Study design and statistical methods to account for unexplained heterogeneity in medical studies with a time to event outcome

• Applicant: Professorin Antje Jahn-Eimermacher, Ph.D.
• Subject Area: Epidemiology and Medical Biometry/Statistics
• Term: from 2015 to 2019
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 276859231

In many medical studies, the evaluation of the time to some particular event is of primary interest, for example the time to disease progression. Regression models are commonly fitted to these kinds of data relying on the assumption that heterogeneity in the individuals' risk to experience events can be explained by known covariates. Nevertheless, individuals sharing the same covariate values often appear to be still heterogeneous in their event risk with the sources of heterogeneity being unknown, at least at the beginning of the study. Unobserved or unknown attributes might affect the outcome, an (unknown) part of the study population might be nonsusceptible for the event, or studies are defined on clusters of observational units that share unknown attributes affecting the outcome. Also more than one source of heterogeneity might be present. A heterogeneous study population has important implications for clinical research: First, it decreases the efficiency of medical studies and thus calls for larger sample size. Secondly, it decreases the precision of study results and thus limits their use for decision making for future patients, for example with respect to therapeutic decisions or the proper scheduling of follow-up visits. With the applied research project we aim to derive statistical designs and methods that decrease and/or properly model the patients heterogeneity in medical studies with a time to event outcome and thus will improve the (cost-)efficiency of medical studies, and the accuracy of clinical decisions that are derived from study results.Particularly, we aim to investigate the statistical estimates, statistical inference and the prediction performance of established time to event models when the source of heterogeneity is misspecified, derive methods to dynamically predict an individuals outcome in the presence of a potentially time-dependent chance of being nonsusceptible for the event (cured) and investigate flexible designs to focus a study on a more homogeneous subpopulation. All methods will be evaluated and illustrated on simulated and real study data.

• DFG Programme: Research Grants