Joint nested frailty models for clustered recurrent and terminal events: An application to colonoscopy screening visits and colorectal cancer risks in Lynch Syndrome families

• Published in: Statistical Methods in Medical Research
• Main Authors: CHOI, Y. H., Jacqmin-Gadda, Helene, Krol, A., PARFREY, P., BRIOLLAIS, L., Rondeau, Virginie
• Other Authors: Bordeaux population health (BPH), Université de Bordeaux (UB)-Institut de Santé Publique, d'Épidémiologie et de Développement (ISPED)-Institut National de la Santé et de la Recherche Médicale (INSERM)
• Format: Article in Journal/Newspaper
• Language: English
• Published: HAL CCSD 2020
• Subjects: Biostatistics; EPICENE; [SDV.SPEE]Life Sciences [q-bio]/Santé publique et épidémiologie; Lynch; Newfoundland
• Online Access: https://hal.archives-ouvertes.fr/hal-03116171; https://doi.org/10.1177/0962280219863076

Joint models for recurrent and terminal events have not been yet developed for clustered data. The goals of our study are to develop a statistical framework for modelling clustered recurrent and terminal events and to perform dynamic predictions of the terminal event in family studies. We propose a joint nested frailty model for colonoscopy screening visits and colorectal cancer onset in Lynch Syndrome families. The screening and disease processes could each depend on individuals' screening history and other measured covariates and be correlated within families; our approach allows for familial correlations to affect both the visit process and the terminal event and the dependence between the two processes is specified through frailty distributions. We provide dynamic predictions of colorectal cancer risk for an individual conditional on his/her own screening history, his/her family history of screening and disease and other important clinical covariates. We apply our model to 18 Lynch Syndrome families from Newfoundland for individualized dynamic predictions of colorectal cancer risks. We demonstrate that the screening visits are non-ignorable for estimating the disease risks, and the joint nested frailty model improves dynamic prediction accuracies compared to existing joint frailty models after accounting for familial and individual screening and cancer histories.