Advertisement

Evaluating key characteristics of ideal colorectal cancer screening modalities: the microsimulation approach

Open AccessPublished:February 15, 2021DOI:https://doi.org/10.1016/j.gie.2021.02.013

      Background and Aims

      Screening for colorectal cancer (CRC) can effectively reduce CRC incidence and mortality. Besides colonoscopy, tests for the detection of biomarkers in stool, blood, or serum, including the fecal immunochemical test (FIT), ColoGuard, Epi proColon, and PolypDx, have recently been advanced. We aimed to identify the characteristics of theoretic, highly efficient screening tests and calculated the effectiveness and cost effectiveness of available screening tests.

      Methods

      Using the microsimulation-based colon modeling open-source tool (CMOST), we simulated 142,501 theoretic screening tests with variable assumptions for adenoma and carcinoma sensitivity, specificity, test frequency, and adherence, and we identified highly efficient tests outperforming colonoscopy. For available screening tests, we simulated 10 replicates of a virtual population of 2 million individuals, using epidemiologic characteristics and costs assumptions of the United States.

      Results

      Highly efficient theoretic screening tests were characterized by high sensitivity for advanced adenoma and carcinoma and high patient adherence. All simulated available screening tests were effective at 100% adherence to screening and at expected real-world adherence rates. All tests were cost effective below the threshold of 100,000 U.S. dollars per life year gained. With perfect adherence, FIT was the most effective and cost-efficient intervention, whereas Epi proColon was the most effective at expected real-world adherence rates. In our sensitivity analysis, assumptions for patient adherence had the strongest impact on effectiveness of screening.

      Conclusions

      Our microsimulation study identified characteristics of highly efficient theoretic screening tests and confirmed the effectiveness and cost-effectiveness of colonoscopy and available urine-, blood-, and stool-based tests. Better patient adherence results in superior effectiveness for CRC prevention in the whole population.

      Graphical abstract

      Abbreviations:

      ADR (adenoma detection rate), CMOST (colon modeling open-source tool), CRC (colorectal cancer), dLYG (discounted life-year gained), dUSD (discounted U.S. dollars), FAST (Fourier amplitude sensitivity test), FIT (fecal immunochemical test), ICER (incremental cost-effectiveness ratio), LYGs (life-years gained)
      Colorectal cancer (CRC) is the third most common cancer in the world
      • Bray F.
      • Ferlay J.
      • Soerjomataram I.
      • et al.
      Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries.
      and causes the fourth highest number of cancer-related deaths, resulting in high economic impact.
      • Brenner H.
      • Kloor M.
      • Pox C.P.
      Colorectal cancer.
      However, CRC offers optimal opportunities for preventive efforts because malignant transformation occurs slowly via adenomatous precursors, and early CRC has much lower mortality compared with advanced disease.
      • Baxter N.N.
      • Goldwasser M.A.
      • Paszat L.F.
      • et al.
      Association of colonoscopy and death from colorectal cancer.
      • Hewitson P.
      • Glasziou P.
      • Watson E.
      • et al.
      Cochrane systematic review of colorectal cancer screening using the fecal occult blood test (hemoccult): an update.
      • Atkin W.S.
      • Edwards R.
      • Kralj-Hans I.
      • et al.
      Once-only flexible sigmoidoscopy screening in prevention of colorectal cancer: a multicentre randomised controlled trial.
      • Pan J.
      • Xin L.
      • Ma Y.F.
      • et al.
      Colonoscopy reduces colorectal cancer incidence and mortality in patients with non-malignant findings: a meta-analysis.
      Therefore, CRC screening is recommended by current gastroenterologic guidelines,
      • Rex D.K.
      • Boland C.R.
      • Dominitz J.A.
      • et al.
      Colorectal cancer screening: recommendations for physicians and patients from the U.S. Multi-Society Task Force on Colorectal Cancer.
      ,
      • Sung J.J.
      • Ng S.C.
      • Chan F.K.
      • et al.
      An updated Asia Pacific Consensus Recommendations on colorectal cancer screening.
      although CRC screening is limited by low patient adherence.
      • Stoop E.M.
      • de Haan M.C.
      • de Wijkerslooth T.R.
      • et al.
      Participation and yield of colonoscopy versus non-cathartic CT colonography in population-based screening for colorectal cancer: a randomised controlled trial.
      • Quintero E.
      • Castells A.
      • Bujanda L.
      • et al.
      Colonoscopy versus fecal immunochemical testing in colorectal-cancer screening.
      • Siegel R.L.
      • Miller K.D.
      • Fedewa S.A.
      • et al.
      Colorectal cancer statistics, 2017.
      • Chen C.
      • Stock C.
      • Hoffmeister M.
      • et al.
      Public health impact of colonoscopy use on colorectal cancer mortality in Germany and the United States.
      • Segnan N.
      • Senore C.
      • Andreoni B.
      • et al.
      Comparing attendance and detection rate of colonoscopy with sigmoidoscopy and FIT for colorectal cancer screening.
      Increasing patient participation in CRC screening remains a central task in current gastroenterology practice.
      The effectiveness of 2 screening principles has been established: (1) detection of biomarkers (eg, occult blood by the guaiac fecal occult blood test) in stool and (2) visual detection of (pre-)cancerous lesions by endoscopy. The guaiac fecal occult blood test and rectosigmoidoscopy can reduce the relative risk for CRC-related mortality by 16% and 31%, respectively.
      • Hewitson P.
      • Glasziou P.
      • Watson E.
      • et al.
      Cochrane systematic review of colorectal cancer screening using the fecal occult blood test (hemoccult): an update.
      ,
      • Atkin W.S.
      • Edwards R.
      • Kralj-Hans I.
      • et al.
      Once-only flexible sigmoidoscopy screening in prevention of colorectal cancer: a multicentre randomised controlled trial.
      The effect of colonoscopy has not been tested in randomized controlled trials, but many observational studies suggest a risk reduction of more than 60%.
      • Pan J.
      • Xin L.
      • Ma Y.F.
      • et al.
      Colonoscopy reduces colorectal cancer incidence and mortality in patients with non-malignant findings: a meta-analysis.
      Recently, tests detecting DNA with CRC-associated mutations in stool or blood have been advanced. Epi proColon (Epigenomics Inc, Berlin, Germany) is based on the detection of DNA in the blood with methylated cytosine in the v2 region of the SEPT9 gene, as found predominantly in colon cancer tissue.
      • Jin P.
      • Kang Q.
      • Wang X.
      • et al.
      Performance of a second-generation methylated SEPT9 test in detecting colorectal neoplasm.
      ColoGuard (Exact Sciences, Madison, Wisc, USA) is a stool-based multitarget DNA test that includes quantitative molecular assays for K-ras mutations, aberrant NDRG4 and BMP3 methylation, and β-actin, plus a hemoglobin immunoassay.
      A stool DNA test (Cologuard) for colorectal cancer screening.
      For both, high sensitivity for CRC has been shown, whereas for adenoma it was lower.
      • Jin P.
      • Kang Q.
      • Wang X.
      • et al.
      Performance of a second-generation methylated SEPT9 test in detecting colorectal neoplasm.
      ,
      A stool DNA test (Cologuard) for colorectal cancer screening.
      Another similarly effective approach is the urine metabolomic test called PolypDx (PolypDx, Metabolomic Technologies Inc, Edmonton, AB, Canada), in which key metabolites are quantified and analyzed by a multivariate algorithm.
      • Wang H.
      • Tso V.
      • Wong C.
      • et al.
      Development and validation of a highly sensitive urine-based test to identify patients with colonic adenomatous polyps.
      • Deng L.
      • Chang D.
      • Foshaug R.R.
      • et al.
      Development and validation of a high-throughput mass spectrometry based urine metabolomic test for the detection of colonic adenomatous polyps.
      • Deng L.
      • Fang H.
      • Tso V.K.
      • et al.
      Clinical validation of a novel urine-based metabolomic test for the detection of colonic polyps on Chinese population.
      • Deng L.
      • Ismond K.P.
      • Liu Z.
      • et al.
      Urinary metabolomics to identify a unique biomarker panel for detecting colorectal cancer: a multicentre study.
      However, for none of these newer tests, effectiveness regarding reduction of CRC incidence and mortality has been shown in clinical trials.
      In the search for a highly efficient screening test, sensitivity and specificity are just one part of the equation. Clinical experience shows that adherence of patients to diagnostic tests may differ substantially and could depend on different aspects, such as invasiveness, required preparation, or associated feeling of shame. The test modality seems to play an important role. In the case of colon cancer screening, no comparative study regarding patient preferences of testing modality has been undertaken. In the literature, uptake of colonoscopy screening is reported to be ∼56% over a 10-year period.
      • Siegel R.L.
      • Miller K.D.
      • Fedewa S.A.
      • et al.
      Colorectal cancer statistics, 2017.
      ,
      • Chen C.
      • Stock C.
      • Hoffmeister M.
      • et al.
      Public health impact of colonoscopy use on colorectal cancer mortality in Germany and the United States.
      Clinical trials involving screening of the general public for various diseases show moderate willingness to undertake stool- and urine-based tests (∼49.3% and 51.4%)
      • Quintero E.
      • Castells A.
      • Bujanda L.
      • et al.
      Colonoscopy versus fecal immunochemical testing in colorectal-cancer screening.
      ,
      • van Roon A.H.
      • Wilschut J.A.
      • Hol L.
      • et al.
      Diagnostic yield improves with collection of 2 samples in fecal immunochemical test screening without affecting attendance.
      • van Roon A.H.
      • Goede S.L.
      • van Ballegooijen M.
      • et al.
      Random comparison of repeated faecal immunochemical testing at different intervals for population-based colorectal cancer screening.
      • Zubero M.B.
      • Arana-Arri E.
      • Pijoan J.I.
      • et al.
      Population-based colorectal cancer screening: comparison of two fecal occult blood test.
      • Singh H.
      • Bernstein C.N.
      • Samadder J.N.
      • et al.
      Screening rates for colorectal cancer in Canada: a cross-sectional study.
      • Singal A.G.
      • Corley D.A.
      • Kamineni A.
      • et al.
      Patterns and predictors of repeat fecal immunochemical and occult blood test screening in four large health care systems in the United States.
      • Diercks G.F.
      • Janssen W.M.
      • van Boven A.J.
      • et al.
      Rationale, design, and baseline characteristics of a trial of prevention of cardiovascular and renal disease with fosinopril and pravastatin in nonhypertensive, nonhypercholesterolemic subjects with microalbuminuria (the Prevention of REnal and Vascular ENdstage Disease Intervention Trial [PREVEND IT.).
      • Bello A.K.
      • Peters J.
      • Wight J.
      • et al.
      The Kidney Evaluation and Awareness Program in Sheffield (KEAPS): a community-based screening for microalbuminuria in a British population.
      and higher willingness for blood tests (∼68%).
      • Connolly S.
      • Whyte R.
      Uptake of cancer screening services among middle and older ages in Ireland: the role of healthcare eligibility.
      ,
      • Bucholz E.M.
      • Rodday A.M.
      • Kolor K.
      • et al.
      Prevalence and predictors of cholesterol screening, awareness, and statin treatment among us adults with familial hypercholesterolemia or other forms of severe dyslipidemia (1999-2014).
      In addition, the interval timing of a screening test can have a significant impact on its effectiveness, especially in regard to the previously mentioned parameters. Hypothetically, repetitive use of a screening test with a lower sensitivity or adherence but short test interval can outperform another test with better test characteristics but longer test intervals (ie, colonoscopy). Last, a screening test has to be (the most) cost effective to justify its application. For many questions regarding CRC screening and prevention, randomized controlled trials are impracticable because of the large parameter space and the ethical and logistical challenges. Therefore, computational approaches have been developed, which simulate the natural history of CRC and screening interventions in a large patient population.
      • Buskermolen M.
      • Gini A.
      • Naber S.K.
      • et al.
      Modeling in colorectal cancer screening: assessing external and predictive validity of MISCAN-colon microsimulation model using NORCCAP trial results.
      ,
      • van Ballegooijen M.
      • Rutter C.M.
      • Knudsen A.B.
      • et al.
      Clarifying differences in natural history between models of screening: the case of colorectal cancer.
      Colon modeling open source tool (CMOST) is an open-source computational microsimulation tool, with similar predictions compared with randomized CRC prevention trials and other well-established microsimulation tools.
      • van Ballegooijen M.
      • Rutter C.M.
      • Knudsen A.B.
      • et al.
      Clarifying differences in natural history between models of screening: the case of colorectal cancer.
      ,
      • Prakash M.K.
      • Lang B.
      • Heinrich H.
      • et al.
      CMOST: an open-source framework for the microsimulation of colorectal cancer screening strategies.
      In this study, we first simulated theoretic CRC screening with a broad variation of key test characteristics in comparison with colonoscopy screening, to identify the characteristics of highly efficient screening tests. Then we used CMOST to calculate the incidence reduction, mortality reduction, life-years gained (LYGs), and cost effectiveness of screening with the available screening modalities: colonoscopy, fecal immunochemical test (FIT, Eiken Chemical Co, Ltd, Tokyo, Japan), ColoGuard, Epi proColon, and PolypDx, including various assumptions regarding screening adherence.

      Methods

      We simulated the use of several theoretic screening tests and varied 5 parameters independently: (1) sensitivity for early adenoma detection, (2) sensitivity for advanced adenoma detection, (3) sensitivity for carcinoma detection, (4) test specificity, and (5) adherence to testing. The first and second tests were varied in a range from 0% to 100% in 10% steps, and the third, fourth, and fifth tests were varied from 10% to 100% in 10% increments. Furthermore, for plausibility, sensitivity for cancer detection needed to be at least as good as the sensitivity for advanced adenoma detection, and the sensitivity for advanced adenoma detection needed to be at least as good as the sensitivity for early adenoma detection (SensCancer ≥ SensAdvAd ≥ SensEarlAd), resulting in the exclusion of several potential tests. Each test was applied either yearly or every 2, 3, 4, or 5 years, resulting in 142,501 theoretic tests. Adherence to follow-up colonoscopy after a positive test result was assumed to be 82%.
      • Jensen C.D.
      • Corley D.A.
      • Quinn V.P.
      • et al.
      Fecal immunochemical test program performance over 4 rounds of annual screening: a retrospective cohort study.
      ,
      • Corley D.A.
      • Jensen C.D.
      • Quinn V.P.
      • et al.
      Association between time to colonoscopy after a positive fecal test result and risk of colorectal cancer and cancer stage at diagnosis.
      The results of theoretic tests were compared with the results of CRC screening with colonoscopy at 100% adherence at ages 50, 60, and 70 years. Predictive models for discounted life-years gained (dLYG) and discounted U.S. dollars (dUSD) saved were calculated by a stepwise regression method that automatically adds or removes predictors (test characteristics in our case) starting from a constant model. More precisely, at each step of the method we added or removed a predictor term to the model if the P-value for an F-test of the change in the resulting sum of squared error was smaller than <.05 and and larger than >.1, respectively. We constrained the model to linear terms only (no interactions were considered), and the method stopped when no term was selected for removal or addition.
      The model structure and basic assumptions of CMOST have been described previously.
      • Prakash M.K.
      • Lang B.
      • Heinrich H.
      • et al.
      CMOST: an open-source framework for the microsimulation of colorectal cancer screening strategies.
      CMOST observes a given population from birth until death (maximum age, 100 years). The age distribution corresponds to a stable population without migration, with declining population size with increasing age. The ethnicity would correspond to the average distribution in the United States in the years 2005 to 2009 (66% white, 15% Latino-American, 12% African-American, 4% Asian-American) but has not been explicitly modeled. The sex distribution is 1:1 for male:female. For the natural history, the model was calibrated to reproduce the age- and sex-specific adenoma prevalence rates as well as carcinoma incidence and mortality from the Surveillance Epidemiology and End Results (SEER) Program database for 2005 to 2009. CRC develops through adenomatous precursors or spontaneous. CMOST considers 6 distinct adenoma stages, depending on size, and 4 cancer stages. Adenoma initiation and progression as well as cancer progression, screening, and surveillance are all modeled in time increments of 3 months. Adenomas can be detected during diagnostic examinations such as endoscopy. Cancer can be diagnosed by screening at an early stage or by symptoms. After detection, treatment can cure cancer with a stage-dependent probability.
      Compared with the original code,
      • Prakash M.K.
      • Lang B.
      • Heinrich H.
      • et al.
      CMOST: an open-source framework for the microsimulation of colorectal cancer screening strategies.
      the program code has been optimized for faster computational speed and now enables calculations with larger patient populations of ≤2 million individuals in each run. In contrast to the original model, CMOST now uses SEER CRC incidence and mortality data from the years 1988 to 2002,
      • Duggan M.A.
      • Anderson W.F.
      • Altekruse S.
      • et al.
      The Surveillance, Epidemiology, and End Results (SEER) program and pathology: toward strengthening the critical relationship.
      before the onset of widespread CRC screening. The new code is available at https://github.com/poljan/CMOSTv2.
      For the current study, the following screening strategies were applied: colonoscopy screening according to practice guidelines,
      • Rex D.K.
      • Boland C.R.
      • Dominitz J.A.
      • et al.
      Colorectal cancer screening: recommendations for physicians and patients from the U.S. Multi-Society Task Force on Colorectal Cancer.
      ,
      • Sung J.J.
      • Ng S.C.
      • Chan F.K.
      • et al.
      An updated Asia Pacific Consensus Recommendations on colorectal cancer screening.
      from age 50 to 75. As recommended by practice guidelines and/ or the individual manufacturers, screening with FIT was conducted yearly or biennially and Epi proColon yearly, whereas ColoGuard and PolypDx were performed every 3 years. Sensitivity and specificity for adenoma and carcinoma detection for ColoGuard,
      • Imperiale T.F.
      • Ransohoff D.F.
      • Itzkowitz S.H.
      • et al.
      Multitarget stool DNA testing for colorectal-cancer screening.
      Epi proColon,
      • Lamb Y.N.
      • Dhillon S.
      Epi proColon® 2.0 CE: a blood-based screening test for colorectal cancer.
      ,
      • Song L.
      • Peng X.
      • Li Y.
      • et al.
      The SEPT9 gene methylation assay is capable of detecting colorectal adenoma in opportunistic screening.
      FIT,
      • Quintero E.
      • Castells A.
      • Bujanda L.
      • et al.
      Colonoscopy versus fecal immunochemical testing in colorectal-cancer screening.
      ,
      • van Roon A.H.
      • Wilschut J.A.
      • Hol L.
      • et al.
      Diagnostic yield improves with collection of 2 samples in fecal immunochemical test screening without affecting attendance.
      • van Roon A.H.
      • Goede S.L.
      • van Ballegooijen M.
      • et al.
      Random comparison of repeated faecal immunochemical testing at different intervals for population-based colorectal cancer screening.
      • Zubero M.B.
      • Arana-Arri E.
      • Pijoan J.I.
      • et al.
      Population-based colorectal cancer screening: comparison of two fecal occult blood test.
      • Singh H.
      • Bernstein C.N.
      • Samadder J.N.
      • et al.
      Screening rates for colorectal cancer in Canada: a cross-sectional study.
      • Singal A.G.
      • Corley D.A.
      • Kamineni A.
      • et al.
      Patterns and predictors of repeat fecal immunochemical and occult blood test screening in four large health care systems in the United States.
      and PolypDx
      • Deng L.
      • Chang D.
      • Foshaug R.R.
      • et al.
      Development and validation of a high-throughput mass spectrometry based urine metabolomic test for the detection of colonic adenomatous polyps.
      ,
      • Deng L.
      • Ismond K.P.
      • Liu Z.
      • et al.
      Urinary metabolomics to identify a unique biomarker panel for detecting colorectal cancer: a multicentre study.
      were taken from the literature (Table 1). A positive result led to colonoscopy with a given adherence. Detection probabilities for adenomas of various sizes were provided (Table 1); detection probabilities for adenomas would translate nonlinearity to adenoma detection rates (ADRs) (Supplementary Fig. 1, available online at www.giejournal.org). ADR is defined as the percentage of screening colonoscopies with ≥1 adenoma and is widely used in clinical practice. For a single screening colonoscopy at age 60 years, our baseline adenoma detection probabilities would translate to an ADR of ∼15% in women and ∼27% in men. The detection and removal of adenoma and/or cancer led to endoscopic surveillance, with intervals as recommended by practice guidelines (5 years after nonadvanced adenoma detection, 3 years after advanced adenoma detection with lifelong 5-year intervals).
      • Lieberman D.A.
      • Rex D.K.
      • Winawer S.J.
      • et al.
      Guidelines for colonoscopy surveillance after screening and polypectomy: a consensus update by the US Multi-Society Task Force on Colorectal Cancer.
      Noninvasive screening was resumed at the earliest 10 years after inconspicuous colonoscopy. Finally, for the purpose of comparison, a no-screening strategy was also run.
      Table 1Characteristics and assumptions regarding colorectal cancer (CRC) screening tests
      VariableColonoscopyFITColoGuardEpi proColonPolypDxPolypDx

      low specificity
      Screening interval, y1013133
      Specificity10.950.870.960.9130.800
      Sensitivity
       Adenoma 2-5 mm0.65-0.750.050.170.200.430.53
       Adenoma 6-9 mm0.81-0.870.1010.170.200.430.53
       Adenoma ≥10 mm0.950.220.420.220.430.53
       CRC0.95-10.70.920.690.750.80
      Scenario with full adherence
       Yearly adherence111111
       Cumulative lifetime adherence111111
       Adherence to diagnostic colonoscopy-11111
      Scenario with real-world adherence
       Yearly adherence0.08220.4930.4930.680.51350.5135
       Cumulative lifetime adherence0.8830.5810.7850.7440.9310.931
       Adherence to diagnostic colonoscopy-0.820.820.820.820.82
      For PolypDx, in addition to the regular tests, characteristics for a variant with lower specificity are provided.
      FIT, Fecal immunochemical test.
      Regarding adherence to screening, 2 scenarios were of interest: full (100%) and real-world adherence. The purpose of the prior scenario was to capture the full potential of each screening modality and thereby provide a contrast point to better understand the influence of adherence on the simulation outcome. Here, adherence to screening was 100% for all available screening tests, as well as diagnostic and surveillance colonoscopy after a positive result. For the real-world scenario, adherence was derived from the literature. Here, colonoscopy screening was assumed to have an annual adherence of 0.0822, resulting in a 10-year adherence of 0.56.
      • Siegel R.L.
      • Miller K.D.
      • Fedewa S.A.
      • et al.
      Colorectal cancer statistics, 2017.
      ,
      • Chen C.
      • Stock C.
      • Hoffmeister M.
      • et al.
      Public health impact of colonoscopy use on colorectal cancer mortality in Germany and the United States.
      Several large studies have shown the adherence to FIT to average ∼0.493.
      • Quintero E.
      • Castells A.
      • Bujanda L.
      • et al.
      Colonoscopy versus fecal immunochemical testing in colorectal-cancer screening.
      ,
      • van Roon A.H.
      • Wilschut J.A.
      • Hol L.
      • et al.
      Diagnostic yield improves with collection of 2 samples in fecal immunochemical test screening without affecting attendance.
      • van Roon A.H.
      • Goede S.L.
      • van Ballegooijen M.
      • et al.
      Random comparison of repeated faecal immunochemical testing at different intervals for population-based colorectal cancer screening.
      • Zubero M.B.
      • Arana-Arri E.
      • Pijoan J.I.
      • et al.
      Population-based colorectal cancer screening: comparison of two fecal occult blood test.
      • Singh H.
      • Bernstein C.N.
      • Samadder J.N.
      • et al.
      Screening rates for colorectal cancer in Canada: a cross-sectional study.
      • Singal A.G.
      • Corley D.A.
      • Kamineni A.
      • et al.
      Patterns and predictors of repeat fecal immunochemical and occult blood test screening in four large health care systems in the United States.
      The adherence to ColoGuard (stool) was assumed to be the same. Uptake of PolypDx (urine) and Epi proColon (blood), was assumed to be similar to the willingness of the general public to undergo microalbuminuria or prostate specific antigen (PSA) and dyslipidemia screening, respectively (0.514, 0.68).
      • Diercks G.F.
      • Janssen W.M.
      • van Boven A.J.
      • et al.
      Rationale, design, and baseline characteristics of a trial of prevention of cardiovascular and renal disease with fosinopril and pravastatin in nonhypertensive, nonhypercholesterolemic subjects with microalbuminuria (the Prevention of REnal and Vascular ENdstage Disease Intervention Trial [PREVEND IT.).
      • Bello A.K.
      • Peters J.
      • Wight J.
      • et al.
      The Kidney Evaluation and Awareness Program in Sheffield (KEAPS): a community-based screening for microalbuminuria in a British population.
      • Connolly S.
      • Whyte R.
      Uptake of cancer screening services among middle and older ages in Ireland: the role of healthcare eligibility.
      • Bucholz E.M.
      • Rodday A.M.
      • Kolor K.
      • et al.
      Prevalence and predictors of cholesterol screening, awareness, and statin treatment among us adults with familial hypercholesterolemia or other forms of severe dyslipidemia (1999-2014).
      Adherence to diagnostic and surveillance colonoscopy after positive screening was 0.82 in the real-world scenario.
      • Jensen C.D.
      • Corley D.A.
      • Quinn V.P.
      • et al.
      Fecal immunochemical test program performance over 4 rounds of annual screening: a retrospective cohort study.
      ,
      • Corley D.A.
      • Jensen C.D.
      • Quinn V.P.
      • et al.
      Association between time to colonoscopy after a positive fecal test result and risk of colorectal cancer and cancer stage at diagnosis.
      The cost assumptions for the United States were taken from a recent study
      • van Hees F.
      • Saini S.D.
      • Lansdorp-Vogelaar I.
      • et al.
      Personalizing colonoscopy screening for elderly individuals based on screening history, cancer risk, and comorbidity status could increase cost effectiveness.
      and transformed to 2018 U.S. dollars (Supplementary Table 1, available online at www.giejournal.org).

      US Bureau of Labor 2013. Medical care costs. Medical care in U.S. city average, all urban consumers, not seasonally adjusted 2019. Available at: Statisticshttps://data.bls.gov/timeseries/CUUR0000SAM?output_view=pct_12mths. Accessed September 12, 2019.

      In CMOST, costs are computed at 3-month intervals. Three time periods are distinguished: the first 12 months after diagnosis, the last 12 months before CRC-related death, and the follow-up period in between, with a maximum total of 5 years after cancer diagnosis. Overlap of periods is not possible. In a case of short survival after diagnosis, the last period took precedence. CRC-related costs differ according to CRC stage (Supplementary Table 1). In a case of death unrelated to CRC, additional costs apply as described.
      • van Hees F.
      • Saini S.D.
      • Lansdorp-Vogelaar I.
      • et al.
      Personalizing colonoscopy screening for elderly individuals based on screening history, cancer risk, and comorbidity status could increase cost effectiveness.
      When indicated, costs and life-years gained were discounted by 3% per year after the start of screening at age 50. Costs for colonoscopy, FIT, Epi pro Colon, ColoGuard, and PolypDx screening were derived from the Centers for Medicare and Medicaid Services, 2018 Clinical Laboratory Fee Schedule Public Use File.

      Centers for Medicare and Medicaid Services 2016, 18CLABQ2. Available at: https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/ClinicalLabFeeSched/Clinical-Laboratory-Fee-Schedule-Files. Accessed June 12, 2019.

      Here, the costs for colonoscopy include only moderate sedation by the gastroenterologist, not full anesthesia care. The costs calculated in the nonendoscopic screening scenarios include those for colonoscopy resulted by a positive screening test result. The model does not distinguish between true- and false-positive tests. Cost effectiveness was calculated on a comparison with no screening, with a cutoff of 100,000 dUSD per dLYGs considered cost effective and values below zero considered cost saving.
      For all analyses, CMOST was used, and calculations were performed on a high-performance Linux cluster. A population of 2 million individuals was simulated in each individual run. The results are the average of 10 replicate analyses. For all outcomes, normal distribution of the replicates was confirmed through the Anderson-Darling test and are shown as means with the respective standard deviations. Postprocessing analyses were performed with customized scripts in Matlab, version R2018b. Data were further processed with Excel (version 16, Microsoft, Redmond, Wash, USA) and Graphpad Prism (version 8, Graphpad, San Diego, Calif, USA).
      For the sensitivity analysis we focused on 3 model output variables: (1) incidence reduction, (2) life-years gained, and (3) dUSD per dLYG. We assessed the uncertainty (perturbations) of 8 model parameters: (1) adherence to screening, (2) adherence to test follow-up, (3) probability of adenoma detection during colonoscopy (eg, due to suboptimal bowel preparation
      • Sulz M.C.
      • Kroger A.
      • Prakash M.
      • et al.
      Meta-analysis of the effect of bowel preparation on adenoma detection: early adenomas affected stronger than advanced adenomas.
      ), (4) risk for colonoscopy adverse events, (5) screening cost, (6) treatment cost, (7) test sensitivity, and (8) test specificity. Each set of variables was perturbed within a specified range (Supplementary Table 1) by use of a uniform random variable. All variables and sensitivity indices—defined as fraction of total output variance generated by the uncertainty in the respective parameter value—were perturbed simultaneously by the Fourier Amplitude Sensitivity Test (FAST) sampling method.
      • Cukier R.
      • Fortuin C.
      • Shuler K.
      • et al.
      Study of the sensitivity of coupled reaction systems to uncertainties in rate coefficients. I theory.
      The higher the value of the sensitivity index, the more influential is the uncertainty (variation) in the parameter compared with the others. Correlation coefficients between analyzed model output variables and sampled model parameters were calculated by the Pearson method.

      Results

      Key characteristics of highly efficient screening tests

      Colonoscopy screening is assumed to be highly efficient for prevention of life-years lost to CRC. We were looking for highly efficient screening tests that would be able to match or outperform the efficiency of colonoscopy screening. We systematically varied the sensitivity for early and advanced adenoma, carcinoma, test specificity, adherence, and the frequency with which this hypothetical test was applied. All theoretic tests were compared with colonoscopy screening regarding dLYG and costs (see Methods).
      Out of 142,501 theoretic tests, 6299 tests saved more life-years than standard colonoscopy screening (Fig. 1A). When characteristics of efficient screening tests were looked at, all of them were associated with moderate adherence (at least ∼50%). Further, efficient tests had a sensitivity threshold for advanced adenomas of at least ∼40% and carcinomas of at least ∼70%, with a specificity of at least ∼40% (Fig. 1B). Most theoretic efficient tests were applied yearly or biyearly, and almost no efficient tests were used less frequently than every 4 years (Fig. 1B). Life-years gained and costs saved associated with the use of these tests followed closely simple linear equations (Fig. 1C).
      Figure thumbnail gr1
      Figure 1Characteristics of highly efficient colorectal cancer (CRC) screening tests. A, Simulation of 142,501 theoretical CRC screening tests with varying sensitivities for early and advanced adenoma as well as CRC, varying specificity, adherence to testing, and test intervals. Displayed are discounted life-years gained as well as discounted costs compared with colonoscopy screening with 100% adherence. Blue dots, inefficient tests compared with colonoscopy. Red dots, tests that save a higher number of discounted life-years compared with colonoscopy. B, Key characteristics of highly efficient theoretic tests, surpassing effectiveness of colonoscopy. Highly efficient tests tended to have higher sensitivity for advanced adenoma and CRC, as well as high specificity and adherence to testing and a test interval of 1 or 2 years. C, Effectiveness (discounted life-years gained, upper panel) and costs (discounted U.S. dollars, lower panel) of highly efficient tests are plotted according to the test interval. We performed stepwise linear regression to predict effectiveness and costs of highly efficient tests from the indicated screening parameters (red line).

      Simulation of CRC screening strategies with hypothetical 100% adherence to screening

      In the hypothetical scenario with 100% patient adherence, colonoscopy screening reduced CRC incidence by 55% and mortality by 62% compared with no screening (Fig. 2A and B). Colonoscopy screening resulted in the gain of 71.6 discounted life-years per 1000 individuals (Fig. 2C). Colonoscopy-based CRC screening was cost effective compared with no screening, with 2101 dUSD for each dLYG (Fig. 2D, Table 2).
      Figure thumbnail gr2
      Figure 2Colorectal cancer (CRC) screening with full and real-world adherence. A, CRC incidence reduction. B, CRC mortality reduction. C, Life-years gained because of CRC screening. D, Cost effectiveness for each screening interventions.
      Table 2CRC screening with a theoretic 100% and real-world adherence to the screening intervention
      VariableNo screeningColonoscopyFITFIT, 2 yColoGuardEpi proColonPolypDx
      CRC cases (per 1000)67.4 (0.2)30.3 (0.1)

      42.5 (0.1)
      34.6 (0.1)

      44.0 (0.2)
      40.4 (0.1)

      48.3 (0.2)
      36.7 (0.1)

      42.8 (0.2)
      33.4 (0.1)

      38.1 (0.2)
      34.1 (0.1)

      39.1 (0.1)
      Screening tests (per 1000)0 (0)

      0 (0)
      14,248 (4.9)

      8559 (4.1)
      8831 (2.3)

      6203 (1.3)
      5482 (1.7)

      4445 (2.1)
      13,587 (3.7)

      10,673 (3.8)
      5420 (1.3)

      4444 (1.4)
      Colonoscopies (per 1000)242 (0.8)3145 (1.6)

      1632 (1.0)
      1464 (2.0)

      934 (1.2)
      1084 (1.5)

      768 (1.6)
      1438 (0.8)

      1087 (1.5)
      1627 (1.6)

      1281 (1.2)
      1545 (1.8)

      1227 (1.3)
      Incidence reduction (%)55.0 (0.16)

      36.9 (0.25)
      48.7 (0.24)

      34.7 (0.27)
      40.0 (0.3)

      28.3 (0.4)
      45.5 (0.3)

      36.5 (0.3)
      50.4 (0.3)

      43.5 (0.2)
      49.4 (0.2)

      41.9 (0.2)
      CRC stage distribution (%)
       Stage I23.4 (0.1)31.2 (0.2)

      27.6 (0.1)
      40.5 (0.2)

      33.8 (0.2)
      36.8 (0.2)

      31.5 (0.2)
      35.1 (0.2)

      31.6 (0.1)
      39.6 (0.2)

      35.9 (0.1)
      33.4 (0.2)

      30.7 (0.2)
       Stage II32.6 (0.1)33.4 (0.2)

      32.6 (0.2)
      31.1 (0.2)

      32.9 (0.2)
      33.3 (0.2)

      33.3 (0.1)
      33.8 (0.2)

      33.4 (0.2)
      31.3 (0.2)

      32.5 (0.2)
      33.4 (0.2)

      33.1 (0.2)
       Stage III25.2 (0.1)21.1 (0.2)

      23.2 (0.2)
      17.2 (0.2)

      20.0 (0.1)
      18.3 (0.1)

      21.1 (0.1)
      19.2 (0.2)

      21.0 (0.2)
      17.5 (0.2)

      19.2 (0.1)
      20.1 (0.1)

      21.6 (0.1)
       Stage IV18.9 (0.1)14.3 (0.1)

      16.6 (0.1)
      11.3 (0.2)

      13.3 (0.1)
      11.6 (0.1)

      14.2 (0.1)
      12.0 (0.2)

      14.0 (0.1)
      11.5 (0.2)

      12.4 (0.1)
      13.1 (0.1)

      14.6 (0.1)
      CRC deaths (per 1000)25.6 (0.1)9.8 (0.04)

      14.8 (0.1)
      9.6 (0.04)

      13.6 (0.1)
      11.6 (0.1)

      15.5 (0.1)
      10.8 (0.1)

      13.7 (0.1)
      9.4 (0.1)

      11.3 (0.1)
      10.4 (0.1)

      12.7 (0.1)
      Mortality reduction (%)61.7 (0.2)

      42.3 (0.3)
      62.4 (0.2)

      47.0 (0.3)
      54.7 (0.4)

      39.4 (0.5)
      57.8 (0.3)

      46.7 (0.4)
      63.1 (0.2)

      56.0 (0.4)
      59.2 (0.3)

      50.4 (0.4)
      Total costs (per capita, in USD)7858 (21.0)6473 (10.7)

      6296 (13.7)
      5181 (11.1)

      5613 (18.0)
      5290 (13.2)

      5926 (20.0)
      7963 (16.4)

      7821 (21.3)
      7627 (13.0)

      7220 (10.8)
      6206 (12.2)

      6280 (17.4)
      Screening costs02481 (0.9)

      1120 (0.5)
      1171 (0.6)

      642 (0.4)
      759 (0.8)

      476 (0.4)
      3672 (0.8)

      2871 (1.2)
      3575 (0.8)

      2738 (0.7)
      1976 (0.7)

      1535 (0.7)
      Treatment costs7660 (20.2)3063 (10.5)

      4525 (13.7)
      3274 (9.3)

      4414 (17.8)
      3911 (13.0)

      4965 (19.8)
      3580 (15.7)

      4359 (19.8)
      3187 (12.4)

      3727 (11.0)
      3382 (11.9)

      4013 (17.4)
      Follow-up costs199 (0.9)928 (2.2)

      650 (1.3)
      735 (2.1)

      557 (1.6)
      620 (1.4)

      485 (1.8)
      710 (1.0)

      591 (1.5)
      865 (1.4)

      756 (1.1)
      847 (1.7)

      732 (1.5)
      LYG (per 1000)177.4 (1.3)

      112.2 (1.8)
      184.7 (1.8)

      135.7 (1.7)
      160.9 (1.9)

      113.1 (2.5)
      168.1 (1.7)

      132.3 (1.7)
      184.6 (1.8)

      160.3 (2.0)
      168.7 (1.9)

      139.4 (1.8)
      dLYG (per 1000)71.6 (0.5)

      43.3 (0.7)
      74.9 (0.7)

      54.3 (0.7)
      65.1 (0.8)

      45.1 (1.1)
      67.9 (0.7)

      53.0 (1.0)
      74.8 (0.8)

      64.2 (0.8)
      67.6 (0.8)

      55.0 (0.7)
      dUSD per dLYG2101 (201)

      -8419 (332)
      -12,643 (190)

      -16,086 (272)
      -15,352 (213)

      -16,992 (334)
      16,169 (338)

      14,965 (486)
      12,209 (267)

      7530 (230)
      -2696 (156)

      -4931 (229)
      For each screening modality, a population of 2 million individuals was simulated 10 times. The numbers shown are the means of these 10 simulations. The values of incidence and mortality reduction as well as LYG, dLYG, and dUSD per dLYG are compared with no screening. Italic numbers represent the real-world scenario; standard deviation is provided in parenthesis.
      CRC, Colorectal carcinoma; FIT, fecal immunochemical test; LYG, life-years gained; dLYG, discounted life-years gained; USD, U.S. dollars, dUSD, discounted U.S. dollars.
      All nonendoscopic screening tests reduced CRC incidence and mortality and resulted in a gain in life-years (Fig. 1). Here, the yearly FIT and Epi proColon tests were the most effective, surpassing even colonoscopy screening (Fig. 2C). Performing FIT biennially led to a significant reduction of this effectiveness. Compared with no screening, all tests were cost effective with <50,000 dUSD per dLYG; only the FIT and PolypDx were cost saving (Fig. 2D, Table 2) (Supplementary Fig. 2, available online at www.giejournal.org).

      Simulation of CRC screening with expected, test-specific adherence rates

      Compared with the hypothetical 100% adherence, the incidence and mortality reduction of all screening tests were reduced, as were the life-years gained (Fig. 2). Owing to higher adherence to blood-based tests, Epi proColon now achieved a discrepantly higher gain of discounted life-years. However, compared with the prior scenario, the cost effectiveness of all screening tests increased and now, next to FIT and PolypDx, colonoscopy was also cost saving (Fig. 2D, Table 2, Supplementary Fig. 2) when compared with no screening.

      Incremental cost-effectiveness ratio

      To compare the cost-effectiveness of all screening methods, incremental cost-effectiveness ratios (ICERs) were calculated. For hypothetical 100% adherence, yearly FIT was the dominating strategy, yielding the highest number of dLYG (74.9 discounted life-years per 1000 individuals compared with no screening) at the second lowest discounted costs (Table 3). Only biennial FIT was slightly more cost effective, however, yielding in the lowest number of discounted life-years. For real-life adherence rates, yearly FIT was surpassed by PolypDx and Epi proColon in efficacy (55 and 64.2 vs 54.3 dLYG) but was still the most cost-efficient screening strategy. PolypDx and Epi proColon came with an incremental cost of 860,378 and 137,060 dUSD per dLYG, respectively (Table 3).
      Table 3Incremental cost effectiveness ratio (ICER) table


      100% Adherence
      No screeningFIT, 2 yPolypDxColoGuardColonoscopyEpi proColonFIT
      No screeningDominatedDominated$16,169$2101$12,209Dominated
      FIT, 2 y$326,866$749,032$176,900$197,180$5353
      PolpypDx$4,267,082$83,170$152,150Dominated
      ColoGuardDominatedDominatedDominated
      Colonoscopy$238,376Dominated
      Epi proColonDominated
      FIT
      Real-world adherence
      No screeningColonoscopyFIT, 2 yColoGuardFITPolypDxEpi proColon
      No screeningDominatedDominated$14,965DominatedDominated$7530
      ColonoscopyDominated$119,349Dominated$7978$40573
      FIT, 2 y$19,7403Dominated$50,014$65433
      ColoGuardDominatedDominatedDominated
      FIT$860,378$137,060
      PolypDx$82,025
      Epi proColon
      Read left to right. The tables shows the relationship of index screening method (left) compared with another screening method shown above. The numbers illustrate the costs per life-year gained. If a screening method is surpassed at life-years gained at lower costs, it is considered to be dominated by the other. All calculations were done with discounted life-years and costs. Upper part of table: 100% adherence. Lower part of table: real-word adherence rates.
      FIT, Fecal immunochemical test.

      Systematic variation of adherence rates

      To demonstrate the impact of screening adherence on screening effectiveness, we calculated dose-response curves for each screening strategy (Fig. 3) (Supplementary Fig. 3, available online at www.giejournal.org). Higher adherence rates yielded superior effectiveness of all measures of screening effectiveness. As expected, higher adherence resulted in higher ICERs compared with no screening for all screening tests (Fig. 3D). In other words, the lowest number of dUSD per dLYG was found at very low adherence rates, and a rise in adherence increased the number of dUSD for each dLYG.
      Figure thumbnail gr3
      Figure 3Systematic variation of screening adherence. A, Colorectal cancer (CRC) incidence reduction. B, CRC mortality reduction. C, Life-years gained because of CRC screening. D, Cost effectiveness for each screening intervention. Cost effectiveness at low adherence showed pronounced stochastic variations for some tests even with large study populations; therefore, the range from 0% to 5% is not shown. FIT, Fecal immunochemical test.

      Sensitivity analysis

      To determine the sensitivity of various performance parameters including ADR, incidence reduction, life-years gained, and cost effectiveness to our choice of parameters, we performed a sensitivity analysis (Fig. 4) (Supplementary Fig. 4, Supplementary Table 2, available online at www.giejournal.org). A considerable variation in incidence reduction and life-years gained was observed (Fig. 4A, Supplementary Fig. 3). Cost effectiveness also varied; however, results were almost consistently below a threshold of 100,000 dUSD per dLYG (Fig. 4A, Supplementary Fig. 4). Incidence reduction and life-years gained were most sensitive to variations in screening adherence and follow-up colonoscopies (Fig. 4B and C, Supplementary Fig. 4). Both outcomes were also sensitive to adenoma detection, but to a lower degree. Cost effectiveness was mainly sensitive to variations in screening and treatment costs (Fig. 4B and C) (Supplementary Fig. 4).
      Figure thumbnail gr4
      Figure 4Sensitivity analysis results for colonoscopy (colo), FIT, and PolypDx. Selected model parameters were varied simultaneously. A, Boxplots of analyzed output variables. Shown are medians (horizontal lines), 25th and 75th percentiles (bottom and top of boxes, respectively). Whiskers show minimal and maximal values after removal of outliers (values outside of 1.5 × interquartile range). B, Pearson correlation coefficients between each perturbed parameter and analyzed output variables. ∗, ∗∗, and ∗∗∗ denote P values < .05, < .01, and < .001, respectively. C, First-order sensitivity indices, defined as the fraction of the total variance in the output variable explained by the variation in each parameter value, calculated by the Fourier amplitude sensitivity test. Unexplained part refers to the part of variance that cannot be attributed to any single parameter and is related to higher-order interactions between parameters. LY, Life-years; USD, U.S. dollars; disc., discounted. FIT, fecal immunochemical test.

      Discussion

      In this study, we used the open-source microsimulation tool CMOST to model effectiveness and cost effectiveness of various screening strategies for CRC. Simulating a large range of theoretic screening tests, we identified tests characterized by good advanced adenoma and carcinoma sensitivity and test adherence that could outperform even colonoscopy screening with perfect adherence. When simulated available screening tests, yearly FIT, and yearly Epi proColon saved the highest number of life-years when 100% adherence to screening was assumed. The efficiency of all available screening modalities did not differ greatly, but their cost effectiveness did. Here, the cheap FIT performed best and was even cost saving.
      As demonstrated by our adherence titration, incidence reduction and the number of life-years gained by each screening modality critically depends on adherence to screening protocol. When real-world adherence rates were applied, our simulation thus shows a superior effectiveness of some noninvasive screening methods compared with colonoscopy. Our sensitivity analysis also demonstrates that the effectiveness of screening strategies was strongly dependent on adherence to the screening test and follow-up colonoscopies, more so than on the individual test’s sensitivity for detecting CRC. Therefore, a noninvasive test with high acceptance could outperform even colonoscopy regarding CRC and adenoma detection. Given that patient preferences naturally differ between individuals, an ability to offer a variety of tests and patient-centered counseling might be optimal for improving adherence.
      We identified several theoretic tests that could outperform colonoscopy screening. This seems counterintuitive because all adenomas and carcinomas were only detected and/or removed by colonoscopy. However, the higher testing frequency of noninvasive screening tests allowed in our simulations for an efficient early detection of advanced lesions. The effects of an increased testing frequency would be similar to those of improved testing adherence.
      In our calculations, all screening strategies were effective and cost effective (<100,000 dUSD/dLYG) at preventing CRC. Cost effectiveness was primarily dependent on screening and treatment costs. Having the lowest price by far allows FIT to remain highly cost effective despite yearly application. The high test specificity for FIT compared favorably with those of most other noninvasive screening tests except Epi proColon and likely also contributed to its cost effectiveness. At real-world adherence rates, the effectiveness of FIT was only slightly surpassed by Epi proColon and PolypDx at additional costs (ICER) of 137,060 and 860,378 dUSD/dLYG, respectively. In the same scenario, 3 screening tests (colonoscopy, FIT, and PolypDx) were cost saving. However, with an increasing number of expensive drug therapies for advanced CRC, the cost effectiveness of all screening tests can be expected to rise.
      Our study has several strengths and limitations. (1) We use a novel, versatile, and open-source microsimulation tool, validated against existing tools and literature data, and freely available to reproduce all simulation results.
      • Prakash M.K.
      • Lang B.
      • Heinrich H.
      • et al.
      CMOST: an open-source framework for the microsimulation of colorectal cancer screening strategies.
      In contrast to proprietary microsimulation tools, our calculations can be independently validated and advanced. (2) Improved CMOST implementation combined with substantial computational power allowed us to simulate very large populations and enabled us (3) to explore >140,000 theoretic tests to identify characteristics of highly efficient tests and (4) to simulate a high number of scenarios in our extensive sensitivity analysis, thus increasing confidence in our results. Limitations include (1) the in silico nature of our study, necessarily relying on assumptions from the literature for all variables. Therefore, recent increases in the incidence of colorectal cancer are not taken into account, and serrated polyps are not included. (2) The natural history of CRC is partially unknown, and the adenoma dwell time—the average time an adenoma would reside in the colon until transformation to CRC—has not been empirically determined.
      • Buskermolen M.
      • Gini A.
      • Naber S.K.
      • et al.
      Modeling in colorectal cancer screening: assessing external and predictive validity of MISCAN-colon microsimulation model using NORCCAP trial results.
      ,
      • van Ballegooijen M.
      • Rutter C.M.
      • Knudsen A.B.
      • et al.
      Clarifying differences in natural history between models of screening: the case of colorectal cancer.
      CMOST assumes an average dwell time of 13 years, but shorter assumptions of the dwell time might result in a lower efficiency of strategies partially relying on early adenoma detection, such as colonoscopy. (3) Calibration of CMOST and cost assumptions were specific to the CRC epidemiology and health economy of the United States, and our calibrations might not be applicable to other geographic regions. For instance, a lower CRC incidence and/or lower CRC treatment costs in developing countries would fundamentally change the results of our calculations. (4) For the United States, the costs for colonoscopy provided by the Medicare and Medicaid 2018 Clinical Laboratory Fee Schedule Public Use File include only moderate sedation but not anesthesia care. Performing colonoscopy under full anesthesia would result in a less favorable cost effectiveness. (5) A scenario with patient-specific screening preferences and application of CRC screening according to patient choices has not been modeled because of the complexity of that scenario.
      In summary, we report the efficacy and cost effectiveness of a variety of CRC screening approaches. Our results point to the crucial importance of adherence to screening. Thereby, incidence and mortality reduction by a test with a lower efficacy but better adherence and/or testing frequency can surpass the effectiveness of colonoscopy in a population-based screening program.

      Acknowledgments

      This project was supported by a grant from the Swiss Cancer League (grant No. KFS-2977-08-2012) and the Horten Foundation to Benjamin Misselwitz.

      Appendix 1. Methods

      Two thresholds for PolypDx were analyzed in this study: (1) Specificity of 91% and sensitivity 43% for adenomas
      • Bray F.
      • Ferlay J.
      • Soerjomataram I.
      • et al.
      Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries.
      and 74% for colorectal cancer (CRC).
      • Brenner H.
      • Kloor M.
      • Pox C.P.
      Colorectal cancer.
      (2) Specificity of 80% and sensitivity of 53% for adenomas
      • Bray F.
      • Ferlay J.
      • Soerjomataram I.
      • et al.
      Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries.
      and 80% for CRC.
      • Brenner H.
      • Kloor M.
      • Pox C.P.
      Colorectal cancer.

      Results

      Variation in PolypDx Specificity

      Specificity for colorectal cancer was similar between the current version of PolypDx and FIT (0.913 for PolypDx vs 0.95 for FIT). We simulated another version of PolypDx with reduced specificity (0.80 instead of 0.91) but higher sensitivity for adenomas (0.53 instead of 0.43) and CRC (0.8 instead of 0.74). With 100% adherence, incidence and mortality reduction slightly increased (incidence reduction 51.5% compared with 49.4%; mortality reduction 60.3% compared with 59.2% along with a decrease in cost effectiveness (−586 discounted U.S. dollars [dUSD] per discounted life-year gained [dLYG] compared with −2969 dUSD per dLYG). This was due to a high rise in colonoscopy numbers after PolypDx with reduced specificity (Supplementary Table 3).
      Figure thumbnail fx2
      Supplementary Figure 1Relationship between detection probabilities of adenomas and adenoma detection rates (ADR). We simulated a single screening colonoscopy at age 60 years with detection of adenomas according to the probabilities provided in . The ADR was defined as the percentage of colonoscopies with at least 1 early or advanced adenoma detected and is provided separately for women and men. Baseline detection parameters were decreased and increased as indicated (censored at 100%). Error bars indicate mean and standard deviation of 10 repeated calculations with 20 million individuals.
      Figure thumbnail fx3
      Supplementary Figure 2Effects of colorectal cancer (CRC) screening with full and real-world adherence (Adh.). As in , but A, Total number of colonoscopies. B, Total number of screening tests. C, Number of colonoscopies per CRC case prevented. D, Number of colonoscopies per life-year gained. E, Total number of CRC cases. FIT, Fecal immunochemical test; exp., expected.
      Figure thumbnail fx4
      Supplementary Figure 3Systematic variation of screening adherence. As in , but A, Number of CRC cases. B, Colorectal cancer (CRC) mortality cases. C, Colonoscopies per CRC case prevented. D, Colonoscopies per life-year gained. E, Total number of colonoscopies. F, Total number of screening tests. FIT, Fecal immunochemical test.
      Figure thumbnail fx5
      Supplementary Figure 4Sensitivity analysis results for Epi proColon and ColoGuard. As in ; selected model parameters were varied simultaneously. A, Boxplots of analyzed output variables. Shown are medians (horizontal lines), 25th and 75th percentiles (bottom and top of boxes, respectively). Whiskers show minimal and maximal values after removal of outliers (values outside of 1.5 × interquartile range). B, Pearson correlation coefficients between each perturbed parameter and analyzed output variables. indicated; ∗, ∗∗, and ∗∗∗ denote P values < .05, < .01, and < .001, respectively. C, First-order sensitivity indices, defined as the fraction of the total variance in the output variable explained by the variation in each parameter value, calculated by the Fourier amplitude sensitivity test. Unexplained part refers to the part of variance that cannot be attributed to any single parameter and is related to higher order interactions between parameters. disc., discounted; LY, life-years.
      Supplementary Table 1Costs of screening interventions, colonoscopy adverse events, and CRC treatment
      CRC screening
      Costs screening testsCosts colonoscopy complications
      FIT$19.64Bleeding$1333
      Epi proColon$192Severe bleeding$7723
      PolypDx$200Serosa burn$9269
      ColoGuard$508.87Perforation$14,349
      Colonoscopy$1020
      Colonoscopy with polypectomy$1260
      CRC treatmentStage IStage IIStage IIIStage IV
       Year of diagnosis$42,187$56,621$68,725$89,462
       Follow-up year$3508$3301$4625$14,005
       Year of death$73,383$73,091$77,100$101,627
       Year of death, other causes$22,053$24,676$19,872$57,348
      Costs were used as described
      • Baxter N.N.
      • Goldwasser M.A.
      • Paszat L.F.
      • et al.
      Association of colonoscopy and death from colorectal cancer.
      ,
      • Hewitson P.
      • Glasziou P.
      • Watson E.
      • et al.
      Cochrane systematic review of colorectal cancer screening using the fecal occult blood test (hemoccult): an update.
      and adjusted for inflation to 2018 USD.
      • Atkin W.S.
      • Edwards R.
      • Kralj-Hans I.
      • et al.
      Once-only flexible sigmoidoscopy screening in prevention of colorectal cancer: a multicentre randomised controlled trial.
      CRC, Colorectal cancer; FIT, fecal immunochemical test.
      Supplementary Table 2Strategy for sensitivity analysis
      CategoryStarting valueMinimumMaximum
      Adherence to screening
      Urine-based test0.8360.20.95
       Stool-based test0.4930.20.95
       Blood-based test0.70.20.95
       Colonoscopy0.08220.050.95
      Adherence to follow-up
      0.750.201.00
      Adenoma detection by colonoscopy
       3-mm adenoma0.300.85
       5-mm adenoma0.750.350.90
       7-mm adenoma0.810.400.90
       9-mm adenoma0.870.400.92
       Advanced adenoma ≥10 mm0.950.500.99
      Sensitivity of screening tests for adenoma and carcinoma detection
       FIT: 3-5 mm adenoma0.050.020.125
       FIT: 6-9 mm adenoma0.1010.040.25
       FIT: advanced adenoma ≥10 mm0.220.080.40
       FIT: CRC0.70.300.90
       PolypDx: 3-9 mm adenoma0.430.1720.80
       PolypDx: advanced adenoma ≥10 mm0.430.1720.80
       PolypDx: CRC0.750.300.90
       Epi pro Colon: 3-9 mm adenoma000.25
       Epi pro Colon: advanced adenoma ≥10 mm000.25
       Epi pro Colon: CRC0.8060.3450.92
       ColoGuard: 3-9 mm adenoma0.170.07290.4
       ColoGuard: advanced adenoma ≥10 mm0.420.180.8
       ColoGuard: CRC0.920.3940.95
      Specificity of screening tests for adenoma and carcinoma detection
       FIT0.950.700.98
       PolypDx0.9130.680.96
       Epi proColon0.9690.720.99
       ColoGuard0.870.640.94
      Colonoscopy risk (25%–300%)
       Perforation0.00070.0001750.0021
       Death after perforation0.0520.0130.156
       Bleeding0.00110.0002750.0033
       Severe bleeding0.00040.00010.0012
       Death after bleeding0.00520.00130.0156
       Serosa burn0.00030.0000750.0009
      Screening costs (50%–180%)
       Colonoscopy$1020.00$510.00$1836.00
       Colonoscopy with polypectomy$1260.00$630.00$2268.00
       FIT$25.00$12.50$45.00
       PolypDx$475.00$237.50$855.00
       ColoGuard$508.87$254.44$915.97
       Epi proColon$192.00$96.00$345.60
       Costs perforation$14,349.00$7174.50$25,828.20
       Costs bleeding$1333.00$666.50$2399.40
       Costs severe bleeding$7723.00$3861.50$13,901.40
       Costs serosa burn$9269.00$4634.50$16,684.20
      Treatment costs (50%–180%)
       First year, stage I$42187.00$21,093.50$75,936.60
       First year, stage II$56,621.00$28,310.50$101,917.80
       First year, stage III$68,725.00$34,362.50$123,705.00
       First year, stage IV$89,462.00$44,731.00$161,031.60
       Follow-up stage I$3508.00$1754.00$6314.40
       Follow-up stage II$3301.00$650.50$5941.80
       Follow-up stage III$4624.00$2312.00$8323.20
       Follow-up stage IV$14,005.00$7,002.50$25,209.00
       Final year, stage I$73,383.00$36,691.50$132,089.40
       Final year, stage II$73,091.00$36,545.50$131,563.80
       Final year, stage III$77,100.00$38,550.00$13,8780.00
       Final year, stage IV$101,627.00$50,813.50$182,928.50
       Death other causes, stage I$22,053.00$11,026.50$39,695.40
       Death other causes, stage II$19,872.00$9,936.00$35,769.60
       Death other causes, stage III$24,676.00$12,338.00$44,416.80
       Death other causes, stage IV$57,348.00$28,674.00$103,226.40
      For each parameter group, all indicated parameters were varied in the same direction to the same relative extent, assuming a flat distribution. For the combined variation, all parameters groups were varied in different directions to different extends. The starting value and the minimum and maximum of variations are indicated.
      CRC, Colorectal cancer.
      Supplementary Table 3PolypDx versus low-specificity PolypDx


      Scenario
      PolypDxPolypDx (low spec.)
      100% adherenceReal-world adherence100% adherenceReal-world adherence
      CRC cases (per 1000)34.139.132.736.8
      Screening tests (per 1000)5420444445923995
      Colonoscopies (per 1000)1545122719451560
      Incidence reduction (%)49.441.951.545.5
      CRC deaths (per 1000)10.412.710.211.9
      Mortality reduction (%)59.250.460.353.5
      Total costs (per capita, USD)6206628063366531
      LYG (per 1000)168.7139.4172.4149.0
      dLYG (per 1000)67.655.069.359.0
      dUSD (per dLYG)−2696−4931−586−3408
      We provide outcome parameters of 2 versions of PolypDx, with a high or lower specificity (0.91 vs 0.80 for CRC (for complete test characteristics see Table 1), for 100% adherence and real-world adherence, as indicated. The values of incidence and mortality reduction as well as LYG, dLYG, and dUSD per dLYG are compared with no screening.
      CRC, Colorectal cancer; dLYG, discounted life-year gained; dUSD, discounted U.S. dollars; LYG, life-years gained; USD, U.S. dollars.

      References

        • Bray F.
        • Ferlay J.
        • Soerjomataram I.
        • et al.
        Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries.
        CA Cancer J Clin. 2018; 68: 394-424
        • Brenner H.
        • Kloor M.
        • Pox C.P.
        Colorectal cancer.
        Lancet. 2014; 383: 1490-1502
        • Baxter N.N.
        • Goldwasser M.A.
        • Paszat L.F.
        • et al.
        Association of colonoscopy and death from colorectal cancer.
        Ann Inter Med. 2009; 150: 1-8
        • Hewitson P.
        • Glasziou P.
        • Watson E.
        • et al.
        Cochrane systematic review of colorectal cancer screening using the fecal occult blood test (hemoccult): an update.
        Am J Gastroenterol. 2008; 103: 1541-1549
        • Atkin W.S.
        • Edwards R.
        • Kralj-Hans I.
        • et al.
        Once-only flexible sigmoidoscopy screening in prevention of colorectal cancer: a multicentre randomised controlled trial.
        Lancet. 2010; 375: 1624-1633
        • Pan J.
        • Xin L.
        • Ma Y.F.
        • et al.
        Colonoscopy reduces colorectal cancer incidence and mortality in patients with non-malignant findings: a meta-analysis.
        Am J Gastroenterol. 2016; 111: 355-365
        • Rex D.K.
        • Boland C.R.
        • Dominitz J.A.
        • et al.
        Colorectal cancer screening: recommendations for physicians and patients from the U.S. Multi-Society Task Force on Colorectal Cancer.
        Gastrointest Endosc. 2017; 86: 18-33
        • Sung J.J.
        • Ng S.C.
        • Chan F.K.
        • et al.
        An updated Asia Pacific Consensus Recommendations on colorectal cancer screening.
        Gut. 2015; 64: 121-132
        • Stoop E.M.
        • de Haan M.C.
        • de Wijkerslooth T.R.
        • et al.
        Participation and yield of colonoscopy versus non-cathartic CT colonography in population-based screening for colorectal cancer: a randomised controlled trial.
        Lancet Oncol. 2012; 13: 55-64
        • Quintero E.
        • Castells A.
        • Bujanda L.
        • et al.
        Colonoscopy versus fecal immunochemical testing in colorectal-cancer screening.
        N Engl J Med. 2012; 366: 697-706
        • Siegel R.L.
        • Miller K.D.
        • Fedewa S.A.
        • et al.
        Colorectal cancer statistics, 2017.
        CA Cancer J Clin. 2017; 67: 177-193
        • Chen C.
        • Stock C.
        • Hoffmeister M.
        • et al.
        Public health impact of colonoscopy use on colorectal cancer mortality in Germany and the United States.
        Gastrointest Endosc. 2018; 87: 213-221.e2
        • Segnan N.
        • Senore C.
        • Andreoni B.
        • et al.
        Comparing attendance and detection rate of colonoscopy with sigmoidoscopy and FIT for colorectal cancer screening.
        Gastroenterology. 2007; 132: 2304-2312
        • Jin P.
        • Kang Q.
        • Wang X.
        • et al.
        Performance of a second-generation methylated SEPT9 test in detecting colorectal neoplasm.
        J Gastroenterol Hepatol. 2015; 30: 830-833
      1. A stool DNA test (Cologuard) for colorectal cancer screening.
        JAMA. 2014; 312: 2566
        • Wang H.
        • Tso V.
        • Wong C.
        • et al.
        Development and validation of a highly sensitive urine-based test to identify patients with colonic adenomatous polyps.
        Clin Transl Gastroenterol. 2014; 5 (e54-e)
        • Deng L.
        • Chang D.
        • Foshaug R.R.
        • et al.
        Development and validation of a high-throughput mass spectrometry based urine metabolomic test for the detection of colonic adenomatous polyps.
        Metabolites. 2017; 7: 32
        • Deng L.
        • Fang H.
        • Tso V.K.
        • et al.
        Clinical validation of a novel urine-based metabolomic test for the detection of colonic polyps on Chinese population.
        Int J Colorectal Dis. 2017; 32: 741-743
        • Deng L.
        • Ismond K.P.
        • Liu Z.
        • et al.
        Urinary metabolomics to identify a unique biomarker panel for detecting colorectal cancer: a multicentre study.
        Cancer Epidemiol Biomarkers Prev. 2019; 28: 1283-1291
        • van Roon A.H.
        • Wilschut J.A.
        • Hol L.
        • et al.
        Diagnostic yield improves with collection of 2 samples in fecal immunochemical test screening without affecting attendance.
        Clin Gastroenterol Hepatol. 2011; 9: 333-339
        • van Roon A.H.
        • Goede S.L.
        • van Ballegooijen M.
        • et al.
        Random comparison of repeated faecal immunochemical testing at different intervals for population-based colorectal cancer screening.
        Gut. 2013; 62: 409-415
        • Zubero M.B.
        • Arana-Arri E.
        • Pijoan J.I.
        • et al.
        Population-based colorectal cancer screening: comparison of two fecal occult blood test.
        Front Pharmacol. 2014; 4: 175
        • Singh H.
        • Bernstein C.N.
        • Samadder J.N.
        • et al.
        Screening rates for colorectal cancer in Canada: a cross-sectional study.
        CMAJ Open. 2015; 3: E149-E157
        • Singal A.G.
        • Corley D.A.
        • Kamineni A.
        • et al.
        Patterns and predictors of repeat fecal immunochemical and occult blood test screening in four large health care systems in the United States.
        Am J Gastroenterol. 2018; 113: 746-754
        • Diercks G.F.
        • Janssen W.M.
        • van Boven A.J.
        • et al.
        Rationale, design, and baseline characteristics of a trial of prevention of cardiovascular and renal disease with fosinopril and pravastatin in nonhypertensive, nonhypercholesterolemic subjects with microalbuminuria (the Prevention of REnal and Vascular ENdstage Disease Intervention Trial [PREVEND IT.).
        Am J Cardiol. 2000; 86: 635-638
        • Bello A.K.
        • Peters J.
        • Wight J.
        • et al.
        The Kidney Evaluation and Awareness Program in Sheffield (KEAPS): a community-based screening for microalbuminuria in a British population.
        Nephron Clin Pract. 2010; 116: c95-c103
        • Connolly S.
        • Whyte R.
        Uptake of cancer screening services among middle and older ages in Ireland: the role of healthcare eligibility.
        Public Health. 2019; 173: 42-47
        • Bucholz E.M.
        • Rodday A.M.
        • Kolor K.
        • et al.
        Prevalence and predictors of cholesterol screening, awareness, and statin treatment among us adults with familial hypercholesterolemia or other forms of severe dyslipidemia (1999-2014).
        Circulation. 2018; 137: 2218-2230
        • Buskermolen M.
        • Gini A.
        • Naber S.K.
        • et al.
        Modeling in colorectal cancer screening: assessing external and predictive validity of MISCAN-colon microsimulation model using NORCCAP trial results.
        Med Decis Making. 2018; 38: 917-929
        • van Ballegooijen M.
        • Rutter C.M.
        • Knudsen A.B.
        • et al.
        Clarifying differences in natural history between models of screening: the case of colorectal cancer.
        Med Decis Making. 2011; 31: 540-549
        • Prakash M.K.
        • Lang B.
        • Heinrich H.
        • et al.
        CMOST: an open-source framework for the microsimulation of colorectal cancer screening strategies.
        BMC Med Inform Decis Mak. 2017; 17: 80
        • Jensen C.D.
        • Corley D.A.
        • Quinn V.P.
        • et al.
        Fecal immunochemical test program performance over 4 rounds of annual screening: a retrospective cohort study.
        Ann Intern Med. 2016; 164: 456-463
        • Corley D.A.
        • Jensen C.D.
        • Quinn V.P.
        • et al.
        Association between time to colonoscopy after a positive fecal test result and risk of colorectal cancer and cancer stage at diagnosis.
        JAMA. 2017; 317: 1631-1641
        • Duggan M.A.
        • Anderson W.F.
        • Altekruse S.
        • et al.
        The Surveillance, Epidemiology, and End Results (SEER) program and pathology: toward strengthening the critical relationship.
        Am J Surg Pathol. 2016; 40: e94-e102
        • Imperiale T.F.
        • Ransohoff D.F.
        • Itzkowitz S.H.
        • et al.
        Multitarget stool DNA testing for colorectal-cancer screening.
        N Engl J Med. 2014; 370: 1287-1297
        • Lamb Y.N.
        • Dhillon S.
        Epi proColon® 2.0 CE: a blood-based screening test for colorectal cancer.
        Mol Diagn Ther. 2017; 21: 225-232
        • Song L.
        • Peng X.
        • Li Y.
        • et al.
        The SEPT9 gene methylation assay is capable of detecting colorectal adenoma in opportunistic screening.
        Epigenomics. 2017; 9: 599-610
        • Lieberman D.A.
        • Rex D.K.
        • Winawer S.J.
        • et al.
        Guidelines for colonoscopy surveillance after screening and polypectomy: a consensus update by the US Multi-Society Task Force on Colorectal Cancer.
        Gastroenterology. 2012; 143: 844-857
        • van Hees F.
        • Saini S.D.
        • Lansdorp-Vogelaar I.
        • et al.
        Personalizing colonoscopy screening for elderly individuals based on screening history, cancer risk, and comorbidity status could increase cost effectiveness.
        Gastroenterology. 2015; 149: 1425-1437
      2. US Bureau of Labor 2013. Medical care costs. Medical care in U.S. city average, all urban consumers, not seasonally adjusted 2019. Available at: Statisticshttps://data.bls.gov/timeseries/CUUR0000SAM?output_view=pct_12mths. Accessed September 12, 2019.

      3. Centers for Medicare and Medicaid Services 2016, 18CLABQ2. Available at: https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/ClinicalLabFeeSched/Clinical-Laboratory-Fee-Schedule-Files. Accessed June 12, 2019.

        • Sulz M.C.
        • Kroger A.
        • Prakash M.
        • et al.
        Meta-analysis of the effect of bowel preparation on adenoma detection: early adenomas affected stronger than advanced adenomas.
        PloS One. 2016; 11e0154149
        • Cukier R.
        • Fortuin C.
        • Shuler K.
        • et al.
        Study of the sensitivity of coupled reaction systems to uncertainties in rate coefficients. I theory.
        J Chem Phys. 1973; 59: 3873-3878

      Linked Article

      • Challenges for colorectal cancer screening decision modeling
        Gastrointestinal EndoscopyVol. 94Issue 2
        • Preview
          Colorectal cancer (CRC) screening is effective, but only if it is completed. One of the giants of CRC screening and surveillance, Sidney Winawer, would often say: “The best CRC screening test is the one that gets done.” Is there a “best” CRC screening test? The answer is not known, because few head-to-head studies have been performed. Few comparative studies have measured CRC incidence, mortality, life-years gained, and cost for more than one screening program. Such studies are difficult and costly.
        • Full-Text
        • PDF