Exclusion criteria included body mass index (BMI) < 18.5 and ≥ 35 kg/m2, multiple gestation, weight-affecting medical or psychiatric conditions, and no e-mail address [12]. The age limits were set by the Early Adult Reduction of Weight through LifestYle interventions (EARLY) consortium of weight management studies of which this trial was a part [13]. Eligible women provided written informed consent and signed a form for release of their medical records. Upon consent, participants were electronically randomized via computer to 2 intervention arms and 1 control arm within 4 income (with low income defined by Medicaid eligibility) and BMI (normal BMI and overweight plus obese class 1 BMI) strata in blocks of six: strata 1 - normal BMI low income; strata 2 - normal BMI high income; strata 3 - overweight plus obese class 1 low income; and strata 4 - overweight plus obese class 1 high income. The study protocol was approved by the University of Rochester Research Subject Review Board and the Cornell University Institutional Review Board.
Intervention
Participants assigned to the intervention arms received access to the intervention website and those assigned to the placebo control condition received access to the control website, both of which were password protected. A more detailed description of the theory, development and implementation of the pregnancy intervention is available [14, 15]. Briefly, the self-directed, integrated online and mobile phone behavioral intervention was designed using the Integrative Model of Behavior Prediction [16] and the Behavior Model for Persuasive Design [17] in addition to formative research with the target population [14]. It was based on a non-electronic pregnancy lifestyle intervention that was demonstrated to be efficacious in low-income women [18]. Women in the intervention arms received access to three behavior change tools including a weight gain tracker, a diet and a physical activity goal-setting and self-monitoring tool, as well as, health information including tips, articles, frequently asked questions; a description of pregnancy and parenting-related resources available in the local community; a blogging tool; and an event and appointment reminder. Fig. 2 shows these tools, their behavioral targets and leverage points from the theoretical model [17]. Fig. 3 shows the dashboard available to participants in the intervention group. The placebo control arm received access to all the features above except the weight gain tracker and the diet and physical activity goal-setting and self-monitoring tools since the latter were hypothesized to be the active ingredients of the intervention. Reminders and informational content, that differed by arm, were distributed weekly via e-mail messages to all participants. In summary, two different suites of tools were made available to trial participants on a password protected study website and mobile phone platform. Women were reminded weekly to login, and they decided what, when, and how much they would use the tools made available to them. All women in the trial received standard prenatal care from their self-selected health care provider.
Measures and data collection
Data were collected at screening by research staff, with two online surveys completed by participants during pregnancy (one before 28 weeks gestation and one from 32 weeks until delivery), and through an audit of the women’s prenatal, labor and delivery, and 6 week postpartum medical records by research staff. Socio-demographic, behavioral, psychosocial and adverse event data were collected through online surveys [12]. Health and weight data were abstracted from the medical record [12]. Participants received financial incentives for enrolling in the study and for each of the online questionnaires they completed. They could earn up to $45 for completing all data collection activities during pregnancy. They were not paid for engaging with intervention or control treatments. An online satisfaction survey using EARLY questions and for which no financial incentive was provided was made available to participants to complete at 6 weeks postpartum [13].
The pre-specified primary outcome for evaluating the effectiveness of the intervention was the proportion of women with total GWG above the upper limit of the range for total GWG defined by the Institute of Medicine (IOM) for each BMI group [1]. Total GWG was calculated as the difference between the first weight at < 14 weeks gestation and the last weight at ≥37 weeks in pregnancy. The binary outcome, the proportion of women with excessive total GWG, was determined by comparing the difference for each woman to the IOM upper limit for GWG range for each BMI group: normal BMI - > 16 kg; overweight BMI - > 11.5 kg; and obese class 1 BMI - > 9 kg [1]. Excessive average weekly GWG in the last half of pregnancy and total GWG in kg were pre-specified secondary outcomes. Average weekly GWG was calculated as the difference between the last weight at ≥37 weeks of gestation and the weight nearest to 20 weeks gestation (+/− 2 weeks) divided by the number of weeks between the two weights. This value was defined as excessive if it exceeded the upper limit for weekly weight gain for each BMI group as specified by the IOM [1]. Adherence to the treatment protocol was defined as logging into the treatment arm specific project website at least once in each 45 day interval during pregnancy. This time interval was based on the schedule of prenatal care visits which are on average every 30 days during pregnancy. Formative research showed that a large proportion of women did not have scales in their homes and thus in order to use the weight gain tracker they needed to use the weights measured at their prenatal care visits [14]. This level of adherence was considered as providing a minimal possibly effective dose of exposure to treatment.
Sample size
The target sample size of 1641 (547/arm) was determined by the weight retention endpoint at 12 months postpartum, the ending time point for the overall study [12]. Assuming 15% attrition during the pregnancy period, yielding 465 controls and 930 intervention subjects, we estimated 87% statistical power to detect a reduction of 10 percentage points from the overall expected 55% of women with excessive total GWG in the control group. The expected rate was calculated from data made available through the Finger Lakes Perinatal Data System which aggregates birth certificate information from hospitals in the region. (See Additional file 1: Table S1.) The study was not powered to examine intervention effects within strata. A Bonferroni-corrected significance level of 1.67% (2-sided), reflecting the three primary comparisons in the entire study, was applied to the primary analysis of excessive total GWG. Otherwise, no multiple comparison procedures were applied and a significance level of 5% was used.
Statistical analysis
Missing data were handled using multiple imputation to address issues of bias which may result from analyzing only complete cases [19]. Sufficient weight information for the calculation of the primary outcome required having a measured weight at both < 14 weeks and ≥ 37 weeks of gestation. If weight information was insufficient, the first, 20 week, and/or last weights were imputed using Statistical Analysis System (SAS) Proc MI. A previous evaluation of the non-electronic version of the intervention indicated that both income and BMI affected GWG outcomes, leading to the stratified randomization design for the present study [18]. Therefore, to allow for a potential interaction between income, BMI, and intervention within the imputed data, 60 imputed data sets were created within each income and BMI group and used in the analysis.
The overall proportion of women with excessive GWG on each arm was estimated by pooling across the imputed data sets, accounting for the variability between imputations, as described by Ratitch, Lipkovich and O’Kelly [20]. Relative risk (RR) estimates for the binary outcomes (proportion of women with excessive total and weekly GWG) were obtained from log-binomial regression models. In two of the sensitivity analyses, some of the log-binomial models did not converge in a few of the 60 imputed data sets. The COPY method was then applied to address the computational issues [21] and was successful in addressing convergence issues in all cases. Mean differences in total GWG (kg) were similarly estimated from least-squares regression models. All multivariable model estimates for the overall effect of intervention versus control arms were adjusted for strata, early pregnancy BMI, gestational age at delivery (to adjust for length of gestation and thus the amount of time over which weight could be gained), and two variables to additionally adjust for the timing of weight measurements: the number of weeks between the first and last pregnancy weight, and the number of weeks between the last pregnancy weight and delivery. SAS Proc MIANALYZE was then used to combine the model results from within each imputed data set in the standard way according to Rubin’s Rules. The interaction between treatment arm and the four category strata variable was assessed as a secondary analysis. Several sensitivity analyses were conducted with various subsamples of study participants.
Descriptive information is presented on engagement within treatment arm. Satisfaction with study participation was compared by treatment arm using chi-square analysis. To describe the safety of the intervention, 10 infant and maternal health outcomes and complications were compared by arm within each of the four strata using Fisher’s Exact Test. Birthweight was compared using the T-Test.
