gms | German Medical Science

Background: The idea of this dataset challenge is that a variety of researchers introduce different approaches for the analysis of time-to-event data applied on the same example-dataset. Ideally, this results in interesting discussions and shows the variety of alternative analysis methods for a more adequate description of time-to-event data. We aim to shed light on the large research field of time-to-event analysis and to set the basis to establish maybe more adequate types of analysis for capturing disease processes holistically.

Methods: All analyses that will be shown are based on follow-up data of the Interdisciplinary Network for Heart Failure (INH) study [1]. This multi-centre randomized controlled trial investigates the efficacy of a nurse-coordinated disease management program (HNC) in heart failure compared to usual care for patients that were first hospitalized for systolic heart failure. In the original study a composite of time to death or time to first rehospitalization was considered as primary endpoint. However, for this extended study it was up to the researchers to use all available information on rehospitalizations. A total of 1022 patients (513 in usual care, 509 in HNC group) with 663 deaths and 3016 rehospitalizations will be included in the analyzed dataset. The introduced analyses methods cover inter alia a joint frailty model for recurrent events with an associated terminal event for which SAS code by Toenges and Jahn-Eimermacher is used [2]. In addition, we will present a parametric, hazard-free model that generates the semi-competing risk model as described by Xu et al. [3] to the situation with recurrent events. Semi-and non-parametric methods like Cox or Aalen's model and Nelson-Aalen or Aalen-Johansen estimator are applied as well. Furthermore, a multi-state modelling framework [4] is proposed, which includes the model of competing risks and the illness-death model. Finally, attention is paid to the multi-centre structure of the data (for further details see the abstract by Unkel et al. entitled “Multistate modelling of time-to-event data from a multi-centre trial in patients with heart failure”).

Results: The introduced methods cover different aspects of the given data and therefore might lead to different results. However, we believe that they all are still more adequate in describing disease processes than the common time-to-first event analysis.

Conclusion: There is a huge variety of approaches that should be considered in future research for a more adequate analysis of complex time-to-event data, especially with recurrent events.

The authors declare that they have no competing interests.

The authors declare that a positive ethics committee vote has been obtained.