Optimal Statistical Inference for Individualized Treatment Effects in High-Dimensional Models

Abstract The ability to predict individualized treatment effects (ITEs) based on a given patient's profile is essential for personalized medicine. We propose hypothesis testing approach choosing between two potential treatments individual in the framework of high-dimensional linear models. methodological novelty lies construction debiased estimator ITE and establishment its asymptotic normality uniformly an arbitrary future observation, while existing methods can only handle certain specific forms observations. introduce procedure with type I error controlled establish power. proposed method be extended making inference general contrasts, including both average effect outcome prediction. optimality from minimaxity adaptivity perspectives procedure. An extension approximate models also considered. finite sample performance demonstrated simulation studies further illustrated through analysis electronic health records data patients rheumatoid arthritis.