Validation of Causal Analysis for Obtaining Intervention-Study Results from Non-Intervention Studies

摘要

Recent research on breast cancer etiology has two important characteristics. (1) Most studies focus on a single potential causal agent, and (2) a large number of such agents have been studied, mostly retrospectively. This project performed a literature search in order to categorize recent studies with respect to their inferential structure, and risk factors investigated. Based on this review, this project developed new methods in time-to-event analysis that support a simulation/causation approach to the study of breast cancer. A new method of representing time-to-events was developed. It shows that the Kaplan-Meier method is appropriate for the simulation/causation approach, but that it cannot be used in retrospective studies. The bias was computed explicitly, and a new complementary exponential method for unbiased estimation of incidence rates in retrospective studies was developed. Although the methods of modern causal analysis can be extended to retrospective studies, their atemporal nature makes them less useful from a simulation/causation perspective.