Sensitivity Analyses for Missing in Repeatedly Measured Outcome Data

We discuss practical aspects of conducting sensitivity analyses for missing data with a repeatedly measured outcome. Our motivation is a SMART trial with a repeatedly measured outcome subject to missingness. We discuss and describe delta-based controlled imputation approaches to conducting sensitivity analyses for such trials that typically use linear mixed models for their primary analysis. We find that delta-based sensitivity analyses for trials with repeatedly measured outcome variables are enhanced by using MICE for the imputation. Further, including last-observed-before-time covariates is critical for a repeatedly observed outcome. We also develop some novel metrics for judging the adequacy of sensitivity analyses. Trial Registration: Tailoring Mobile Health Technology to Reduce Obesity and Improve Cardiovascular Health in Resource-Limited Neighborhood Environments: NCT03288207.