Exploring multi-state modelling in an epidemiological and health economics context

Competing risks analysis is more appropriate than standard survival analysis when there are two or more mutually exclusive possible events, and investigates which one of the events occurs first. Competing risks analysis constitutes the simplest form of multi-state modelling. Multi-state modelling more generally extends the competing risks approach to consider events of interest that can occur after the first event. However, competing risks and multi-state modelling have not been used to their full potential in health research. The aim of this thesis is to demonstrate the potential of multi-state modelling in an epidemiological and health economics context, in areas where it is not widely applied. Focus is on two case studies – one in epidemiology and one in health economics. The first case study is in stroke epidemiology and investigates the outcomes stroke recurrence and death. The research is thought to be the first to comprehensively examine the competing risks stroke recurrence and death without recurrence. It demonstrates the clinical insights that can be gained by decomposing a composite outcome and by studying the cumulative incidence of each event alongside the hazards that drive them. Furthermore, an illustration of the flexibility in predictions of multi-state modelling is given. Predictions at the start of the study and as time progresses are demonstrated. The second study is in health economics and is based on a technology appraisal submitted to the National Institute for Health and Care Excellence in the UK. An objective of this thesis is to compare multi-state modelling with the two common approaches of Markov decision-analytic modelling and partitioned survival. This comparison shows that the conventional decision-analytic modelling and multi-state modelling differ substantially when the assumptions vary between the approaches, but produce equivalent results when they make the same transition assumptions. Therefore, the greatest influence on the clinical and cost-effectiveness results is the choice of assumptions rather than the modelling approach used itself. The research highlights it is imperative to check that any assumptions made are realistic. The comparison of the approaches shows any output required from the conventional approaches can just as easily be produced using multi-state modelling. It is hoped this research will encourage further adoption of multi-state modelling, in the many areas where it has not yet reached its full potential