Statistical modelling of competing risks with incomplete data: with applications to allogeneic stem cell transplantation

Abstract

Incomplete data often pose a challenge in observational studies in medicine. This is especially true when studying the time from a clinically relevant starting point, to a particular event. For example, we may be interested in the time to an infection after an allogeneic haematopoietic stem cell transplantation (alloSCT), a treatment primarily given to individuals with blood cancer. Here, the time to infection may be unknown for an individual if a) the study ends before an infection has occurred (right censoring); or b) the individual dies before becoming infected (competing risk). Additionally, individual-specific characteristics that could be predictive of the time to infection, such as the age of a patient at alloSCT (a baseline covariate) or the number of immune cells circulating in their blood over time (a longitudinal covariate), may also be partially observed. Inappropriate handling of missing values, such as naive exclusion of cases with missing data, can dramatically reduce statistical power, and can yield biased estimates of targeted quantities. This dissertation therefore aims to develop, assess, and apply statistical methodology for dealing with missing data (with a focus on multiple imputation of missing covariates) in the context of alloSCT studies with competing risks outcomes.