Background: Heart failure-related hospital readmissions and mortality are often outcomes in clinical trials. Patients may experience multiple hospital readmissions over time with mortality acting as a dependent terminal event. Univariate composite end points are used for the analysis of readmissions. We may amend these approaches to include emergency department visits as a further outcome. An alternative multivariate modeling approach that categorizes hospital readmissions and emergency department visits as separate event types is proposed.
Methods and results: We seek to compare the modeling approach which handles event types as separate, correlated end points against composites that amalgamate them to create a unified end point. Using a heart failure data set for illustration, a model with random effects for event types is estimated. The time-to-first event, unmatched win-ratio, and days-alive-and-out-of-hospital composites are derived for comparison. The model provides supplementary statistics such as the correlation among event types and yields considerably more power than the competing composite end points.
Conclusions: The effect on individual outcomes is lost when they are intermingled to form a univariate composite. Simultaneously modeling different outcomes provides an alternative or supplementary analysis that may yield greater statistical power and additional insights. Improvements in software have made the multitype events model easier to implement and thus a useful, more efficient option when analyzing heart failure hospital readmissions and emergency department visits.
Keywords: atrial fibrillation; disease progression; heart failure; life; software.
© 2017 American Heart Association, Inc.