The mortality of ovarian cancer is higher than any other female genital malignant tumors, while there exists a strong correlation between early-stage detection and cure for it. CA125 and HE4 are two most common and effective serum markers in recent screening research of ovarian cancer. This paper derives a sequential screening strategy for ovarian cancer by jointly modeling the longitudinal profiles of CA125 and HE4. We construct a Bayesian hierarchical mixture model with changepoint, and propose two approaches for diagnosis： the risk of cancer index and the hypothesis test on the true incidence time. We simulated a 7-year sequential screening research and compared with the standard approach based on a fixed cutoff level. Our approach achieves a 15% higher sensitivity for a fixed specificity, indicating that the sequential strategy combining multiple markers is more effective in the early-stage detection of ovarian cancer.
