Fig. 4: Trial primary outcome and subgroup analyses.

We used generalized estimating equations with an identity link and normally-distributed errors to evaluate the effect of the intervention on adherence to medication measured by pill bottles compared with control. The points on the figure are the point estimates and the error bars are the 95% confidence intervals from the relevant sample sizes. These models were adjusted for baseline characteristics and the block randomized design. The primary outcome is shown at the top. The results of exploratory subgroup analyses by key demographic and clinical characteristics also shown; these were performed by repeating the same models within each subgroup, using interaction p-values to assess between subgroups.