Statistical methods for the simultaneous analysis of quality of life and survival data

The aim of the thesis is to critically review, apply and where appropriate develop statistical methodology for the analysis of longitudinal quality of life data collected as part of a clinical trial where survival is also a key endpoint on which treatments are being compared. The thesis focuses on methods that simultaneously analyse quality of life and survival data, partly in order to provide an overall assessment of the treatments in terms of both endpoints and partly as a means to overcome the problem of missing data that results from patients dropping out of the quality of life study due to death. The thesis also extends the methodology to deal with the additional dropout of patients from the quality of life study prior to death. Three key methods are considered: quality-adjusted survival analysis, multistate modelling and joint modelling. Quality-adjusted survival analysis compares treatments in terms of a composite measure of quality and quantity of life, created by down-weighting survival time according to the reduction in quality of life experienced by patients. Multistate models describe the movement of patients between various health states, defined by levels of quality of life and death, and explore how treatments differ in terms of the transition rates between health states. Joint models describe the change in quality of life over time and the time to death as two interlinked models. The key pursuit is the practical application of methods to data and the thesis makes use of two real datasets, from the MIC trial in lung cancer and the ESPAC trial in pancreatic cancer, that encompass the typical problems faced by analysts tackling this type of data in the real world. The results from this research provide statisticians analysing quality of life data with a variety of possible methods for the analysis of such data that should yield unbiased results.