Aller au contenu principal


Challenges in longitudinal and time-to-event data analysis: new numerical approaches for robust, stable and accurate joint model estimation

The present proposal relates to methodological and computational developments for longitudinal and time-to event data analysis.

Période du projet :
Investigateur principal :

Roch Giorgi, Aline Campos Reis de Souza

Membre(s) SESSTIM du projet :
Commanditaires :

Aix-Marseille Université | France 2030 | Fondation A*Midex.


After the diagnosis of a disease, a key question in clinical practice is accurate prediction of patient prognosis. In this context, there is an increasing interest in joint models for longitudinal and time-to-event data due to their ability of updating estimates of conditional survival probabilities as additional longitudinal information (such as individual biomarkers measurements, possibly obtained from digital health) are recorded. The main drawback of this modelling approach, however, is of a practical nature: complex model specifications are still difficult to implement and computationally expensive to estimate, limiting their use for real time analysis. As a consequence, the usage of joint models for prognostic purposes can be burdensome.

Méthode :

A stochastic optimization algorithm will be adapted for the joint modelling framework to increase the processing and storage capacity of maximum likelihood estimation. Matrix corrections will be proposed to reduce numerical instabilities during the estimation in order to contribute with the automation of the process. These advances in the state-of-art will result in a robust and easy to implement methodology for longitudinal and time-to event data analysis that can reach the non-academic health sector, providing innovative digital solutions in personalised medicine and disease prevention.