CMStatistics 2015: Start Registration
View Submission - CMStatistics
B0594
Title: Testing for informative observation in multi-state models subject to panel observation Authors:  Andrew Titman - Lancaster University (United Kingdom) [presenting]
Abstract: Many observational studies into disease processes do not continually monitor patient status, but instead only observe patients at clinic examination times. In the majority of cases, analysis of the data proceeds by assuming the process which generates the examination times is ignorable. When some or all clinic visits are patient initiated there is a risk that the times will be informative, if for instance patients go to clinic if their condition has deteriorated, and standard estimates may be biased. Building upon previous work which considered a joint parametric model, a joint semi-parametric model is proposed comprising of the underlying parametric multi-state process and an Andersen-Gill counting process model for the observation process. The likelihood for the model can be shown to have a hidden Markov model representation. Moreover, under a null hypothesis of non-informative observation, the two parts of the model can be maximized separately which leads to a convenient construction of a direct score test for informative observation. The test is applied to datasets relating to post-transplantation patients and finite sample properties are investigated by simulation.