Treatment effect measures for recurrent event endpoints with and without presence of terminal events

Invitation to PhD defense

 Julie Kjærulff Furberg

Treatment effect measures for recurrent event endpoints with and without presence of terminal events

Date              Wednesday October 4, 2023, at 14.00

Venue           Øster Farimagsgade 5, 1353 Copenhagen K, room 35.01.44 (CSS)

The defense will be followed by a reception in room 5.2.46 (CSS)

Academic Advisors

Professor Per Kragh Andersen

Section of Biostatistics, Department of Public Health, University of Copenhagen, Denmark

Dr Henrik Ravn

Biostatistics OSCD & Outcomes 1 and Innovative Medical Evidence Generation, Novo Nordisk, Denmark

Trine Saugstrup

Programme Management, Novo Nordisk, Denmark

Assessment Committee

Associate professor Frank Eriksson (chair)

Section of Biostatistics, Department of Public Health, University of Copenhagen, Denmark

Dr Susanne Rosthøj

Statistics and Data Analysis, Kræftens Bekæmpelse (Danish Cancer Society), Denmark

 Professor Richard J. Cook

Department of Statistics and Actuarial Science (SAS), Mathematics 3 (M3), University of Waterloo, Canada

Summary

This thesis focuses on statistical methodology and its application to recurrent event data with or without the presence of terminal events. The research concerns the complex task of quantifying a relevant treatment effect, estimand, from recurrent event data. The complexity is increased by the presence of terminal events. Novo Nordisk is conducting several large, randomized trials which collect data that may be subject to right censoring and high mortality rates. Within recent years, there has been a big desire to perform inference on treatment effects for recurrent events from such large trials. This task, however, is not trivial. Moreover, no industrial nor regulatory preferences are in place for the analysis of recurrent events. Guidance and good practice are particularly lacking when analyzing recurrent events which may be influenced by terminal events.

The thesis discusses and characterizes current recurrent event methods and their ability to describe clinically relevant treatment effects from randomized trials. It is recommended to use marginal models as opposed to models based on intensities. The thesis proposes a bivariate method based on pseudo-observations to characterize treatment effects on recurrent events and mortality simultaneously. This idea can be generalized to perform joint inference on treatment effects with many domains of interest.

Further, the thesis suggests a simulation-based procedure for conducting sample size estimation using marginal models for future trials with confirmatory recurrent event endpoints and competing terminal events. Publicly available software is provided for several parts of the thesis. Finally, the thesis discusses how to perform inference on recurrent events using marginal models under dependent censoring. This can be useful as sensitivity analyses for recurrent events. The research contributions are both methodological and practical. Overall, the thesis discusses and suggests how to plan, analyze, and report recurrent event endpoints from randomized trials.

The work was funded by Innovation Fund Denmark and Novo Nordisk.