Propensity Score Adjustment in Iterative Study

We began by calculating descriptive statistics at baseline and follow-up for each cycle. We identified nine variables that could plausibly be related to the outcomes and that differed significantly across cycles. We included these nine variables in a multiple logistic regression analysis and saved the predicted probabilities. These predicted probabilities were entered into the outcome models as propensity scores [39 (link)]. Cycle 1 was used as the reference group to compare subsequent cycles to the first cycle, as the study design allowed for modifications and refinements after each cycle through the mixed methods design [36 (link)]. We used a generalized estimating equations (GEE) approach with an exchangeable correlation matrix to logistic regression to adjust for correlations among participants within each cycle. Analyses were conducted in SAS Enterprise Guide Version 7.15 and Stata MP/IC Version 16.