Thèse « Optimisation des Techniques d'Inférence pour les Métaanalyses sur Données Individuelles de Patients Portant sur des Critères de Survie avec une Application au Carcinome du Nasopharynx H/F - Doctorat.Gouv.Fr

Établissement : Université Paris-Saclay GS Santé publique École doctorale : Santé Publique Laboratoire de recherche : Centre de Recherche en épidémiologie et Santé des populations Direction de la thèse : Stefan MICHIELS ORCID 0000000269632968 Début de la thèse : 2026-10-01 Date limite de candidature : 2026-05-08T23:59:59 Fast and well informed decision making in oncology clinical trials often relies on early outcomes or on different types of endpoints that may represent competing events. Furthermore, because clinical trials frequently involve limited sample sizes, it can be critical to leverage information from multiple outcomes or complementary data sources. In this context, the meta analysis toolbox can help strengthen the level of evidence, both in the overall study population and across clinically relevant subgroups.
The Oncostat team has previously conducted an individual patient data meta analysis to evaluate the optimal timing and combination of radiotherapy and chemotherapy in locally advanced nasopharyngeal carcinoma (Blanchard et al., 2015; 2021; Petit, 2023), a cancer highly prevalent in South East Asia. In this meta analysis, adding chemotherapy to standard radiotherapy was associated with a significant improvement in the composite endpoint progression free survival, defined as the time from randomization to the first event, either locoregional or distant progression, or death from any cause.
Several methodological questions remain open, including how to optimally model the relationship between treatment effects on early endpoints and on long term survival, how to characterize the natural disease behaviour in terms of competing risks across clinically relevant patient subgroups, and how to combine clinical trial evidence with real world data (Yao et al., JCE 2025).
Fast and well informed decision making in oncology clinical trials often relies on early outcomes or on different types of endpoints that may represent competing events. Furthermore, because clinical trials frequently involve limited sample sizes, it can be critical to leverage information from multiple outcomes or complementary data sources. In this context, the meta analysis toolbox can help strengthen the level of evidence, both in the overall study population and across clinically relevant subgroups.
The Oncostat team has previously conducted an individual patient data meta analysis to evaluate the optimal timing and combination of radiotherapy and chemotherapy in locally advanced nasopharyngeal carcinoma (Blanchard et al., 2015; 2021; Petit, 2023), a cancer highly prevalent in South East Asia. In this meta analysis, adding chemotherapy to standard radiotherapy was associated with a significant improvement in the composite endpoint progression free survival, defined as the time from randomization to the first event, either locoregional or distant progression, or death from any cause.
Several methodological questions remain open, including how to optimally model the relationship between treatment effects on early endpoints and on long term survival, how to characterize the natural disease behaviour in terms of competing risks across clinically relevant patient subgroups, and how to combine clinical trial evidence with real world data (Yao et al., JCE 2025).
This PhD project aims to enhance decision-making through three key objectives:
1 Using pseudovalues in surrogate endpoint validation in an individual patient data based meta-analysis
2 Meta-Analysis of Competing Outcomes for Causal Inference
3 Meta-analytic combination of clinical trials and real world data 1. Using pseudovalues for surrogate endpoint validation in an individual patient data meta analysis
While gold standard oncology endpoints such as overall survival are time to event outcomes, many early candidate endpoints (e.g., disease free survival, progression free survival, time to ctDNA clearance) are also censored time to event variables. Surrogate endpoint evaluation is therefore methodologically challenging due to this double time to event structure, typically requiring complex approaches such as copula models, joint models (Rotolo et al., 2019), or mediation based methods (Le Coënt et al., 2023).
The first objective is to investigate the use of the pseudovalue framework to simplify meta analytic surrogacy modelling for time to event endpoints, by transforming the problem into a continuous-continuous setting. Recent methodological work has suggested pseudovalues in a single trial context with a binary surrogate (Chernofsky Lod, 2025; Ocampa et al., 2024), but their extension to meta analysis and double time to event surrogates remains unexplored.
A simulation study will compare this approach with state of the art methods under various data generating mechanisms. The surrogacy of locoregional relapse free survival, distant metastasis free survival, and progression free survival for overall survival will then be evaluated in the Chemotherapy in Nasopharyngeal Carcinoma individual patient data meta analysis and compared with previous findings (Rotolo et al., 2017).
2. Meta analysis of competing outcomes for causal inference
In randomized clinical trials included in meta analyses, treatment effects are often estimated on competing endpoints (e.g., cancer specific vs. non cancer mortality; locoregional failure vs. distant failure vs. death), requiring a competing risks framework (Meddis et al., 2020). The second objective of the PhD is to extend network meta analytic methods for competing risks toward causal estimands, using: the restricted mean time lost (RMTL) statistic (Conner & Trinquart, 2021), and g estimation approaches (Young et al., 2020), while adjusting for clinical covariates. An additional aim is to explore subgroup specific treatment effects, particularly across age groups and other clinically relevant covariates, in the Nasopharyngeal Carcinoma meta analysis. A simulation study will assess bias and variance of treatment effects under various scenarios.

3. Meta analytic combination of clinical trials and real world data (RWD)
Beyond the limited number of randomized trials typically available in oncology, real world data on control treatments can provide valuable complementary information. The third objective is to extend existing causal inference methods combining trial data and RWD (e.g., Wei et al., 2024 or Zhang, 2026) to: the meta analysis setting, the network meta analysis setting (Ollier, Stat Med, 2022), both for single time to event endpoints and for the competing risk framework developed in Objective 2. A particular focus will be placed on handling covariate distribution differences between trials and RWD sources.
The Oncostat individual patient data meta analysis in locally advanced nasopharyngeal carcinoma will serve as a case study to illustrate and apply these methodological developments in collaboration with Prof P. Blanchard.

master degree in biostatistics or statistics. Interest in statistical methods for clinical trials