Robust estimation of state occupancy probabilities for interval-censored multistate data: an application involving spondylitis in psoriatic arthritis

We formulate a three-state illness-death model to estimate the proportion of psoriatic arthritis patients developing spondylitis over time. Data from a longitudinal cohort of patients are available but the transitions in this model are interval-censored for the onset of spondylitis; times of deaths are right-censored. Robust methods for estimating the prevalence of spondylitis over time are described based on differences in marginal survivor functions for state entry times in the spirit of Pepe et al. (1991). Nonparametric estimates (Turnbull, 1976) and local likelihood estimates (Loader, 1999) of the marginal distributions are derived. Multiplicative intensity Markov regression models are used to examine covariate effects