A0751
Title: Individual treatment rule estimation with M-learning
Authors: Bo Lu - The Ohio State University (United States) [presenting]
Abstract: Individualized treatment rules (ITRs) tailor treatments to individuals based on their unique characteristics to optimize clinical outcomes. Current approaches use outcome modeling or propensity score weighting to control confounding in complex medical data. To avoid model misspecification and the impact of extreme weights, matched-learning (M-learning) was recently proposed for continuous outcomes. We expand the existing M-learning methodology to estimate optimal ITRs under the right censored data. Matched sets are constructed for individuals by comparing observed times, and an inverse probability censoring weight is incorporated into the value function to handle censored observations. Additionally, a full matching design is proposed in M-learning to reduce the potential overuse of a single subject when matching with replacement. The proposed value function is demonstrated to be unbiased for the true value function without censoring. To assess the method's performance, an extensive simulation study is conducted to compare the proposed method with the existing M-learning approach and a weighted learning approach. Results are evaluated based on winning probabilities and estimated values. The simulation reveals that all methods are generally fine in the absence of unmeasured confounders, and different methods show somewhat different performance under various scenarios. But their performances drop substantially in the presence of unmeasured confounders.
