The analysis of a bivariate multi-state Markov transition model for rheumatoid arthritis with an incomplete disease history.

Abstract

In many long-term chronic diseases, patients pass through an observable sequence of ordered clinical states as their condition progressively worsens. Often the information on which disease state the patient is in is incompletely recorded, usually with information only available on the occasion of a clinic visit. This article describes a novel analysis of data from a clinical trial, in which several such outcome measures of disease state have been recorded simultaneously. The article is motivated by the analysis of a multi-centre double-blind placebo-controlled clinical study into the effect of continual low dose corticosteroid treatment on the progression of X-ray scores for patients with rheumatoid arthritis. Previous methods of analysis of such data have been based on an independence analysis, thus ignoring any correlation that may exist between the outcomes. This article shows that such an approach can lead to biased underestimates of the covariate effects if an independence model is used. Biased estimates of the covariate effects were found when the model was fitted to the trial data. The bivariate model was also shown to provide a significantly better fit to the data. However, the bivariate model did prove more difficult to fit, and both models demonstrated a highly significant treatment effect with comparable clinical effect.