Statistical assessment of repeated measures data with application in oncology

Abstract

Statistical analysis methods that analyse complex data appropriately are necessary for medical research to fulfil its aims of improving cancer treatment. To move towards this goal, this thesis uses data from clinical studies and simulation to explore two different areas: (1) genetic associations with efficacy and safety of chemotherapy, and (2) repeated measures analysis of clinical trial data.
Several studies have reported that safety events occurring during chemotherapy are predictors of significantly longer survival for cancer patients. Recent research has identified germline genetic variants associated with either safety or efficacy drug response outcomes. We sought to understand the similarities and differences in the genetic association signals between safety and efficacy endpoints. Our work highlights the difficulties in combining cohorts of patients across cancer types, since differences between cohorts with respect to baseline characteristics and efficacy responses prohibit the use of meta-analysis for the discovery of response-associated factors. Our work confirms that baseline patient characteristics can be important prognostic factors in drug response, however, we conclude that the addition of baseline factors as covariates does not assist in the identification of genetic variants associated with response. Lastly, we develop a novel graphical method to describe the similarities in genetic association results between any two clinical endpoints measured in cancer studies.
Baseline values are commonly measured in clinical trials to help assess drug response following randomisation. Treatment effects on mean change from baseline can be analysed: 1) including the baseline value as part of the treatment response, 2) using only the post-randomisation values in the response analysis only the post-randomisation values and baseline as a covariate, or 3) using the calculated change from baseline value as the dependent variable. We consider each of these analysis methods for their accuracy and precision in estimating the between-group difference in the mean change from baseline. We conclude that the method by which the baseline responses are used in the analysis influences both the accuracy and the efficiency of identifying the response slope difference between treatment arms. The difference in accuracy and precision between methods depends on the number of post-randomisation assessments, with-patient correlation strength and correlation structure of repeated measurements.

Date of Award	2017
Original language	English
Awarding Institution	King's College London
Supervisor	Cathryn Lewis (Supervisor), Ton Coolen (Supervisor) & Irene Rebollo-Mesa (Supervisor)