Abstract:
|
Intention-to-treat (ITT) is widely considered the recommended approach for analyzing primary outcomes in randomized clinical trials, but can yield biased treatment effects in the presence of missing outcome data (Little & Kang, 2015, Statistics in Medicine). One option that requires more research for obtaining accurate treatment effect estimates when outcome data are missing is the use of outcome data obtained at intermediate time points between the pretest and the posttest. The current study examines how such data may be used analytically to reduce bias and increase precision of treatment effects in RCTs when outcome data are missing. Various missing data properties (e.g., MAR vs MNAR missing data mechanisms, differential drop-out types), analytic approaches (e.g., pre-post ANCOVA, mixed-model approaches which allow MAR data), and longitudinal RCT designs with intermediate observations are examined. Practical recommendations are provided to researchers concerning when it is worthwhile to employ a research design with intermediate observations to obtain more accurate inference and estimates for treatment effects in RCTs when outcome data are missing.
|