|
Abstract:
|
In personalized medicine, optimal individualized treatment rules can be estimated via either outcome regression estimators or inverse probability weighted estimators. The strengths of both estimating methods can be combined using the targeted learning, leading to targeted maximum likelihood estimators (TMLE). To simplify estimated treatment rules, we can apply the decision-tree procedure to prune them and make them more interpretable in practice. In addition, we can use the data-splitting strategy to unbiasedly estimate the values of the pruned treatment rules. The performance of the proposed approach is demonstrated via simulations and a real application.
|
