Catalog Home Page

Incorporating endpoint uncertainty into biomedical survival analyses

Koh, Jeanette (2015) Incorporating endpoint uncertainty into biomedical survival analyses. Honours thesis, Murdoch University.

[img]
Preview
PDF - Whole Thesis
Download (2MB) | Preview

Abstract

A key problem in medical research and practice is that of accurately identifying endpoints or times-to-event in stages of disease progression or treatment outcomes. In chronic or long-term conditions such as Human Immunodeficiency Virus (HIV) infection, the reaching of the endpoint is determined by comparing an individual’s biomarker measurements over a period of time to a pre-determined threshold. The small number and irregular spacing of such measurements often make it difficult to determine when a patient has truly reached the endpoint, and current methods do not take this uncertainty into account. In this thesis, we consider the impact of endpoint uncertainty on statistical inferences in Cox regression for time-to-event data, review the current use of weights on individuals in Cox modelling, and propose a novel approach of weighting an individual’s biomarker measurement times, based on schemes which determine the probability of that biomarker measurement being indicative of a true endpoint. Two weighting schemes are developed and illustrated — the first-of-two weighting (FOTW) scheme, based on the practice of taking the first of two consecutive above-threshold measurements as the endpoint measurement (termed the first-of-two unweighted, or FOT U/W scheme); and the longitudinal conditional probability weighting (LCPW) scheme, which applies weights to all measurements based on a longitudinal mixed-effects model (LMEM) of the data. We show through simulations that both the FOTW and LCPW schemes have good power in detecting significant differences between study groups using Cox analysis but are sensitive to differences in observational frequency between the groups, and we explore the robustness of the LCPW scheme to misspecifications of the LMEM. Finally, we illustrate the use of these schemes on a dataset from the Western Australian HIV study and discuss areas for future investigation, including ways to improve the reliability and robustness of the schemes.

Publication Type: Thesis (Honours)
Murdoch Affiliation: School of Engineering and Information Technology
Supervisor: Admiraal, Ryan and James, Ian
URI: http://researchrepository.murdoch.edu.au/id/eprint/29985
Item Control Page Item Control Page

Downloads

Downloads per month over past year