Augmented estimation for t-year survival with censored regression models.

摘要: Reliable and accurate risk prediction is fundamental for successful management of clinical conditions. Estimating comprehensive models precisely, however, a difficult task, especially when the outcome interest time to rare event number candidate predictors, p, not very small. Another challenge in developing arises from potential model misspecification. Time-specific generalized linear estimated with inverse censoring probability weighting are robust misspecification, but may be inefficient setting. To improve efficiency such estimation procedures, various augmentation methods have been proposed literature. These procedures can also leverage auxiliary variables as intermediate outcomes that predictive risk. However, most existing do perform well setting, p In this article, we propose two-step, imputation-based procedure We develop regularized settings where small, along individualized treatment effect reduction. Numerical studies suggest our substantially outperform gains. The applied an AIDS trial treating HIV-infected patients.