Inverse probability weighted estimation of dynamic treatment regimen means in sequential multiple assignment...

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Health Services Research

Background:

• Dynamic treatment regimens (DTRs) personalize sequential treatment decisions in various diseases.
• Sequential multiple assignment randomized trials (SMARTs) evaluate and optimize DTRs through multi-stage randomization.
• Inverse probability weighting (IPW) is commonly used to estimate DTR means in SMARTs, but missing data poses a challenge.

Purpose of the Study:

• To evaluate the performance of complete case analysis (CCA) and multiple imputation (MI) in handling missing data for estimating DTR mean outcomes using IPW within a two-stage SMART.
• To compare the bias and standard errors of CCA and MI under various missing data scenarios and proportions.

Main Methods:

• Simulated 1000 datasets with 400 participants each, based on a two-stage SMART design where non-responders are re-randomized.
• Estimated four DTR means using IPW, assessing performance of CCA and MI under missing data scenarios defined by m-DAGs.
• Evaluated missing data proportions (20%, 40%) and dependence on stage 1 outcome, stage 2 treatment, and final outcome.

Main Results:

• MI exhibited minimal bias in most scenarios, except when stage 1 intermediate outcome was missing dependent on baseline variables and stage 1 treatment.
• CCA showed greater bias than MI when data were missing dependently on other variables (e.g., stage 2 treatment missing based on stage 1 outcome).
• Empirical standard errors were comparable, with MI generally yielding slightly lower values.

Conclusions:

• For the studied prototypical SMART design, MI is generally preferred over CCA for estimating DTR mean outcomes via IPW due to minimal bias and slightly lower standard errors.
• MI proves effective in handling missing data complexities inherent in sequential randomization and intermediate outcome dependencies within SMARTs.