|
Abstract:
|
In the era of precision medicine, sequential multiple assignment randomized trials (SMARTs) are increasingly needed to develop adaptive interventions for disorders with heterogeneous treatment effects. However, missing data can compromise the validity of inference and the complex structure of SMARTs presents unique challenges when handling missing data. While multiple imputation can facilitate unbiased and efficient estimation for longitudinal data, it relies on the unverifiable assumption that the data is missing at random (MAR). It is unclear how violations of the MAR assumption affect inference in a SMART setting, highlighting the need for sensitivity analyses to assess the robustness of conclusions to departures from MAR. However, existing sensitivity analysis methods for longitudinal data do not address the structural missingness unique to a SMART design. We propose a flexible framework to perform sensitivity analyses in SMARTs based on the pattern-mixture model factorization. We validate our approach with a simulation study and implement our framework to explore how departures from MAR affect conclusions in a SMART study addressing college binge drinking.
|
