Variable selection based on A-learning for optimal treatment regimes with survival data under semiparametric model

Key Points

• This research aims to optimize treatment decision regimes using variable selection based on the A-learning framework for survival data.
• Utilized A-learning framework with estimating equations and a loss function integrated with regularization penalties.
• Conducted extensive simulation experiments to validate the methodology.
• Analyzed data from the ACTG 175 clinical trial involving HIV-infected patients.
• Achieved effective variable selection when baseline covariate or propensity score models were correctly specified.
• Showed improved treatment decision-making in high-dimensional time-to-event data.
• Simulation results confirmed the performance of the method under various conditions.