Are You in the State of Mind for Multistate Models?

Enzmann and colleagues applied multistate models on published data from a randomised controlled trial comparing vein bypass surgery with nitinol stent angioplasty for femoropopliteal lesions. They provide a step by step explanation of this alternative statistical approach to analysing time to event data and discuss the potential benefits of this method.
Time to event data are often visualised using Kaplan–Meier estimators, with group comparisons typically made via the log rank test or Cox models when covariable adjustments are necessary. These traditional methods summarise comparisons using a single measure: the hazard ratio (HR), which represents the relative likelihood of an event occurring in one group compared with another at any time during follow up. However, the HR is a complex measure, and its interpretation can be challenging. Cox models assume that the risk between groups is constant throughout the follow up period, even though the underlying baseline risk may vary over time. These models are well suited to analysing survival time, focusing on a single event type. However, when attention shifts to events other than death, death itself becomes a competing risk, potentially compromising the results. Cox models and the potential problems of competing risk have been discussed in recent European Journal of Vascular and Endovascular Surgery Edutorials.3