STatistical Methods for Precision Medicine (StaMPM)

The research project is build around the 3 axis

Axis 1) The development of a framework for the validation of prognostic and predictive biomarker

This axis will benefit from the interaction with Allory and Bidard’ Team.

Project 1: Our work aims to improve the reproducibility of the evaluation of the impact of biomarkers on the therapeutic decision and its consequences. Finding the optimal threshold for a continuous biomarker at a clinically usefull time horizon is an essential step for guiding treatment and possibility avoiding over-treatment. To this end, the Phd of B. Mboup aims at providing reproducible methods to estimate this threshold either from clinical trial data or observational data.

An crucial assumption of risk prediction model used to estimate risk distribution conditional on the biomarker and the treatment is the perfect calibration assumption (observed and estimated  risks are in close agreement). This assumption is not tenable in practice and no useful nor efficient test of calibration are yet available. We thus proposed to address this issue by developing inference for simulating the calibration assumption employing constrained nonparametric maximum likelihood.

Project 2: To be able to rely on a biomarker, clinicians want to know to which extend a biomarker can help predicting the outcome of a patient. Common strategy is to ignore heterogeneity in the phase of evaluation of the performance of a promising biomarker, but to confirm its discriminatory capacity it is of prime importance to account for heterogeneity while adjusting for clinical covariates. A. Meddis’s Phd aims at developing methods for addressing these issues. Notably, estimation of time-dependent ROC curves with clustered failure times, and adjusting for relevant covariates. Since covariates can influence both the clinical endpoint and the biomarker under study, we have to qualify the magnitude of the confounding in order to guide which adjusted ROC curve should be estimated. In fact assessing the amount of confounding is important for determining whether or not a specific factor should adjusted in the analysis.

To illustrate the relevance of the proposed methods we are currently investigating the incremental value of patient specifc outputs derived from boolean model prediction, as compared toa purely clinical pronostic model (J. Beal’s Phd).

Axis 2) The analysis and modelisation of response to treatment

Project 1  COMBIMMUNO : In collaboration with F. Reyal Team we plan to decipher the relationships between co-morbidities, comedications, immune infiltration, response to treatment and toxicities, as well as breast cancer outcomes in a very large dataset of breast cancer patients. Several independent cohorts from international or national or real life cohorts will be aggregated. A robust statistical methodology will be developed to take into account both confounding factors and the aggregation of heterogeneous data (Causal inference from Targeted Maximum Likehood Estimator). A large integrative analysis will be analyzed on a pooled analysis of nearly 50000 BC patients.

This project is funded by InCA SHS 2018 Call

Project 2: In recent years, immune checkpoint inhibitors have been shown to be effective for overall survival in several tumor sites. However, the peculiarity of this new therapeutic class, unlike cytotoxic or targeted therapies, is that its survival benefit is due to the long response time obtained in responder patients. Its delayed effect results in a late separation of the survival curves between the chemotherapy and immunotherapies arms with late-stage plateau in the arm immunotherapy related to long-term survivors estimated to be 15-25%. This late effect renders quantification of efficacy more difficult with the usual statistical methods which wrongly conclude that there is no efficiency. This project aims to propose appropriate methods for (i) a robust estimate of the antitumor response of immunotherapy (ii) the planning of future clinical trials.

Axis 3) Innovative Clinical Trials design and patients monitoring mHealth

Project 1: Early phase clinical trial incorporating either repeated biomarker (such as liquid biopsies) or dose-escalation and pharmacokinetic

Project 2: Mobile-health clinical trial

The expected impact of our team, in the precision medicine era, will range from early clinical trials to large multi centric phase 3 trials and observational studies.