Presented by Alise Mendoza
Masters Candidate in Biostatistics
Plan B Adviser: Dr. Thomas Murray
Behavioral surveys are widely used in clinical and public health research to measure patient-reported outcomes such as self-efficacy, health beliefs, and medication adherence. These multi-item instruments are typically analyzed using a total score that reflects the sum or average of the item responses, with randomized groups compared on these total scores using standard regression approaches such as ANOVA. This approach may overlook item-specific treatment effect heterogeneity and limit potential for efficiency gains from baseline covariate adjustment. In my thesis, I conducted a novel item-level analysis using generalized estimating equations (GEE) for questionnaire outcomes collected in the mGlide randomized controlled trial, which evaluated a mobile health intervention for hypertension management. Item-level GEE methods allow estimation of item-specific treatment effects while accounting for within-person correlation in responses across items and allowing estimation of the effect on a derived total score outcome via regression contrasts. These methods also enable item-matched adjustment for baseline responses, which may facilitate efficiency gains. I analyzed responses from the MASES-R and HB-MAS questionnaires, comparing traditional total-score models with item-level marginal models. I also conducted a complementary simulation study to evaluate the potential for item-level modeling to provide efficiency gains.
