Proximal Survival Analysis

Specialeforsvar: Hjalte Søberg Mikkelsen

Titel: Proximal Survival Analysis

Abstract:  Many clinical and epidemiological studies seek to analyze the time to an event. A common complication in such studies is right censoring, which needs to be addressed properly by the practitioner, as to not introduce biases in the estimates. One popular way of addressing right censoring is to assume that it is non-informative. If, however, the censoring is informative, one can utilize possibly time-varying covariates to get conditional independent right censoring. Realistically, the covariates can
rarely capture all associations between the censoring and the event time, which can mean that the covariates are at best proxies of the true underlying mechanism. I will in this thesis show the proximal survival analysis framework, which is a new way to handle dependent right censoring. Three proximal estimators will be derived, and I will show that these estimators are consistent and asymptotically normal. The proximal estimators will then be tested in a simulation study, where they will be compared to already popular estimators.

Vejleder: Munir Haibu
Censor:   Sören Möller. SDU