|
Abstract:
|
Estimating an optimal individualized treatment rule (ITR) is very important in precision medicine. Binary and multiple treatment problems have been extensively explored in the literature. However, combination therapy or multi-channel treatments are not well studied, yet they can provide more effective treatments than a single-channel treatment, especially for chronic disease. In this talk, we propose a double encoder deep learning strategy to estimate optimal multi-channel ITR (MCITR). Our method estimates the treatment e?ects using concordance between latent embeddings of treatments and patients, which can be effciently captured through neural networks. Further, we study MCITR estimation under budget constraints to facilitate optimal decisions with limited resources. We provide a non-asymptotic value reduction bound and show a faster convergence rate. Finally, we demonstrate the superior performance of the proposed method through extensive numerical simulations and real-data applications.
|
