Screening Based on the Risk of Cancer Calculation From Bayesian Hierarchical Changepoint and Mixture Models of Longitudinal Markers

The standard approach to early detection of disease with a quantitative marker is to set a population-based fixed reference level for making further individual screening or referral decisions. For many types of disease, including prostate and ovarian cancer, additional information is contained in the subject-specific temporal behavior of the marker, which exhibits a characteristic alteration early in the course of the disease. In this article we derive a Bayesian approach to screening based on calculation of the posterior probability of disease given longitudinal marker levels. The method is motivated by a randomized ovarian cancer screening trial in the United Kingdom comprising 22,000 women screened over 4 years with an additional 5 years of follow-up on average. Levels of the antigen CA125 were recorded annually in the screened arm. CA125 profiles of cases and controls from the U.K. trial are modeled using hierarchical changepoint and mixture models, posterior distributions are calculated using Markov chain Monte Carlo methods, and the model is used to calculate the Bayesian posterior risk of having ovarian cancer given a new subject's single or multiple longitudinal CA125 levels. A screening strategy based on the risk calculation is then evaluated using data from an independent screening trial of 5,550 women performed in Sweden. A longitudinal CA125 screening strategy based on calculation of the risk of ovarian cancer is proposed. Simulations of a prospective trial using a strategy based on the risk calculated from longitudinal CA125 values indicate potentially large increases in sensitivity for a given specificity compared to the standard approach based on a fixed CA125 reference level for all subjects.