Cause-specific regression model for correlated data in the presence of interval censoring

Enrico Colosimo, Universidade Federal de Minas Gerais

Co-authors: Marcio A. F. Rodrigues, Universidade Federal de Góias; Juliana V. Bastos, Universidade Federal de Minas Gerais;  Sylvia C. Coste, Universidade Federal de Minas Gerais

Abstract: The present work was motivated by a longitudinal study conducted at the Dental Trauma Clinic at the Federal University of Minas Gerais, Brazil aiming to evaluate pulp prognosis, as well as its prognostic factors, after luxation of permanent teeth. In this study two possible mutually exclusive outcomes (competing risks) are considered, pulp vitality or pulp necrosis. The same patient may contribute with more than one tooth, forming a cluster. Finally, patients return for follow-up at regular intervals, its is only possible to know the time interval in which the event occurred. The proposed methodology adopted a GEE-type model using an independent work matrix to accommodate the presence of clusters. A Taylor series is used to approximate the log baseline hazard function in Cox proportional hazards regression. The model performance was evaluated using a simulated data set and showing good small sample properties.