Post-Doctoral Research Visit F/M Causal inference and mechanistic digital twins for individualized adjuvant treatment in resected lung cancer

Context
This position is funded by the
RHU LUCA-pi
project (Recherche Hospitalo-Universitaire, France 2030), dedicated to personalized medicine in early-stage non-small cell lung cancer (NSCLC). LUCA-pi brings together clinical, biological and computational partners around a large retrospective cohort (LUCA-pi RETRO) and a prospective cohort of patients treated with surgery.
The postdoc will join the
COMPO
team (COMPutational pharmacology and clinical Oncology, Inria – Inserm – Aix-Marseille Université), located at the Cancer Research Center of Marseille (CRCM). Within LUCA-pi, COMPO develops a mechanistic model of metastatic relapse — a
digital twin
— that predicts each patient's risk of relapse after surgery. The postdoc will work closely with the PhD student building this model, and with the project's clinicians.
Assignment
Adjuvant therapies (chemotherapy, immunotherapy, targeted therapy) are prescribed according to clinical criteria, so their real-world effect cannot be read directly from observational data because of confounding by indication, immortal time bias and lack of positivity. Combining causal inference with mechanistic modeling is still largely unexplored.
The goal of the postdoc is to add a causal layer to the LUCA-pi digital twin, along two lines:
Applied:
estimate the causal effect of adjuvant treatment on metastatic relapse by emulating a target trial on LUCA-pi RETRO (clone-censor-weighting; risk differences and RMST, accounting for competing risks). Average effects will be estimated with robust methods (AIPW/TMLE) and heterogeneous effects with causal survival forests and meta-learners, then compared with published randomized trials (LACE, IMpower010, ADAURA).
Methodological:
connect causal inference and mechanistic modeling with calibration of treatment-effect parameters of nonlinear mixed-effects models under confounding, hybrid estimators using the mechanistic model as outcome model, and a semi-synthetic benchmark with known counterfactuals generated by the digital twin.
Main activities
Audit and curate the clinical data (treatment dates, confounders, biomarkers).
Design and implement causal survival analyses, including sensitivity analyses for unmeasured confounding.
Develop and benchmark new methods combining mechanistic and causal approaches.
Collaborate with the PhD student and the clinicians of the LUCA-pi consortium.
Publish the results and present them at international conferences.
Skills
Strong background in causal inference (potential outcomes, propensity scores, doubly-robust estimation) and survival analysis.
Proficiency in R and/or Python.
Interest in dynamical/mechanistic models; experience with mixed-effects models is a plus.
Interest in clinical applications; good communication skills in English (French not required).