sst |
Joint Modeling of Complex Longitudinal Outcomes and Time-to-Event Data : With a Focus on Quality of Life Outcomes in Cancer Clinical Trials by Hortense DOMS
Vendredi 3 avril 2026 à 15h00 - Auditoire MORE51 - Place Montesquieu, 2 - 1348 Louvain-la-Neuve
In clinical trials, longitudinal measurements and time-to-event outcomes are repeatedly collected to assess treatment efficacy and disease progression. Joint models provide a natural framework to analyse these data simultaneously, but their use becomes challenging in the presence of complex longitudinal structures. In oncology, health-related quality of life (HRQoL) outcomes are increasingly incorporated alongside survival endpoints. However, they are typically multidimensional, ordinal, and subject to informative dropout, which may lead to biased inference if not properly accounted for. This thesis aims to develop joint modelling approaches adapted to these challenges, with a particular focus on HRQoL data. The proposed work builds on latent variable models to capture the underlying structure of HRQoL outcomes and to better represent patients’ quality of life over time. In particular, we develop a joint modelling framework for the analysis of multiple longitudinal ordinal outcomes in the presence of informative and competing dropouts. The proposed model links a latent variable underlying HRQoL responses to cause-specific dropout hazards. Special attention is given to the role of dropout by explicitly incorporating information on dropout mechanisms into the longitudinal process, allowing for a more accurate assessment of the underlying latent HRQoL dimension. To accommodate more complex longitudinal models, we also investigate two-stage approaches as a computationally efficient alternative to full joint estimation. A bias-corrected strategy is proposed and applied to multidimensional latent trait joint models, enabling the joint analysis of multiple HRQoL domains while retaining substantial computational advantages. In parallel, the thesis explores more flexible specifications of the survival submodel to relax common assumptions and improve practical applicability. The methods are motivated and illustrated using data from clinical trials in patients with glioblastoma.
Jury members :
Prof. Catherine Legrand (UCLouvain) (Supervisor)
Prof. Philippe Lambert (ULiège & UCLouvain) (Supervisor)
Prof. Anouar El Ghouch (UCLouvain) (Chairperson)
Prof. Karim Barigou (UCLouvain) (Secretary)
Dr. Murielle Mauer (EORTC)
Dr. Virginie Rondeau (INSERM)
Pay attention : the public defense of Hortense DOMS will also take place in the form of a videoconference
