Comparing the Cost-Effectiveness of Innovative Colorectal Cancer Screening Tests

Elisabeth F. P. Peterse, PhD; Reinier G. S. Meester, PhD; Lucie de Jonge, MSc; Amir-Houshang Omidvari, MD; Fernando Alarid-Escudero, PhD; Amy B. Knudsen, PhD; Ann G. Zauber, PhD; Iris Lansdorp-Vogelaar, PhD


J Natl Cancer Inst. 2021;113(2):154-161. 

In This Article



The MISCAN-Colon model was developed by the Department of Public Health within Erasmus University Medical Center, Rotterdam, the Netherlands, and has been described in detail elsewhere.[17,18] It is part of the US National Cancer Institute's Cancer Intervention and Surveillance Modeling Network[19] and has been used to inform screening recommendations.[20–22] In brief, the model generates, with random variation, a large population similar to the US population in terms of life expectancy and CRC risk. As each simulated person ages, 1 or more adenomas may develop, which can progress in size and can develop into preclinical cancer (stages I-IV). During each stage, CRC may be diagnosed because of symptoms. Screening can alter some simulated life histories, because CRC can be prevented or diagnosed at an earlier stage. Screening may also result in complications, overdiagnosis, and overtreatment, which are also taken into account by the model.

Screening Strategies

We simulated screening from age 50 years through 75 years in an average-risk population, with perfect adherence to screening, diagnostic follow-up, and surveillance recommendations.[4,23] We used the same model assumptions as for the 2018 ACS guidelines, which account for recent trends in CRC incidence.[21,24] The screening strategies evaluated were CTC every 5 years, mtSDNA testing every 1 or 3 years,[4] PillCam every 5 or 10 years, and annual or biennial mSEPT9 testing. These alternative screening strategies were compared with colonoscopy every 10 years and annual FIT. Positive noncolonoscopy tests were followed by a diagnostic colonoscopy, and individuals in whom adenomas were detected and removed received colonoscopy surveillance through age 85 years.[23]

To compare the different screening strategies, an incremental cost-effectiveness analysis was performed, ranking strategies based on costs. Strategies that were more costly and less effective than a (combination of) other strategies were considered dominated. The remaining strategies provided good value for money (ie, were efficient). For the efficient strategies, the incremental cost-effectiveness ratios (ICERs) were obtained by dividing the additional costs by the additional quality-adjusted life-years gained (QALYG) compared with the next less costly alternative strategy. In this analysis, we assumed a willingness-to-pay threshold of $100 000 per QALYG.[25,26]

Test Characteristics

mSEPT9 performance characteristics were based on Potter et al.[27] (Table 1; Supplementary Table 1, available online), which was used for the FDA approval of mSEPT9.[33] In this study, 1544 samples were retrospectively selected from the PRospective Evaluation of SEPTin 9 (PRESEPT) trial.[34] CRC sensitivity and specificity of 68.2% and 78.8% were reported, respectively, with a sensitivity for advanced adenoma of 21.6%. PillCam characteristics were based on the study of Rex et al.[30] in which 695 asymptomatic individuals were successfully screened using the PillCam, followed by colonoscopy several weeks later. This study reported a sensitivity of 92% and 91% for adenomas larger than 10 mm and 6 mm, respectively, with a specificity of 83%.[30] Colonoscopy, FIT, CTC, and mtSDNA characteristics were similar to previous analyses from our group[20,21] (Table 1). All test characteristics were varied in probabilistic sensitivity analyses (see below).

Costs and Disutilities

Costs of screening, screening-related complications, and cancer care were computed from a societal perspective, obtained from various sources, and included (as relevant) payments, coinsurance, cathartic bowel preparation agents, and patient and escort time costs (Table 2; Supplementary Tables 2–6, available online). Costs were updated to 2017 US dollars using the Personal Health Care Deflator Price Index. Estimated test disutilities included those associated with the test itself and those related to fear or anxiety while waiting for the test result or a follow-up colonoscopy after a positive result (Supplementary Table 4, available online). Complication and CRC care disutilities were in line with previous analyses.[16,35]

Scenario Analyses

We repeated analyses under several alternative scenarios. In the first scenario, we evaluated CRC screening from age 45 years instead of 50 years, in line with the most recent ACS screening guideline.[5] In the second scenario, we used the version of MISCAN-Colon that was used to inform the 2016 US Preventive Services Task Force CRC screening recommendations, with CRC incidence based on 1975–1979 data[20] instead of more recent data. In the third scenario, we accounted for suboptimal adherence to diagnostic and surveillance colonoscopy and for decreasing adherence over multiple screening rounds.[36] For this scenario, we assumed a 100% adherence at the first screening and that 90% of the people screened at a given age would participate again at the next recommended age.[37,38] In line with current CRC participation rates,[39] we assumed screening adherence would not drop below 60% at any age by assuming that 15% of the people who previously did not participate would participate at the next recommended age. We further assumed 80% adherence to diagnostic and surveillance colonoscopy.[40,41] Finally, we evaluated a scenario in which 12% of the advanced adenomas and 18% of CRCs were systematically missed by the mSEPT9 due to no methylation of the SEPT9 gene promoter.[42]

Probabilistic Sensitivity Analyses

To evaluate the model parameter uncertainty, a probabilistic sensitivity analysis was performed, varying the characteristics, costs, and disutilities of all screening tests as well as the costs and disutilities of CRC treatment and colonoscopy complications (Supplementary Tables 7 and 8, available online). For every evaluated screening strategy, we performed 1000 simulation runs of 10 million persons in which we sampled parameters values from distributions that reflect the parameter's current level of evidence (Supplementary Tables 7 and 8, available online). The results of the probabilistic sensitivity analysis were displayed with cost-effectiveness acceptability curves and a frontier representing the proportion that each strategy is cost-effective and the strategy with the highest expected net monetary benefit at each cost-effectiveness threshold, respectively.[43] Results were analyzed using R with the package BCEA.[44,45]