Derivative-free optimization of combinatorial problems - A case study in colorectal cancer screening

摘要

In the US, colorectal cancer (CRC) is a significant burden on society as the 2nd most deadly cancer. This burden can be mitigated by early detection or prevention via screening asymptomatic individuals using a screening strategy. The progression of CRC is not known with certainty, exhibiting a challenge for determining an optimal screening strategy for populations. A microsimulation model is utilized to incorporate this uncertainty within a population to estimate the benefits of a given screening strategy. The optimization problem for determining CRC screening strategies is formulated as a combinatorial problem, a challenging problem type for derivative-free optimization (DFO) solvers. We assess ten DFO solvers' ability to handle combinatorial problems using a test problem. Then, a simulation-optimization approach is used to determine the optimal strategy for the population. The best-identified screening strategies were shown to reduce the societal impact of CRC more so than the currently recommended screening strategies.