CAUSAL INFERENCE AND LONGITUDINAL AIDS STUDIES

点击翻译按钮获取中文摘要

DESCRIPTION (Adapted from applicant's abstract) The proposed research will develop statistical semiparametric methods for causal inference in longitudinal studies with time-dependent treatment strategies, time-dependent confounders and covariates, informative censoring on the outcome of interest and covariates and informative monitoring schemes. A template for developing such methods results from the unification of counterfactual causal inference models and censored data models. The counterfactual causal inference model will be used as the actual data model, where the treatment mechanism is generalized to an "action mechanism," representing monitoring, censoring, and treatment action. By also viewing this model as a censored data model, it is possible to exploit the estimation theory developed for censored data models. This provides a blueprint for the construction of locally efficient estimators whose consistency relies only on correct specification of models for the action process (Gill, van der Laan, Robins). This general methods is described in the "research section" of this proposal. The proposed project is to (a) develop the estimators for the most common causal inference and censored data models; (b) develop computer programs to simulate both complex longitudinal data and pure censored data structures and to estimate the causal parameters of interest using the proposed estimators; (c) to apply the methods to various data sets from AIDS studies. The proposed research will allow causal inferences to be drawn from observational data, even in the presence of complicating factors such as informative treatment assignment and informative censoring. The tools to analyze such data will be invaluable in areas such as AIDS research, where action strategies must be tailored to a patient's history and modified in response to changes in confounding variables such as viral load.