Comparative Effectiveness and Cost-Effectiveness of Colorectal Cancer Screening With Blood-Based Biomarkers (Liquid Biopsy) vs Fecal Tests or Colonoscopy

Abstract

BACKGROUND & AIMS:

Colorectal cancer (CRC) screening is highly effective but underused. Blood-based biomarkers (liquid biopsy) could improve screening participation.

METHODS:

Using our established Markov model, screening every 3 years with a blood-based test that meets minimum Centers for Medicare & Medicaid Services’ thresholds (CMS_min) (CRC sensitivity 74%, specificity 90%) was compared with established alternatives. Test attributes were varied in sensitivity analyses.

RESULTS:

CMS_min reduced CRC incidence by 40% and CRC mortality by 52% vs no screening. These reductions were less profound than the 68%–79% and 73%–81%, respectively, achieved with multi-target stool DNA (Cologuard; Exact Sciences) every 3 years, annual fecal immunochemical testing (FIT), or colonoscopy every 10 years. Assuming the same cost as multi-target stool DNA, CMS_min cost $28,500/quality-adjusted life-year gained vs no screening, but FIT, colonoscopy, and multi-target stool DNA were less costly and more effective. CMS_min would match FIT’s clinical outcomes if it achieved 1.4- to 1.8-fold FIT’s participation rate. Advanced precancerous lesion (APL) sensitivity was a key determinant of a test’s effectiveness. A paradigm-changing blood-based test (sensitivity >90% for CRC and 80% for APL; 90% specificity; cost ≤$120–$140) would be cost-effective vs FIT at comparable participation.

CONCLUSIONS:

CMS_min could contribute to CRC control by achieving screening in those who will not use established methods. Substituting blood-based testing for established effective CRC screening methods will require higher CRC and APL sensitivities that deliver programmatic benefits matching those of FIT. High APL sensitivity, which can result in CRC prevention, should be a top priority for screening test developers. APL detection should not be penalized by a definition of test specificity that focuses on CRC only.

Graphical Abstract

Despite the long-standing availability of screening methods that can substantially reduce colorectal cancer (CRC) incidence and mortality,^1,2 many screen-eligible people remain unscreened,³ and it is anticipated that achieving high screening rates in newly eligible 45- to 49-year-olds will take time. Emerging blood-based biomarkers (liquid biopsy) for colorectal neoplasia could potentially increase screening participation.⁴ However, if novel tests do not deliver clinical benefits comparable with those of established strategies like fecal immunochemical testing (FIT) or colonoscopy, substituting them for these accepted strategies could have an adverse impact.

In 2021, the Centers for Medicare & Medicaid Services (CMS) determined that they will cover a blood-based CRC screening test every 3 years if it secures US Food and Drug Administration authorization, and has a sensitivity of ≥74% and a specificity of ≥90% in the detection of CRC.⁵ CMS did not address sensitivity for sub-categories, such as early-stage CRC, or advanced precancerous lesions (APLs), the most consequential targets of screening. A strict interpretation of the CMS definition of “specificity for CRC” would consider detection of APLs as false positives.⁶ The CMS minimum thresholds have become the default targets for screening test development efforts.

We performed a decision analytic modeling study to address the following critical questions that can inform the development and deployment of novel blood-based CRC screening tests: (1) What is the estimated impact of screening with a novel blood-based test that matches the minimum performance thresholds set by CMS, and how does this compare with FIT, colonoscopy, or a multi-target stool DNA (MT-sDNA) test (Cologuard; Exact Sciences)? (2) How would differential participation rates by strategy affect comparative effectiveness and cost-effectiveness? (3) How important is detection of APLs, what are the consequences of test positivity in the presence of nonadvanced polyps, and what are the implications for the definition of test specificity? (4) What attributes would be required to have a truly paradigm-changing blood-based CRC screening test?

Our analysis focused on blood-based screening for colorectal neoplasia only. We did not address the complex questions surrounding multi-cancer early detection (MCED) tests, including the balance of potential benefits, harms, and costs of screening simultaneously across multiple cancer types. However, our CRC-specific results apply to blood-based tests designed exclusively for colorectal neoplasia, as well as MCED tests that include CRC among the targets.

Methods

We performed a decision and cost-effectiveness analysis using an updated version of our validated model of CRC screening,^6–14 now named the Model of Screening and Surveillance for Colorectal Cancer (MOSAIC), version 2023.1 (Appendix 1). The model was constructed using TreeAge Pro 2023 (TreeAge Software, Williamstown, MA) and outputs were analyzed using Excel (Microsoft Corporation, Redmond, WA). In order to determine how to model the sensitivity of blood-based biomarkers for CRC, we performed explorations of the study sample size and power that would be needed to contrast CRC stage–specific test sensitivities using R statistical programming languages, version 4.2.1 (R Core Team, 2022).

Decision-Analytic Model: Model of Screening and Surveillance for Colorectal Cancer, Version 2023.1

The details of our model^7–13 and validations¹⁵ against randomized controlled trials of fecal occult blood testing^16,17 and sigmoidoscopy,^18–20 and the concordance of its projections with those of other established models^14,21 have been detailed in previous publications. For MOSAIC version 2023.1, we performed a comprehensive upgrade, including recalibration to contemporary data on nonadvanced adenoma and APL prevalence by age²² as opposed to older autopsy series, detailed calibration to post-polypectomy data on metachronous adenoma and advanced adenoma detection rates,^23–30 additional validations against post-polypectomy rates of metachronous CRC incidence and CRC mortality,^31–34 incorporation of the most recent stage-specific CRC survival data,³⁵ and use of the longer post-polypectomy surveillance intervals endorsed in the latest US Multi-Society Task Force on Colorectal Cancer guidelines.³⁶ This comprehensive upgrade allowed exploration of a broader range of analyses with increased confidence in the outputs.

Appendix 1 describes the model and upgrade in detail. In brief, the Natural History module reproduces the natural history and age-specific incidence and prevalence of nonadvanced colorectal adenomas, APLs, and CRC by stage in the United States without screening.^7,9,11 People transition between health states of normal, nonadvanced polyp, APL, localized, regional or disseminated CRC, and death, in 1-year cycles. Screening and post-polypectomy surveillance are superimposed on the Natural History module, accounting for test performance characteristics, complications, and costs. Age-specific non-CRC mortality is accounted for. People are followed until age 100 years or death.

Screening strategies are superimposed on the Natural History module from age 45 through 75 years, with post-polypectomy surveillance through age 80 years. A strategy’s impact is determined by test performance characteristics, screening interval, screening ages, cost, and participation rates.

Model inputs are presented in Appendix 2 Tables 1A and B.

The current model version does not explicitly consider a separate sessile serrated lesion (SSL) to CRC pathway. However, the APL health state reflects the age-specific prevalence of advanced adenomas and advanced SSLs combined.²² In sensitivity analyses, we approximated the potential impact of contrasting test sensitivities for adenomas with SSLs based on assumptions about the fraction of CRCs that arise from SSLs (Appendix 1).

Blood-Based Biomarker Sensitivity for Colorectal Cancer

We faced a fundamental decision on whether to model different stage-specific CRC sensitivities for blood-based biomarkers or simply an overall sensitivity for all CRCs. To inform this decision, we performed the following 2 sets of analyses: (1) an exploration of the feasibility of demonstrating significant differences in CRC stage–specific test sensitivities, focusing on the sample sizes that would be required to distinguish early-stage vs late-stage CRC sensitivities in prospective studies comparing blood-based testing with colonoscopy and (2) an exploration of how much of an impact there would be on CRC outcomes if blood-based biomarkers achieved a given overall CRC sensitivity with different underlying stage-specific test sensitivities. The details are in Appendix 2.

First, we determined the statistical power of hypothetical studies with 10,000–80,000 subjects to detect significant differences in test sensitivity for early-stage (stages I and II) vs late-stage (stages III and IV) CRC. The relevant factors included CRC prevalence (0.5% vs 0.3%), CRC stage distribution (with the ratio of early-stage to late-stage CRC informed by large studies of noninvasive screening compared with colonoscopy as a gold standard [78:22,³⁷ 68:32,^37–39 or 58:42³⁹]) (Appendix 2 Table 1C), a range of assumed differences in true sensitivity for early-stage vs late-stage CRC (62% vs 100%, 65% vs 93%, or 68% vs 87%, respectively), and an assumed overall CRC sensitivity of approximately 74%, the minimum set by CMS (Appendix 2 Tables 2A and B and Appendix 2 Figures 1A–F). Under numerous plausible assumptions, >30,000–70,000 subjects would be required in order to have at least 80% power to detect a significant difference between early-stage vs late-stage CRC test sensitivity (Appendix 2 Figure 1A). The conclusion was that the sample sizes of current and anticipated studies in this field (unlikely to include hundreds of CRCs) are unlikely to have the power to distinguish early-stage vs late-stage CRC sensitivities (eg, Appendix 2 Table 3).

Second, we used MOSAIC, version 2023.1 to compare clinical outcomes between hypothetical blood-based tests that meet the CMS minimum overall CRC sensitivity of 74%, but with varying stage-specific sensitivities, and 90% CRC specificity. The differences in predicted outcomes were very small (Appendix 2 Table 4).

Given the results of both of these sets of analyses, we modeled an overall CRC sensitivity for each blood-based biomarker scenario in all subsequent analyses.

Clinical and Economic Outcomes, Perspective, and Cost-Effectiveness Analyses

The principal model outputs were CRC cases and deaths, CRC stage distribution (localized, regional, or disseminated), quality-adjusted life-years (QALYs), costs, and required number of colonoscopies and noninvasive tests in a hypothetical cohort of 100,000 people starting at age 45 years.⁴⁰ Future QALYs and costs were discounted by 3% annually.⁴¹ Health state utilities for CRC by stage were applied for 5 years after CRC diagnosis. Analyses were performed using a health sector perspective.⁴⁰

Base-case cost inputs were derived from 2023 CMS reimbursement rates^42–44 and the most recent data on stage-specific CRC care costs.⁴⁵ In the base case, we assumed a blood-based test cost equivalent to that of MT-sDNA. As in our previous analyses, we assumed higher costs at ages <65 years (commercial insurance vs Medicare)⁴⁶ by factors of 1.8 for cost of colonoscopy with or without polypectomy and 1.76 for CRC care and complication costs based on our previous studies of commercial payments compared with CMS rates.^46,47 Costs were updated to 2023 dollars using the medical care component of the Consumer Price Index.

The incremental cost-effectiveness ratio was used to compare the cost/QALY gained between strategies. The common US willingness-to-pay thresholds of $100,000–$150,000/QALY gained^48,49 were used to designate interventions as cost-effective.

Screening Strategies

In the base case, we focused on a hypothetical blood-based test performed every 3 years that meets the minimum criteria set by CMS (ie, CRC sensitivity of 74% and CRC specificity of 90%, which implies a 10% positivity rate in people with a normal colon, nonadvanced polyp, or APL). The principal comparators were annual FIT, colonoscopy every 10 years, or MT-sDNA every 3 years, the strategies recommended by the US Preventive Services Task Force⁵⁰ that account for the vast majority of CRC screening in the United States.^51,52 We also modeled FIT every 2 years, blood-based testing for methylated Septin9 (mSep9), or the emerging Guardant Shield (Guardant Health) test (based on preliminary test performance data³⁹) every 3 years. We explored the impact of shortening the blood-based test’s interval to every 2 years or yearly.

Comparative Effectiveness of Blood-Based Tests vs Established Strategies, and Impact of Participation

We first compared clinical outcomes between strategies assuming full participation over time. Because blood-based testing could have advantages over stool-based testing or colonoscopy in terms of patient acceptance and utilization, we then performed threshold analyses to determine the relative participation rates at which a blood-based test that meets the minimum CMS thresholds would match the clinical outcomes with annual FIT. We made the simplifying assumption that people either participate fully over time or not at all, expressed as the participation rate in the population with a given strategy. We explored the impact of imperfect follow-up with colonoscopy after noninvasive screening (follow-up rates of 60%, 75%, and 90%), including differential rates across strategies.

Addition of New Screenees vs Substitution for Established Strategies

We explored the potential population-level impact of capturing new screenees with a novel test (“addition” effect) vs substituting a novel test for established alternatives (“substitution” effect). We constructed hypothetical scenarios reflecting varying fractions of the population participating with annual FIT, colonoscopy, MT-sDNA, a blood-based test that meets the CMS minimum performance thresholds, or remaining unscreened.

We anchored the comparisons on a hypothetical current state of 10% participation with annual FIT, 40% with colonoscopy, and 10% with MT-sDNA (ie, 60% of the population screened and 40% unscreened). We modeled the following 3 scenarios in which an absolute 20% of the population took up novel blood-based testing: (1) addition of new screenees only (one-half of the 40% previously unscreened people take up novel blood-based testing); (2) substitution only (one-half of the 10% previously undergoing FIT, one-half of the 10% previously undergoing MT-sDNA, and one-quarter of the 40% previously undergoing colonoscopy switch to novel blood-based testing); and (3) a combination of addition and substitution of equal absolute magnitude (one-quarter of the 40% previously unscreened people take up novel blood-based testing; one-quarter of the 10% previously undergoing FIT, one-quarter of the 10% previously undergoing MT-sDNA, and one-eighth of the 40% previously undergoing colonoscopy switch to novel blood-based testing).

Cost-Effectiveness of Blood-Based Tests vs Established Strategies, and Impact of Participation

We calculated incremental cost-effectiveness ratios vs no screening or vs alternative strategies, using annual FIT as the reference standard in the base case. We then performed threshold analyses, as described above for comparative effectiveness, to determine the relative participation rates at which a blood-based test that meets the minimum CMS thresholds could be considered cost-effective vs annual FIT.

Detection of Advanced Precancerous Lesions and Nonadvanced Polyps, and a Hypothetical Paradigm-Changing Blood-Based Test

First, we explored the impact of improving a blood-based test’s sensitivity for APL vs improving its sensitivity for CRC. Second, we reasoned that if biomarkers to detect APL can be discovered, they might also serendipitously increase the positivity rate in people with nonadvanced polyps, even if nonadvanced polyps are not an intentional target of screening. We therefore explored the impact of increasing the positivity rate for nonadvanced polyps once the necessary APL sensitivity had been achieved to match or exceed the programmatic effectiveness of annual FIT. The purpose of this second exercise was to inform the discussion of how to define test specificity.

We constructed hypothetical sets of blood-based test performance characteristics that would match the clinical effectiveness of annual FIT. Based on these, we performed threshold analyses to determine the test cost at which such a paradigm-changing blood-based test would be cost-equivalent or cost-effective vs annual FIT.

Number of Tests Required

For each strategy, we determined the mean number of colonoscopies and noninvasive tests that would be required per person over a lifetime, assuming full participation over time.

Sensitivity Analyses

In sensitivity analyses focused on the potential importance of the serrated pathway, we explored scenarios reflecting 15% or as high as 25% of CRCs arising from SSLs, assuming that blood-based tests do not currently detect SSLs, but can be positive by chance in the presence of SSLs, based on specificity (Appendix 1). In addition to the multiple threshold and sensitivity analyses described above relating to a blood-based test’s attributes, we performed 1-way sensitivity analyses and a probabilistic Monte Carlo simulation with 10,000 iterations in which we varied all other model inputs simultaneously, as described previously.¹³ We varied all model inputs simultaneously using β distributions for probabilities derived from means, SDs, and ranges in the literature. Costs of screening were varied by a common factor within a range of 20% of the base-case value, and costs of care by a different common factor within the same range. We focused on the comparison between a blood-based test with the minimum CMS specifications vs annual FIT.

Results

Base Case: Clinical Outcomes

Screening every 3 years with a blood-based test with the minimum performance thresholds set by CMS reduced CRC incidence by 40% and CRC mortality by 52% vs no screening (Table 1, Figure 1 [CMS minimum every 3 years]). These reductions were less profound than those achieved with annual FIT (72% and 76%, respectively), colonoscopy every 10 years (79% and 81%, respectively), or MT-sDNA every 3 years (68% and 73%, respectively), and translated into fewer QALYs/person with blood-based screening than with the alternatives (Table 1, Figure 1). Screening with such a blood-based test every year (an interval not currently endorsed by CMS) achieved clinical results that approached those of MT-sDNA (Table 1).

Table 1.

Clinical and Economic Outcomes of Screening Strategies in a Hypothetical Cohort of 100,000 People Starting at Age 45 Years

Screening strategya	Clinical outcomes						Economic outcomes
	CRC distribution, n (% of total CRCs)			CRC incidence, n (% reduction vs no screening)	CRC mortality, n (% reduction vs no screening)	Quality-adjusted life expectancy (discounted, from age 45 y)	Overall cost (discounted)	Cost-effectivenessb
	Localized	Regional	Disseminated	Total CRC	CRC deaths	QALYs/person	Cost ($)/person	Cost ($)/QALY gained vs no screen	Cost ($)/QALY gained vs FIT yearly
No screen	2912 (39)	2907 (39)	1651 (22)	7470	3624	21.2865	5965	NA	NA
FIT yearly	1246 (59)	601 (28)	270 (13)	2117 (72)	868 (76)	21.3828	3529	FIT dominates	NA
FIT every 2 y	1657 (59)	815 (29)	353 (12)	2824 (62)	1133 (69)	21.3758	3367	FIT dominates	NA
Colonoscopy every 10 y	794 (52)	525 (34)	222 (14)	1541 (79)	672 (81)	21.3845	5207	Colonoscopy dominates	958,000
MT-sDNAc every 3 y	1314 (56)	740 (31)	301 (13)	2355 (68)	970 (73)	21.3792	6550	6300	FIT dominates
Blood-based test (CMS minimumd)
Every 3 y	2484 (55)	1439 (32)	576 (13)	4499 (40)	1754 (52)	21.3542	7891	28,500	FIT dominates
Every 2 y	2316 (62)	1019 (27)	430 (11)	3766 (50)	1390 (62)	21.3664	8662	33,800	FIT dominates
Every year	1680 (63)	681 (26)	304 (11)	2665 (64)	988 (73)	21.3768	10,624	51,600	FIT dominates
mSep9 every 3 y	1538 (53)	972 (33)	407 (14)	2917 (61)	1184 (67)	21.3661	6329	4600	FIT dominates
Guardant Shield every 3 y	2217 (53)	1422 (34)	506 (12)	4146 (45)	1622 (55)	21.3585	7664	23,600	FIT dominates
Blood-based test, high sensitivities,e every 3 y	971 (53)	602 (33)	251 (14)	1824 (76)	785 (78)	21.3837	6299	3400	$3.1 million
Blood-based test, high sensitivities adjusted down to meet CRC/APL-specificity,f every 3 y	1381 (53)	853 (33)	353 (14)	2587 (65)	1080 (70)	21.3738	6724	8700	FIT dominates

NA, not applicable; QALY, quality-adjusted life-year (discounted).

^aOne time test costs at ages <65 y/ages ≥65 y are FIT $18/$18; colonoscopy without polypectomy $1415/$786; MT-sDNA $681/$509; blood-based test in the base case assumed to be the same as MT-sDNA; mSep9 $498/$192; Guardant Shield assumed to be the same as MT-sDNA.

^bDominates = more effective (more QALYs/person) and less costly.

^cCologuard (Exact Sciences).

^dSensitivity for CRC was 74%, specificity for CRC was 90% (ie, positivity rate was 10% in people with normal colon, APL, or nonadvanced polyps).

^eSensitivity for CRC was 90%, APL was 70%, nonadvanced polyp was 30%, positivity rate was 10% in people with normal colon (ie, specificity was 90% in these people).

^fSensitivities adjusted down per receiver-operating characteristics curve to meet requirement for 10% positivity in people with normal colon or only nonadvanced polyps (ie, specificity was 90% in these people [positivity rate was 5% in people with normal colon and 15% in people with nonadvanced polyp]); resulting sensitivity for CRC was 75%, APL was 35%.

Figure 1.

Clinical outcomes with established CRC screening tests and with emerging blood-based biomarkers. The minimum thresholds set by CMS are 74% sensitivity and 90% specificity for CRC (CMS minimum). QALYs are discounted and they are from age 45 years.

Base Case: Cost-Effectiveness

Assuming that a blood-based test with the minimum performance thresholds set by CMS has a test cost equal to that of MT-sDNA, screening every 3 years with that test had a cost/QALY gained of $28,500 vs no screening (Table 1, Figure 2). However, screening with annual FIT, colonoscopy, or MT-sDNA was more effective and less costly and, therefore, these strategies were dominant over such a blood-based test (Table 1, Figure 2). Annual FIT cost $23,400/QALY gained vs FIT every 2 years.

Figure 2.

Discounted mean cost per person and QALYs per person with established CRC screening tests and with emerging blood-based biomarkers. The minimum thresholds set by the CMS are 74% sensitivity and 90% specificity for CRC (CMS minimum).

Impact of Differential Participation Rates by Strategy

In a screen-eligible cohort that included people who participated fully with screening over time (defined as the participation rate), and the rest who never participated, the participation rate with a blood-based test with the minimum performance thresholds set by CMS offered every 3 years would have to be approximately 1.8-fold, 1.5-fold, and 1.4-fold the rate with FIT in order to match annual FIT’s impact on CRC incidence, CRC mortality, and QALYs/person, respectively (Table 2 and Appendix 2 Figure 2). For example, compared with annual FIT with a participation rate of 40%, a blood-based test with the minimum performance thresholds set by CMS offered every 3 years would need to achieve a participation rate of 72% in order to match the number of CRC cases prevented with FIT, a participation rate of 59% in order to match the number of CRC deaths prevented with FIT, and a participation rate of 57% in order to match the number of QALYs gained with FIT (Table 2). At FIT participation rates >60%–70%, such a blood-based test could not yield matching clinical benefits even with 100% participation (Table 2).

Table 2.

Overall All-or-None^a Participation Rates Required for Screening Every 3 Years With a Blood-Based Test That Meets Centers for Medicare & Medicaid Services’ Minimum Performance Thresholds in Order for That Strategy to Match the Clinical Outcomes of Annual Fecal Immunochemical Testing at Varying Levels of All-or-None Participation Rates With Fecal Immunochemical Testing

Variable	Participation rate (%) with blood-based test every 3 y that yields outcomes equivalent to annual FIT (all-or-none over time)
	Overall FIT participation rate of 10%	Overall FIT participation rate of 20%	Overall FIT participation rate of 30%	Overall FIT participation rate of 40%	Overall FIT participation rate of 50%	Overall FIT participation rate of 60%	Overall FIT participation rate of 70%
CRC cases prevented	18	36	54	72	90	Blood test cannot match FIT	Blood test cannot match FIT
CRC deaths prevented	15	29	44	59	74	88	Blood test cannot match FIT
QALYs gained vs no screening	14	28	43	57	71	85	Blood test cannot match FIT

QALYs, quality-adjusted life-years (discounted).

^aFor illustrative purposes, scenarios reflect perfect participation with every screening round over time in a given fraction of the population with a given test (defined as “participation rate”), and no screening at all in the remainder.

Assuming a cost for such a blood-based test that is equal to that of MT-sDNA, the differentials in participation rates that would need to be realized by the blood-based test vs annual FIT in order to cost <$100,000–$150,000/QALY gained vs annual FIT were more extreme (Appendix 2 Figure 3). For instance, compared with annual FIT with a 30% participation rate, blood-based screening would need to achieve participation rates >75% in order to cost <$100,000/QALY gained (Appendix 2 Figure 3).

The importance of the colonoscopy follow-up rate after an abnormal screening test is highlighted in Appendix 2 Table 5. For CRC outcomes, failures at this critical step in the screening continuum are equivalent to failing to screen. The economic impact reflects a balance between wasted screening costs and missed opportunities to decrease costs of CRC care vs averted costs for those follow-up colonoscopies and polypectomies that would not affect CRC outcomes.

Addition of New Screenees vs Substitution for Established Strategies

Table 3 illustrates the contrasting potential population-wide impact of a novel blood-based test depending on whether it leads to addition of new screenees, substitution for existing alternatives, or a combination. For a blood-based test with the minimum performance thresholds set by CMS performed every 3 years, a pure addition effect improved population outcomes, a pure substitution effect worsened them, and the combination of equal addition and substitution effects improved outcomes but to a lesser degree than with a pure addition effect (Table 3).

Table 3.

Population-Level Impact of Addition of New Screenees, Substitution for Other Established Tests, or a Combination of Addition and Substitution With a Blood-Based Test That Meets Centers for Medicare & Medicaid Services’ Minimum Performance Thresholds

Scenarios for test utilization at the population level						Clinical outcomes
	Fraction of the population using a particular screening test (%)a					CRC distribution, n (% of total CRCs)			CRC incidence, n (% reduction vs no screening)	CRC mortality, n (% reduction vs no screening)	Quality-adjusted life expectancy (discounted, from age 45 y)
	FIT	Colonoscopy	MT-sDNA	CMSmin	No screening test	Localized	Regional	Disseminated	Total CRC	CRC deaths	QALYs/person
No screening	0	0	0	0	100	2912 (39)	2907 (39)	1651 (22)	7470	3624	21.2865
Hypothetical current state	10	40	10	0	40	1738 (43)	1507 (37)	806 (20)	4052 (46)	1902 (48)	21.3446
Only addition of new screenees with CMSmin	10	40	10	20	20	1653 (48)	1213 (35)	591 (17)	3457 (54)	1528 (58)	21.3581
Only substitution for other tests with CMSmin	5	30	5	20	40	2028 (44)	1675 (37)	871 (19)	4574 (39)	2094 (42)	21.3389
Addition of new screenees and substitution for other tests with CMSmin	7.5	35	7.5	20	30	1840 (46)	1444 (36)	731 (18)	4016 (46)	1811 (50)	21.3485

CMS_min, blood-based test that meets the minimum CMS test performance thresholds every 3 years.

^aAnnual FIT; colonoscopy every 10 years; MT-sDNA (Cologuard; Exact Sciences) every 3 years.

Importance of Detecting Advanced Precancerous Lesions

A blood-based test’s clinical benefits increased substantially as its sensitivity for APLs improved (Figure 3 and Appendix 2 Figures 4A and B). For instance, the QALYs/person achieved with annual FIT, with an assumed APL sensitivity of 24%, could be matched by a blood-based test performed every 3 years with the minimum CMS CRC sensitivity of 74% if the blood-based test also had an APL sensitivity >70% (assuming specificity of 90%, defined here as a false-positive rate of 10% in people with a normal colon or only nonadvanced polyps).

Figure 3.

Impact of sensitivity for APL on the effectiveness of a blood-based test performed every 3 years. Improving APL sensitivity had a much greater impact than improving CRC sensitivity. FIT screening is annual. Colonoscopy screening is every 10 years.

The impact on clinical outcomes of improvements in APL sensitivity were substantially more dramatic than the impact of improving the blood-based test’s sensitivity for CRC, even to a level as high as 95% (Figure 3 and Appendix 2 Figures 4A and B). Across the range of blood test sensitivity for CRC of 74%–95%, the blood test sensitivity for APL that would be required in order to match the QALYs/person with annual FIT was in the range of 70%–80% (Figure 3).

A blood-based test administered every 3 years could match the QALYs/person achieved with annual FIT if the blood-based test’s CRC and APL sensitivities were, for instance, 95% and 67.2%, respectively; 90% and 71.2%, respectively; 85% and 75.3%, respectively; or 80% and 79.6%, respectively (assuming specificity of 90%, defined here as positivity rates of 10%, given a normal colon or only nonadvanced polyps) (Appendix 2 Figure 5A).

Impact of Increased Positivity Rate With Nonadvanced Polyps

Given a blood-based test with CRC and APL sensitivities that yielded results comparable to those with annual FIT, increasing such a test’s positivity rate in the presence of nonadvanced polyps resulted in relatively small marginal benefits (Appendix 2 Figures 5A–C). The impact of improving APL sensitivity (Figure 3), as described above, was much more substantial than the impact of increasing the positivity rate in the presence of nonadvanced polyps once high APL sensitivity has been achieved.

Current Tests: Methylated Septin9 and Guardant Shield

Screening every 3 years with mSep9 (previously marketed as Epi proColon [Epigenomics AG] in the United States) yielded greater clinical benefits than screening with a blood-based test with the minimum performance thresholds set by CMS (Table 1, Figures 1–2). This result, which may seem paradoxical, given mSep9’s lower CRC sensitivity and specificity, is explained by the fact that lower specificity leads to more colonoscopies and incidental detection of APLs, which has a substantial impact on clinical outcomes (see above).

A blood-based test with the performance characteristics reported preliminarily for Guardant Shield yielded clinical benefits that were only slightly higher than those with a test meeting minimum CMS thresholds, but lower than those with mSep9 (Table 1, Figures 1–2).

A Hypothetical Paradigm-Changing Blood-Based Test and Impact of Test Cost

Screening every 3 years with a hypothetical blood-based test with sensitivities for CRC, APL, and nonadvanced polyps of 90%, 70%, and 30%, respectively, and specificity of 90% (defined here as a positivity rate of 10% in people with a normal colon), yielded clinical benefits similar to those of annual FIT or colonoscopy (Table 1).

However, in order for blood-based tests with high sensitivity (90%–95% for CRC; 80% for APL) and 90% specificity (defined as 10% false-positive rate in people with a normal colon) to be cost-effective at thresholds of $100,000–$150,000/QALY gained vs annual FIT, the blood-based tests would need to cost ≤$120–$140 (Appendix 2 Table 6 and Appendix 2 Figure 6). In order to match the total costs with annual FIT, the blood-based tests would need to cost approximately $100–$110 (Appendix 2 Table 7 and Appendix 2 Figure 7).

Required Numbers of Colonoscopies and Noninvasive Tests

With screening at ages 45–75 years, and post-polypectomy surveillance through age 80 years, the mean number of colonoscopies required per person was 4.0 with screening colonoscopy, 2.0 with annual FIT, 1.9 with MT-sDNA, and ranged from approximately 1.5 to 2.5 for the various blood-based screening strategies (Table 4). One consequence of mSep9’s comparatively lower specificity was a higher colonoscopy requirement vs other blood-based tests.

Table 4.

Total Number of Colonoscopies and Noninvasive Tests Required per Person Starting at Age 45 Years Depending on Screening Strategy (Assuming Full Participation Over Time With All Screening, Follow-Up, and Surveillance)

Screening strategy	No. of colonoscopies per person	No. of stool or blood-based tests per person
No screen	0.2	0
FIT yearly	2.0	16.9
FIT every 2 y	1.5	10.8
Colonoscopy every 10 y	4.0	0
MT-sDNAa every 3 y	1.9	6.5
Blood-based test (CMS minimumb)
Every 3 y	1.5	7.3
Every 2 y	1.8	9.5
Every year	2.5	13.3
mSep9 every 3 y	2.3	5.7
Guardant Shield every 3 y	1.5	7.3
Blood-based test, high sensitivities,c every 3 y	2.2	5.8
Blood-based test, high sensitivities adjusted down to meet CRC/APL-specificity,d every 3 y	1.6	7.1

^aCologuard (Exact Sciences).

^bSensitivity for CRC was 74%, specificity for CRC was 90% (ie, positivity rate was 10% in people with normal colon, APL, or nonadvanced polyps).

^cSensitivity for CRC was 90%, APL was 70%, nonadvanced polyp was 30%, positivity rate was 10% in people with normal colon (ie, specificity was 90% in these people).

^dSensitivities adjusted down per receiver-operating characteristics curve to meet requirement for 10% positivity in people with normal colon or only nonadvanced polyps (ie, specificity was 90% in these people [positivity rate was 5% in people with normal colon, 15% in people with nonadvanced polyp]); resulting sensitivity for CRC was 75% and 35% for APL.

Table 4 shows the mean number of noninvasive blood-based or stool-based tests per person required over a lifetime in order to maintain full longitudinal screening participation.

Sensitivity Analyses

The change in results was negligible in scenarios when 15% or up to 25% of CRCs arose from SSLs (Appendix 2 Table 8). In 1-way sensitivity analyses, the results changed minimally for a blood-based test with the minimum CMS thresholds vs no screening, and annual FIT remained dominant (Appendix 2 Table 9). In the Monte Carlo simulation, a blood-based test with the minimum CMS thresholds cost a mean of $28,300 (95% of iterations, $22,800–$34,400) per QALY gained vs no screening. However, at willingness-to-pay thresholds of up to $300,000/QALY gained, annual FIT was dominant over the blood-based test, as well as all other strategies, in 100% of iterations.

Discussion

The conclusions of our decision analysis of CRC screening with blood-based biomarkers include: (1) the estimated clinical benefit of screening with a blood-based test with the minimum CMS thresholds could be large compared with no screening, at an acceptable cost per QALY gained; however, FIT, colonoscopy, and MT-sDNA are estimated to yield substantially higher clinical benefit and to be dominant vs a blood-based test with those minimum-threshold test performance characteristics in terms of cost-effectiveness, assuming comparable participation rates; (2) if a blood-based test with the minimum CMS thresholds could achieve participation levels that are substantially, and perhaps unrealistically, higher vs those of annual FIT, then it could potentially match its effectiveness and cost-effectiveness (eg, 75% with blood-based test vs 30% with annual FIT); however, in settings with annual FIT participation rates >60%–70% (which is unrealistic in opportunistic screening), even a 100% participation rate with a blood test with the minimum CMS thresholds could not deliver outcomes comparable to those with annual FIT; (3) APL sensitivity is a critical determinant of a screening test’s clinical benefits and improving this parameter is likely to have a much more profound impact than improving CRC sensitivity; and (4) a blood-based screening test with CRC and APL sensitivities of 90%–95% and 70%–80%, respectively, could be paradigm-shifting, but in order to be considered cost-effective vs annual FIT, its cost would need to be ≤$120–$140.

The most critical issue is the contrasting consequences of screening previously unscreened people who will not take up any of the established screening methods (an “addition” effect) vs shifting the screening method in those who would be willing to use an established effective strategy (a “substitution” effect). Addition of screenees who would not be screened with stool tests or colonoscopy would improve outcomes at acceptable costs, even if the novel tests just matched the minimum CMS thresholds at a test cost equal to that of MT-sDNA. However, substitution of established, effective alternatives with a blood-based test with the minimum CMS performance thresholds would worsen outcomes. In order to contemplate the substitution scenario, the programmatic benefits of a novel test must, at minimum, match the programmatic benefits of the established alternatives.

APL sensitivity emerged as a critical determinant of a test’s programmatic benefit over time. We believe that research and development efforts should emphasize APL detection. A useful rule of thumb is that a blood-based test done every 3 years might yield similar benefits to those of annual FIT if it achieves better CRC sensitivity than FIT (eg, 85%–90% vs 74%) and APL sensitivity that is approximately 3-fold that of FIT (eg, 70%–75% vs FIT’s sensitivity of 24% for advanced adenoma). This comparison assumes independence in test performance from cycle to cycle, in which case annual FIT would detect 24%, 67%, and 85% of dwelling advanced adenomas by cycles 1, 4, and 7, respectively. Preliminary data for Guardant Shield include sensitivity of 13% for APL.³⁹ Early-stage reports for other assays include higher APL sensitivity,⁵³ but larger studies and prospective confirmation are needed.

Given the importance of APL detection, test specificity should not be defined in a way that penalizes APL detection.⁶ Defining a minimum “specificity for CRC” treats detection of APLs as a false positive. Whether or not this is inadvertent, it is not desirable. We have previously explored how specificity can be a confusing term when there are more than just the 2 health states of “disease” and “not disease,” such as when “disease” can be defined as cancer or significant cancer precursors.⁶ In the current analysis, we further considered that APL detection may come at the expense of higher positivity rates for nonadvanced polyps, even if these are not intentional targets of screening. We would oppose an overall downward adjustment in test sensitivities based on receiver-operator characteristic curves,^37,54 in order to match a certain “CRC specificity” or even “CRC or APL specificity” because such manipulations of test thresholds would worsen outcomes (Table 1). Instead, we propose the alternative of establishing a minimum acceptable false-positive rate in people with a normal colon (eg, 10%, which is in the spirit of the CMS “90% specificity” threshold). This is an easily understood metric that is not subject to the confusion surrounding use of the term minimum specificity.

Longitudinal participation determines a strategy’s real-world effectiveness. It will take time to characterize blood-based test utilization patterns by clinicians and participation patterns by patients. For those receiving routine medical care, including blood tests, adding a blood-based CRC screening test would be easy. For those without such medical contact, mailed FIT has advantages. Whether ambulatory phlebotomy is viable remains to be determined. Critical questions include the following: Will clinicians and patients trust a negative colonoscopy after an abnormal blood test that includes genomic markers? Will there be a temptation to order blood-based screening more often than every 3 years, and will payers cover this? Will the tests perform similarly in younger and older people (eg, DNA methylation accumulates with age)? Will completion of follow-up colonoscopy after an abnormal primary screening test, which is a critical step in the screening continuum,²¹ differ across screening tests?

The cost of a novel test is a major determinant of its cost-effectiveness, as would be expected. A useful rule of thumb is that a novel test’s cost could be up to 3- to 5-fold that of FIT (current CMS reimbursement for FIT is $18.05) in order for it to be cost-equivalent or cost-effective vs FIT at traditional willingness-to-pay thresholds, assuming similar effectiveness and similar overall costs for infrastructure, outreach, and navigation. In the United States, colonoscopy and MT-sDNA are accepted, despite their high costs relative to FIT, in part because patient and physician preferences are accommodated and because most CRC screening is opportunistic. In the case of colonoscopy, its performance characteristics include high sensitivity for both CRC and APL, its screening interval is much longer than with annual FIT, and it is highly cost-effective or even cost-saving compared with no screening. Whether competition will drive down the costs of blood-based or stool-based tests⁵⁴ remains to be seen. In people who adhere with FIT over time, it is challenging for novel tests to compete with the cost-effectiveness of FIT at a cost of <$20.

CMS has not addressed the issue of early-stage CRC sensitivity. It is logical to demand that a novel test detect early-stage CRC because detecting disseminated CRC is unlikely to yield a mortality benefit. However, our analyses set realistic expectations about the level of evidence we are likely to have available regarding test sensitivity for CRC by stage. The study sample sizes required to demonstrate differences in early-stage vs late-stage CRC sensitivities seem prohibitively large (Appendix 2). The stage-specific CRC sensitivity estimates that we will have available are likely to have very wide confidence intervals. Fortunately, our modeling work suggests that determining overall CRC sensitivity is adequate. Furthermore, the higher a test’s overall CRC sensitivity is, the less likely it becomes that very high late-stage CRC sensitivity will mask a comparatively lower early-stage CRC sensitivity.

We modeled blood-based tests that are designed to detect colorectal neoplasia only, and not cancers other than CRC. However, our results have relevance for MCED tests. A “substitution” effect is a particular concern in this context. If MCED tests cannot match the programmatic benefits of established CRC screening methods, then their use would need to be coupled with established forms of CRC screening. Our current study cannot address key questions related to MCED tests, including the potential benefits and harms across cancer types, concerns about tissue-of-origin specificity, and the potential of a diagnostic odyssey after a positive result, including its potential harms and costs.

Our model’s results for CRC incidence and mortality reductions result from assumptions about test sensitivity and specificity for CRC precursors and CRC, and underlying assumptions about the natural history of CRC. Our model’s extensive validations against the results of randomized controlled trials^16–20 and studies of post-colonoscopy metachronous CRC incidence and mortality^31–34 (Appendix 1) provide confidence in our model’s predictions. However, our base-case results are best viewed as idealized estimates of efficacy with optimal implementation and participation. Real-world effectiveness is likely to be lower.

Our study has limitations. We did not model intermittent participation patterns over time (eg, dropoffs, inconsistent participants, or late entrants). We did not include costs for navigation or outreach. We did not explicitly consider an SSL to CRC pathway in the core analyses, but in sensitivity analyses, assuming that 15% and up to 25% of CRCs arise from SSLs, the conclusions were not affected. We did not consider any potential extracolonic testing, or sooner-than-recommended repeat screening, which might be pursued by clinicians after a positive blood-based test and normal colonoscopy. We did not model hybrid strategies with noninvasive screening at younger ages and colonoscopy at older ages; however, the findings of previous modeling of hybrid strategies including fecal tests also apply to hybrid strategies including blood-based tests.^14,55,56

In conclusion, blood-based CRC screening tests that meet the minimum CMS criteria could contribute substantially to CRC control by adding to the screening population who will not use existing, established methods. However, in order for blood-based tests to be substituted for established CRC screening methods, they would need higher CRC and APL sensitivities that deliver programmatic benefits similar to those of FIT and colonoscopy. Studies of emerging tests are likely to lack the power to distinguish early-stage vs late-stage CRC sensitivity, so we will probably rely on overall CRC sensitivity; fortunately, this is adequate. We propose defining an acceptable false-positive rate in people with a normal colon as an alternative to the CMS “minimum CRC specificity” in order to avoid disincentivizing detection of CRC precursors, the removal of which can lead to CRC prevention. APL detection should be a priority for blood-based screening test developers.

Supplementary Material

Note: To access the supplementary material accompanying this article, visit the online version of Gastroenterology at www.gastrojournal.org, and at http://doi.org/10.1053/j.gastro.2024.03.011.

WHAT YOU NEED TO KNOW.

BACKGROUND AND CONTEXT

Colorectal cancer (CRC) screening is highly effective but underused. Emerging blood-based biomarkers (liquid biopsy) could improve screening participation. The Centers for Medicare & Medicaid Services set minimum blood test thresholds of 74% sensitivity and 90% specificity for CRC.

NEW FINDINGS

A blood test that meets the Centers for Medicare & Medicaid Services’ minimum thresholds could be effective in people who decline stool tests or colonoscopy, but it should not substitute for the established methods, which are more effective and cost-effective. To contemplate substitution, high enough CRC and advanced precancerous lesion sensitivity are needed to match the programmatic benefit of established methods.

LIMITATIONS

Complex participation patterns over time or declining sensitivity by test cycle were not modeled over time, and outreach and navigation costs were not considered.

CLINICAL RESEARCH RELEVANCE

Implementation studies are needed to determine real-world participation rates with stool-based, blood-based, or colonoscopy-based CRC screening in the context of patient and provider choice.

BASIC RESEARCH RELEVANCE

Advanced precancerous lesion detection should be a top priority in the development of new blood-based tests to screen for colorectal neoplasia.

Abbreviations used in this paper:

APL

advanced precancerous lesion

CMS

Centers for Medicare & Medicaid Services

CRC

colorectal cancer

FIT

fecal immunochemical testing

MCED

multi-cancer early detection

MOSAIC

Model of Screening and Surveillance for Colorectal Cancer

mSep

methylated Septin9

MT-sDNA

multi-target stool DNA

QALY

quality-adjusted life-year

SSL

sessile serrated lesion

Data Availability

Data inputs are available to other researchers and the methods are described in detail in the Appendices. The decision analytic model software program is not available for dissemination at this time.