Optimizing Colonoscopy Screening for Colorectal Cancer Prevention and Surveillance

Abstract

M illions of Americans undergo colonoscopy screening for colorectal cancer (CRC) prevention and surveillance every year. The efficiency of colonoscopy operations depends on how often patients are screened, which is a complex and controversial decision, as reflected by the discrepancy between clinical practice and guidelines. We develop a partially observable Markov decision process to optimize colonoscopy screening policies for the objective of maximizing total quality-adjusted life years. Our model incorporates age, gender, and risk of having CRC into the screening decisions and therefore provides a novel framework for personalized CRC screening. In addition to deriving the maximum attainable benefit from colonoscopy screening, which reflects the opportunity cost of following current guidelines, our results have several policy implications. Using clinical data, we show that the optimal colonoscopy screening policies may be more aggressive than the guidelines under some conditions. Optimal screening policies recommend that females with CRC history undergo colonoscopy more frequently than males. In contrast, females without CRC history should be screened less frequently than males. This result, which was not recognized before, signifies the role of gender in optimal CRC screening decisions.

Cost‐effectiveness of alternative colonoscopy surveillance strategies to mitigate metachronous colorectal cancer incidence

“…Therefore, more research is needed to determine MCRC risk factors and to model MCRC progression accordingly for deriving more risk-stratified colonoscopy surveillance strategies as advocated in the literature. 31,32 Our results show that the most cost-effective alternative strategy (eg, the one with the greatest health outcome improvements and an ICER closest to $50,000/LY) depends on the cumulative MCRC incidence. Therefore, we can theoretically identify the most cost-effective strategy as the one with the best health outcomes among the strategies with ICERs < $50,000/LY for all MCRC incidence scenarios (ie, one of the promising alternative strategies).…”

“…The following conclusions should be noted: 1) the current US colonoscopy surveillance guideline may not necessarily be the best practice, 2) more intensive surveillance within the first 6 to 8 years (eg, reducing the US guideline's surveillance intervals by 1 to 2 years after the second negative surveillance colonoscopy) may improve health outcomes while moderately increasing costs, and 3) the set of Pareto‐efficient alternative strategies varies with the aggressiveness of MCRC progression. Therefore, more research is needed to determine MCRC risk factors and to model MCRC progression accordingly for deriving more risk‐stratified colonoscopy surveillance strategies as advocated in the literature …”

Operational research techniques applied throughout cancer care services: a review

“…Hosking, Roberts, Uzsoy, and Joseph (2013), Erenay, Alagoz, and Said (2014) and Li, Zhu, Klein, and Kong (2014) focus on colonoscopy only screening strategies, while Li, Dong, Ren, and Yin (2015), Song and Wang (2016) and Campbell, Blake, Kephart, Grunfeld, and MacIntosh (2017) consider two-step strategies that follow a noninvasive test with a colonoscopy for those patients that need it. The majority of these papers compare screening strategies (Li et al, 2014;Song & Wang, 2016) or optimise the yearly screening decision Li et al, 2015).…”

Optimally solving Markov decision processes with total expected discounted reward function: Linear programming revisited

“…We also solve a real-life MDP problem adapted from a partially observable MDP model for optimizing colorectal screening [19], in which the objective is to determine the optimal colonoscopy decisions for early diagnosis of colorectal cancer. The parameters for this problem are estimated from a clinical database, available epidemiological data, and the literature.…”

“…The parameters for this problem are estimated from a clinical database, available epidemiological data, and the literature. More information about the model and parameter estimation is available elsewhere [19]. The MDP model includes 40, 403 states and 2 actions.…”