Séminaire de Santé publique
Vendredi 23 janvier 2026 - 12h15 à 13h15
Amphi Louis - ISPED
Campus Carreire - université de Bordeaux
Ouvert à tous
En présentiel et visioconférence
Title: Insights from joint models of tumor size evolution and survival in cancer clinical trials
Denis Rustand, CR Inserm,
Bordeaux Population Health (BPH), Inserm U1219
Equipe Biostatistique
Bio:
Denis Rustand joined the BPH-BIOSTAT team in July 2025 as an Inserm Research Scientist. He holds a PhD in Biostatistics and Public Health, defended in 2020 at the University of Bordeaux, on the development of joint models to evaluate therapeutic choices in cancer clinical trials, under the supervision of Dr. Virginie Rondeau (BPH-BIOSTAT Bordeaux) and Dr. Laurent Briollais (University of Toronto). He completed a four-year postdoctoral fellowship in Prof. Håvard Rue's team at KAUST in Saudi Arabia. In 2021, he was awarded the Norbert Marx Thesis Prize, a distinction presented every two years by the French Statistical Society.
Abstract:
In solid tumor randomized clinical trials, treatment evaluation heavily relies on RECIST guidelines, which reduce complex longitudinal tumor dynamics into categorical outcomes. While joint models have been proposed to better exploit longitudinal tumor burden data, standard approaches often rely on simplified assumptions which may not reflect the biological reality of treatment response.
In this seminar, I will present a framework based on recent developments in joint modeling to provide a more precise characterization of the disease process. We propose a model that captures the non-linear dynamics of tumor shrinkage and regrowth. We will demonstrate two key benefits of this framework. First, by accounting for inter-individual heterogeneity through Shared Random Effects (SRE), joint models offer a more precise estimation of the treatment effect compared to standard Cox models. Second, by using a Current Value (CV) association structure, we allow for the estimation of time-dependent treatment effects and the decomposition of the treatment effect into direct (survival-specific) and indirect (tumor-mediated) components.
We illustrate the clinical relevance of this framework through the analysis of a phase III clinical trial in metastatic colorectal cancer. The results highlight the gain in precision achieved with the SRE approach and reveal, through the CV approach, time-dependent hazard ratios that provide a better view of treatment efficacy often missed by standard methods. Finally, we will discuss the robustness of these findings based on simulation studies, which quantify the bias introduced by standard approaches when longitudinal tumor dynamics are ignored.
