609 – Recent Developments in Causal Inference
Estimating the Causal Effect of Treatment in Observational Studies with Survival Time Endpoints and Unmeasured Confounding
Jaeun Choi
Harvard Medical School
James O'Malley
Dartmouth
Ongoing research will be presented on the development and evaluation of methods of accounting for unmeasured confounding of treatment and survival-times measured with incomplete ascertainment due to censoring. The confounding-censoring problem is first characterized using Directed Acyclic Graphs (DAGs). We then consider adapting traditional instrumental variable methods to censored survival-time data and discuss general conditions under which causal effects are identifiable. We apply the methods to evaluate the comparative effectiveness of endovascular surgery and open surgical repair for abdominal aortic aneurysm (AAA) in Medicare patients. Because various physician and patient factors affect treatment selection in observational data, unmeasured factors related to both treatment selection and survival are likely, warranting the use of IV methods. Due to limited follow-up near the end of the study period, censoring is extensive. Therefore, on these data, methods that account for both confounding and censoring have the potential to perform substantially better than naïve methods.