Personalized Decision Modeling for Intervention and Prevention of Cancer

Personalized medicine has been utilized in all stages of cancer care in recent years, including the prevention, diagnosis, treatment and follow-up. Since prevention and early intervention are particularly crucial in reducing cancer mortalities, personalizing the corresponding strategies and decisions so as to provide the most appropriate or optimal medical services for different patients can greatly improve the current cancer control practices. This dissertation research performs an in-depth exploration of personalized decision modeling of cancer intervention and prevention problems. We investigate the patient-specific screening and vaccination strategies for breast cancer and the cancers related to human papillomavirus (HPV), representatively. Three popular healthcare analytics techniques, Markov models, regression-based predictive models, and discrete-event simulation, are developed in the context of personalized cancer medicine. We discuss multiple possibilities of incorporating patient-specific risk into personalized cancer prevention strategies and showcase three practical examples. The first study builds a Markov decision process model to optimize biopsy referral decisions for women who receives abnormal breast cancer screening results. The second study directly optimizes the annual breast cancer screening using a regression-based adaptive decision model. The study also proposes a novel model selection method for logistic regression with a large number of candidate variables. The third study addresses the personalized HPV vaccination strategies and develops a hybrid model combining discrete-event simulation with regression-based risk estimation. Our findings suggest that personalized screening and vaccination benefit patients by maximizing life expectancies and minimizing the possibilities of dying from cancer. Preventive screening and vaccination programs for other cancers or diseases, which have clearly identified risk factors and measurable risk, may all benefit from patient-specific policies.