Statistical analysis included descriptive statistics for age, sex, race, ethnicity, education, income, health literacy, employment, health insurance, Diabetes Risk Calculator (DRC), and weight status. Chi-square and independent t-tests were conducted to determine if any of the groups differed on baseline characteristics (Supplementary Table 1 ). Data were examined for the presence of outliers, violations of normality (for those continuous variables) and missing data. No violations of normality were detected. Between group differences in changes in BMI and other weight outcomes were prespecified using intention-to-treat (ITT) analysis. To simultaneously account for individual effects regardless of the condition, we employed a linear mixed effect model to a multi-treatment framework (33 (link)) for the treatment effect analysis (34 ). To be specific, two group dummies are in the model along with assessment time dummies and their interactions. This model allows us to control error non-independence of over time assessment within the same individual and heteroskedasticity caused by between individual heterogeneity, and a-priori-determined covariates that are influencing factors of outcome-specific production. The goal was to make more robust inferences about the treatment effect of main outcomes of interest: for example, the effect of Class/IVR and DVD/IVR in reducing BMI over 18 months when compared to SC group. For those participants with missing outcome measurements, we replaced the missing data with their baseline value following the Baseline Carried Forward approach.
Additionally, we conducted analysis based on participants completing at least 4 sessions (i.e., meeting NDPP threshold for recognition standards) (35 ), at least 6 months (i.e., core intervention effects), and the full 12 months (i.e., post-core effects). For the purposes of these analyses, class and “Live-Call” completion were calculated based on attendance, DVD was based on participant self-report, and IVR call completion was based on the voice files for the lesson of the week being played (24 (link)). Further, for the dichotomous outcome measures (i.e., achieve 5% weight loss goal), we treated those models as linear probability models in order to retain the straight-forward treatment effect interpretation of the results by applying generalized linear models in the analysis. Means and standard deviations for all primary and secondary outcomes at baseline, 6, 12, and 18 months are also presented. All statistical analyses were conducted in Stata v16 and the 5% significance level was used.
Additionally, we conducted analysis based on participants completing at least 4 sessions (i.e., meeting NDPP threshold for recognition standards) (35 ), at least 6 months (i.e., core intervention effects), and the full 12 months (i.e., post-core effects). For the purposes of these analyses, class and “Live-Call” completion were calculated based on attendance, DVD was based on participant self-report, and IVR call completion was based on the voice files for the lesson of the week being played (24 (link)). Further, for the dichotomous outcome measures (i.e., achieve 5% weight loss goal), we treated those models as linear probability models in order to retain the straight-forward treatment effect interpretation of the results by applying generalized linear models in the analysis. Means and standard deviations for all primary and secondary outcomes at baseline, 6, 12, and 18 months are also presented. All statistical analyses were conducted in Stata v16 and the 5% significance level was used.
Free full text: Click here
