Walker, Ann Sarah; (1999) The analysis of multivariate failure time data with application to multiple endpoints in trials in HIV infection. Doctoral thesis (Ph.D.), University College London (United Kingdom).

Text The_analysis_of_multivariate_f.pdf Download (28MB)

Abstract

Endpoints currently in use in clinical trials in HIV infection are a composite which include death, clinical events, measures of quality of life, events based on laboratory markers and adverse events. A composite endpoint, defined as the first occurrence of any one of a set of events (including death), is generally accepted as appropriate for use in Phase III trials of efficacy and safety, and is analysed using univariate failure time methods. However, events in such a composite endpoint are often heterogeneous in their effect on subsequent mortality, and also in their pathology, physiological system affected, and potentially their response to antiretroviral treatment. Statistical methods for the analysis of such multivariate failure time data fall into three broad classes—marginal, frailty and conditional models. A new semi-parametric marginal model based on Poisson regression for failure data and GEE is developed, and the use of correlation structures other than independence investigated. This estimation process is compared in a simulation study with the standard method for multivariate failure time data based on a Cox partial likelihood. A new simple binary frailty model is also developed using both parametric and semi-parametric baseline hazards. Various estimators of the semi-parametric and parametric hazard are compared in a simulation study. Finally, these methods are applied to data from the Delta and Concorde trials. Initially, AIDS events are split into broad classes based on subsequent risk of death, and then multivariate failure time methods applied to these classes. Combination antiretroviral therapy can be shown to delay progression to more severe AIDS events compared to monotherapy. These late events are generally untreatable and prophylaxis is not available. The effect of treatment on individual AIDS events varies considerably. More than one composite endpoint can also be analysed using these methods, providing an overall test of treatment effect across the endpoints which allows for the correlation between them.

Download activity - last month

Export as CSVJSONXML

Download activity - last 12 months

Export as XMLCSVJSON

Downloads by country - last 12 months

Export as XMLCSVJSON

Archive Staff Only

View Item