| Colorectal cancer is a disease that is very treatable if caught early. Colorectal cancer is caught early when individuals are screened for cancer using effective tests. However, the choice of whether or not to screen is an individual decision. The outcome of this decision may directly affect an individual's long-term survival and health outcome. Public health researchers have recognized that the best way to improve health outcomes for colorectal cancer is to encourage screening by educating individuals. However, observational studies do not follow individuals for a sufficiently long term to see the downstream results of behavioral changes. Because long term studies cannot be conducted in a cost effective manner, researchers have turned to simulation modeling to create detailed models of disease so as to "observe" many individuals for their entire life course. In this dissertation, we develop an approach for optimizing public health interventions within a limited health budget aimed at improving screening behavior. To do this, we first use a generalized logistic response curve to model individuals' behavioral responses to public health programs and incorporate it into a detailed computer simulation model of colorectal cancer progression. To make this model efficient at simulating health gains from behavior changes generated by public health policies, we extend the concept of Common Random Numbers to create a "Common Patient" that is modeled in our aggregate simulation model. We then formulate two different policy design questions as mathematical programing problems and solve these problems using our simulation model. The first question addressed is focused on the allocation of spending across multiple decision periods in individuals' age 50 to 75 screening horizon. A response surface approach, using kriging to represent the response surface, is proposed as an efficient way of solving the simulation optimization problem. A public health case study using this model suggests it is better to focus public health spending later in an individual's age 50-75 screening horizon. The second policy decision problem modeled is how to allocate spending between multiple interventions. These interventions are described by what populations or test attributes they affect. Using an analytic method, we develop solutions of how budgets should be allocated between different intervention options given individuals response to that spending. Based on the optimization of a quasiconcave objective function, the analytic approach provides easy and intuitive solutions to practical problems. All of these techniques are applied in the context of evaluating public health interventions proposed for use in the State of North Carolina. |
