Theory and Methods for Causal Inference in Chronic Diseases

点击翻译按钮获取中文摘要

Chronic diseases such as cardiovascular disease and HIV create an immense health and economic burden, both within the USA and globally. With recent technological advances, the chronic disease research enterprise is rapidly becoming data-intensive and data-driven. Massive and complex data provide unprecedented opportunities for discovering optimal treatment strategies for chronic diseases. However, these complex data also present novel challenges for statistical analysis. Patients may visit the clinic at irregular intervals, may drop out of studies, and may discontinue prescribed treatments prematurely. In addition, there may be "confounding by indication", in that some treatments may have been prescribed preferentially to sicker patients. These features can be barriers to effectively translating rich information into meaningful knowledge. The overarching theme of this project is to develop new data analysis methods that tackle these important and recurring challenges. This work aims to advance statistical science through the development of novel approaches to address these difficult challenges, where existing methods do not apply or suffer from major drawbacks. The research will also provide subject matter scientists with a principled way to approach scientific questions in these settings to discover optimal treatment strategies for patients. This research project has the following goals. 1) Develop estimators of survival distributions as a function of time to treatment discontinuation using a dynamic-regime marginal structural models approach. Treatment discontinuation arises frequently in clinical practice, complicating the analysis and interpretation. The objective here is to develop an instructive demonstration of how careful conceptualization of this problem leads to an unambiguous definition of a sensible treatment effect and to valid inferences, shaping a principled approach to dealing with treatment discontinuation. 2) Develop efficient estimators for Structural Nested Mean Models (SNMMs) from longitudinal observational studies in the presence of informative censoring using semiparametric theory. Time-varying confounding by indication is a widespread phenomenon and causes selection bias in the estimation of treatment effect. SNMMs have been proposed to overcome this issue; however, their use in practice is still unpopular, partly because the efficiency of the estimators is highly dependent on the choice of estimating equations, and the theory is still underdeveloped in many settings. The investigator plans to develop improved estimators of causal parameters in SNMMs in the presence of censoring, which gain both efficiency and robustness to nuisance model specification over existing methods. 3) Develop a new framework of continuous-time SNMMs. In many realistic situations, the outcomes and treatments are more likely to be measured at irregularly spaced time points. Most of the existing SNMMs literature uses a discrete-time setup, which is overly simplified and therefore impractical. The investigator aims to provide a unified framework for SNMMs with continuous-time processes, establishing a novel area of research in causal inference.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.