B1647
Title: Assessing patient benefit with semi-Markov multi-state models
Authors: Abdul Haris Jameel - University of Nottingham (United Kingdom) [presenting]
Christopher Fallaize - University of Nottingham (United Kingdom)
Joachim Grevel - BAST Inc Limited (United Kingdom)
Blesson Chacko - BAST Inc Limited (United Kingdom)
Christopher Brignell - University of Nottingham (United Kingdom)
Gilles Stupfler - University of Angers (France)
Abstract: In the context of oncological drug trials, the Cox proportional hazards model is traditionally used to establish treatment efficacy. However, a treatment which causes a high rate of premature discontinuation due to adverse side effects could be considered clinically effective by the Cox model, despite being too toxic for most patients to tolerate. A patient-focused modelling approach would instead seek to answer whether a particular treatment can treat cancer while being sufficiently tolerable for patients. Such information is crucial for patients, their doctors, and caretakers to assess and make better-informed decisions about optimally using the patients' limited remaining lifespans. Multi-state models are well known and can alternatively be used to model the entire event history of patients in drug trials. Furthermore, assuming an underlying semi-Markov process allows for arbitrary distributions of holding times before transitioning to other states. This allows for an alternative set of tools to quantify the various effects of a treatment on patients, and thus whether patients can potentially benefit. The main focus is on the methods used to formally define and quantify the patient benefit, namely the survival function of the holding time in a given state. A proposed hypothesis test is also discussed and validated.
