Public Thesis defense - LIDAM

Time-dependent ROC Curve Estimation and Inference for Censored Data : Some Novel Contributions by Kassu Mehari BEYENE

Pour l’obtention du grade de Docteur en sciences

In recent years, prediction models have become increasingly popular tools to estimate the risk of a person developing a specific event of interest at a particular time, given his/her characteristics. Accuracy of these predictive models is critical as it determines the quality of their predictions that form scientific evidence for informing treatment or other clinical decisions for individual patients. Therefore, nowadays, it is widely accepted that assessing the predictive accuracy of predictive models is a critical step before make use of them for a clinical practice. The ability of the models to distinguish between two classes (e.g., case and control subjects) is one of the most important criteria to evaluate predictive accuracy. To this end, the receiver operating characteristic (ROC) curve and the associated summary indices are the most widely used tools to evaluate the classification accuracy of the model. There exist various methods to estimate the time-dependent ROC curve and the associated summary indices. However, there is a lack of literature that addresses issues such as the presence of cure fraction, interval censoring, and smoothing the ROC curve. This thesis fills these gaps by introducing some new estimation and inference methods for time-dependent ROC curves and associated summary measures. We first investigate the validity of a non-parametric time-dependent area under the ROC curve (AUC) estimator of the classical right-censored data in the presence of cure fraction and, afterward, we propose some non-parametric time-dependent ROC curve and the associated summary measures estimation and inference methods for the right-censored data. Finally, we introduce extensions of these methods for the general case of mixed data that may involve exact, left-censored, right-censored, and interval-censored observations.

Jury members :

• Prof. Anouar El Ghouch (UCLouvain), supervisor
• Prof. Catherine Legrand (UCLouvain), chairperson
• Prof. Ingrid Van Keilegom (UCLouvain), secretary
• Prof. Abderrahim Oulhaj (UAE University, United Arab Emirates)
• Prof. Juan Carlos Pardo-Fernández (University of Vigo, Spain)

Télécharger l'annonce