Discrete-time semi-Markov modeling of human papillomavirus persistence

Multi-state modeling is often employed to describe the progression of a disease process. In epidemiological studies of certain diseases, the disease state is typically only observed at periodic clinical visits, producing incomplete longitudinal data. In this paper we consider fitting semi-Markov models to estimate the persistence of human papillomavirus (HPV) type-specific infection in studies where the status of HPV type(s) is assessed periodically. Simulation study results are presented indicating that the semi-Markov estimator is more accurate than an estimator currently used in the HPV literature. The methods are illustrated using data from the HIV Epidemiology Research Study.