Risk-Based Simulation Optimization of PSA-Based Prostate Cancer Screening

Prostate cancer (PCa) is a serious chronic disease affecting a large number of men and is the second leading cause of men's cancer deaths in the United States. We present a screening simulation model composed of the following: (i) a PCa natural-history submodel based on a discrete-event stochastic process representing a patient's progression through underlying health states over his lifetime; and (ii) a statistical change-point submodel representing a patient's prostate-specific antigen (PSA) level over time. Using specific risk-based parameterizations of this screening simulation model, we search for improved PSA-based PCa screening strategies for certain well-known risk groups based on race (white and African American), family history of PCa, and different levels of comorbid medical conditions.;We first demonstrate how the careful use of common random numbers (synchronized patient histories) allows for more precise estimation of NNS, the expected number of patients needed to be screened in order to prevent 1 death from PCa. We validate the simulation model by comparing model estimates of NNS and other statistics with corresponding estimates from the literature. By comparing 14 strategies from the literature, we found that the strategy of screening annually from age 50 to 75 using the PSA threshold 2.5 ng/mL yielded the smallest estimated NNS.;Next, we present a quality-adjusted life-years (QALYs) parameterization of the simulation model that uses synchronized patient histories to estimate for a given screening strategy the expected QALYs gained (QG) by a patient relative to no screening. This development naturally leads to the statistic NNSQ, the expected number of patients needed to be screened to produce a net gain of 1 QALY in the screened population. We formulate an optimization model to find improved screening strategies with QG as the performance measure. Based on results from this model, we discuss how NNSQ (the reciprocal of QG) can be helpful to policy makers as an alternative to NNS, or as an auxiliary performance measure for evaluation of screening policies.;We develop a three-stage metaheuristic simulation-optimization method based on the combination of a global genetic algorithm (GA), a post-GA clean-up procedure, and an implementation of the COMPASS local search method. The composite global and local metaheuristic is designed to find strategies that yield large expected values of QG while also exhibiting a smoothness in the PSA-threshold time series that render the strategies more suited to actual clinical implementation. We use this method to search for improved strategies in each risk group based on maximizing estimated expected QG. In addition to this metaheuristic, we develop an analytical formulation of a simplified, single-period PCa screening model. Results from both the metaheuristic and the analytical approach suggested that PSA is not beneficial for deciding whether men should have a biopsy. Moreover, our results suggest that for all 5 risk groups a routine biopsy between the ages of 45 and 60, regardless of PSA level, maximizes expected QG.