Colorectal Cancer Screening with Blood-Based Biomarkers: Cost-Effectiveness of Methylated Septin 9 DNA versus Current Strategies

Colorectal cancer is the third most common cancer globally and the second most common cancer in industrialized countries (1–3). Screening can decrease colorectal cancer-related mortality through removal of precancerous lesions and detection of early-stage cancers (4–8). The United States Preventive Services Task Force (USPSTF) recommends guaiac or immunochemical fecal occult blood tests (FOBT, FIT), sigmoidoscopy, or colonoscopy for colorectal cancer screening starting at age 50 years (grade A recommendation; ref. 9).

Despite the established benefits of colorectal cancer screening, approximately 40% of eligible Americans do not undergo colorectal cancer screening (10). Moreover, adherence to screening over time is suboptimal (10–19). Barriers to screening at the system, practice, and individual level can substantially reduce the effectiveness of screening programs (20). Several promising blood-based biomarkers of colorectal cancer are under investigation and development (21, 22). An easily conducted blood-based screening test might be accepted by persons not currently participating in screening and may improve adherence over time, and a highly accurate blood test could even replace current tests if it proved more acceptable to most people.

Plasma methylated Septin 9 DNA is a marker of colorectal cancer (23–25). In a recent large prospective study in average-risk persons undergoing screening colonoscopy in the United States and Germany, a 2-well plasma Septin 9 DNA methylation assay showed an overall sensitivity of 48% for colorectal cancer and a specificity of 92%, with sensitivities of 35% for stage I colorectal cancer, 63% for stage II colorectal cancer, 46% for stage III colorectal cancer, and 11.2% for advanced adenoma (26). This sensitivity is superior to that of FOBT, but inferior to that of FIT. It cannot yet be compared with that of other promising blood-based biomarkers (21, 22) because these other markers have not yet been evaluated in large prospective studies. A commercial assay for this biomarker is available in Europe.

We used our published decision analytic model of colorectal cancer screening, which is validated against randomized controlled trials of FOBT and sigmoidoscopy (27, 28), to address key questions pertaining to the potential use of a blood-based biomarker in a screening program, illustrated by methylated Septin 9 DNA, including the appropriate screening interval, comparative effectiveness and cost-effectiveness in relation to established strategies, and the potential impact at the population level accounting for differential patterns of uptake and adherence among screening strategies over time.

Our published decision analytic model, its validation (27, 28) against the Minnesota Colon Cancer Control Study (4, 5), the United Kingdom Flexible Sigmoidoscopy Trial (8), the SCORE Trial (29), and the PLCO Cancer Screening Trial (30), the data sources used, and a model schematic are presented in the Supplementary Appendix. We modeled the contemporary population in the United States at average risk for colorectal cancer, with age-specific all-cause mortality based on U.S. Life Tables from 2003.

PRESEPT, a prospective multicenter study in the United States and Germany, examined the performance for colorectal cancer detection of a polymerase chain reaction-based assay for methylated Septin 9 DNA in plasma of average risk persons undergoing screening colonoscopy (26). Two plasma aliquots per person were initially tested (2-well assay), and a third aliquot of remaining DNA was tested later post hoc to emulate a 3-well assay (26). Based on these results, we modeled both assays, with the 3-well assay having higher sensitivities but slightly lower specificity than the 2-well assay.

Because there is value in illustrating different sets of test performance characteristics for blood-based biomarkers, we compared screening using 2-well or 3-well methylated Septin 9 DNA assays (^mSEPT9-2well and ^mSEPT9-3well) with established strategies recommended by the USPSTF: annual FOBT, annual FIT, sigmoidoscopy every 5 years, and colonoscopy every 10 years (9), and the combinations sigmoidoscopy/FOBT and sigmoidoscopy/FIT. We recognize that the 3-well assay model inputs are based on an emulation and not a prospective evaluation.

Screening was superimposed on the Natural History module. Screening and surveillance were offered to persons at average risk of colorectal cancer from ages 50 to 80 years. If a screening test was positive, then colonoscopy was offered. If colonoscopy was normal after a positive screening test, the screening test was assumed to be a false-positive and screening resumed in 10 years with the primary screening strategy. With colonoscopy, polyps were removed and colorectal cancers were biopsied if detected. In all strategies, surveillance colonoscopy was carried out on average every 5 years after adenoma detection to reflect guidelines that suggest surveillance at 3, 5, or 5 to 10 years depending on adenoma number and size (31), and within 1 year of diagnosis, 3 years, and then every 5 years after colorectal cancer diagnosis (31).

Assumptions about test performance characteristics, uptake, and adherence are crucial in our analyses. Test sensitivities and specificities are presented in the Supplementary Appendix, Supplementary Table S1. In the base case, we modeled perfect uptake and adherence, and we considered multiple levels of imperfect uptake and adherence in sensitivity analyses.

Our model is calibrated to SEER data on stage distribution by localized, regional, and disseminated colorectal cancer. SEER coding guidelines consider stage IIA as localized and IIB as regional disease. We used the reported population-adjusted sensitivities of ^mSEPT9-2well and ^mSEPT9-3well (26) to estimate the assay's sensitivities for localized, regional, and disseminated colorectal cancer.

In the PRESEPT study, only one-time screening was carried out. Before a ^mSEPT9-2well or ^mSEPT9-3well screening program could be compared with alternatives, a testing interval had to be selected. We examined screening with ^mSEPT9-2well and ^mSEPT9-3well at progressively shorter intervals, as we have described previously (32), and selected 2 years as a screening interval consistent with the common willingness to pay threshold of $50,000/life-year gained (see Results).

The derivation of cost inputs in year 2010 dollars is described in the Supplementary Appendix. The cost of ^mSEPT9-2well and ^mSEPT9-3well in the United States is not yet determined. We used $150 in the base case (range $100–200) based on anticipated costs (C. Lofton-Day, personal communication).

The principal model outputs were quality-adjusted life-years (QALY) and costs per/person (33, 34). Future QALYs and costs were discounted by 3% annually (35). Reported health state utilities for colorectal cancer by stage were used to calculate QALYs by applying these for 5 years after colorectal cancer diagnosis. For each strategy, we estimated colorectal cancer cases by stage and deaths in a cohort of 100,000 persons. The effects on colorectal cancer mortality for ^mSEPT9-based screening are unknown. Estimates of effectiveness were produced by the model for these and all strategies based on the assumptions about natural history, uptake, adherence, and test-performance characteristics.

Analyses from the perspective of an insurer such as Medicare were conducted in TreeAge Pro (TreeAge Software, Inc.) and in Excel 2003 (Microsoft Corporation). Incremental cost-effectiveness ratios were calculated (33, 34).

One-way sensitivity analyses were conducted on all model inputs. Threshold analyses were conducted on influential variables. To estimate the uncertainty of our projections, a Monte Carlo simulation with 1,000 trials was carried out. We used beta distributions for probabilities derived from means, SD, and ranges in the literature (36). 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 conducted analyses reflecting persons who take up and adhere with screening and follow-up colonoscopy after abnormal screening (maximum efficacy), as well as scenarios reflecting imperfect uptake, per-cycle adherence, and follow-up (potential effectiveness). We addressed the potential impact of ^mSEPT9 at the population level by contrasting scenarios with increased overall uptake of screening as a consequence of acceptance of ^mSEPT9 by some of those currently unscreened, versus partial substitution by ^mSEPT9 for current strategies (10–19).

The incremental cost/QALY gained with ^mSEPT9-3well increased from $5,900 to $16,400 as the testing interval decreased from every 5 years to every 2 years, and rose sharply to $49,200 for annual testing, assuming a test cost of $150. This accelerating cost/QALY gained as the testing interval moved to yearly was also observed with ^mSEPT9-2well, and at ^mSEPT9 costs of $100 and $200. Annual testing at a test cost of $200 had incremental costs/QALY gained of $64,600 for^mSEPT9-3well and $52,500 for ^mSEPT9-2well. ^mSEPT9-3well and ^mSEPT9-2well with annual testing were less effective and more costly than annual FIT. Thus, a base case testing interval of 2 years was selected for ^mSEPT9-based screening.

Assuming perfect uptake, adherence, and follow-up after abnormal screening, the greatest reductions in colorectal cancer incidence and mortality compared with no screening were observed with sigmoidoscopy/FIT, and the reductions with ^mSEPT9-3well and ^mSEPT9-2well were lower than with the alternatives (Table 1).

All strategies yielded substantial gains in life expectancy, and some but not all were cost-saving (Fig. 1, Table 1). ^mSEPT9-3well and ^mSEPT9-2well gained fewer QALYs than the other strategies, and incurred the highest costs (Fig. 1, Table 1).

Compared with no screening, ^mSEPT9-2well and ^mSEPT9-3well yielded costs/QALY gained of $11,500 and $8,400, respectively (Table 1). In comparison, sigmoidoscopy/FIT, sigmoidoscopy/FOBT, and colonoscopy yielded costs/QALY gained under $3,000, and FOBT, sigmoidoscopy, and FIT were more effective and less costly (i.e., cost-saving) compared with no screening (Table 1).

While all strategies yielded substantial gains in life expectancy compared with no screening, and the differences in effectiveness between strategies were comparatively small, annual FIT was preferred in incremental comparisons between strategies. It was more effective and less costly than ^mSEPT9-2well, ^mSEPT9-3well, FOBT, sigmoidoscopy, and colonoscopy (Table 1).

Comparisons between strategies were affected by key attributes of the screening tests. Assuming a best-case scenario for ^mSEPT9 based on the CIs for reported sensitivities and specificities, ^mSEPT9-2well and ^mSEPT9-3well became more effective than FOBT but not FIT. With these assumptions, ^mSEPT9-3well would have to be carried out yearly at a test cost of less than $10 to approximate but still not match the effectiveness and costs with FIT. At the lower bound of reported sensitivities for FIT, ^mSEPT9-3well with base case assumptions carried out yearly was nearly as effective as FIT, but ^mSEPT9-3well cost would have to be less than $5 in order for costs to be similar with the 2 strategies. Compared with no screening, ^mSEPT9-2well and ^mSEPT9-3well every 2 years were cost-saving at a cost of $50.

The costs of colonoscopy and colorectal cancer care, and the underlying risk of colorectal cancer in the population affected the results for all strategies (Table 2). For instance, if colonoscopy costs increased by 70% to $1,097 and $1,722 with biopsy, to illustrate costs for commercial insurance instead of Medicare, then FIT was no longer cost-saving, but it was still dominant over the ^mSEPT9 strategies. Other variables did not significantly affect the results, ranking of strategies, or conclusions (Table 2).

FIT remained more effective and less costly than ^mSEPT9-2well and ^mSEPT9-3well in all iterations. Cost-effectiveness acceptability curves are shown Fig. 2. Compared with no screening, the strategies' median costs/QALY gained and the 95% CIs were $11,600 ($8,200–$16,300) for ^mSEPT9-2well, and $8,500 ($5,800–$11,600) for ^mSEPT9-3well. By comparison, the values were $600 (cost saving to $2,900) for sigmoidoscopy/FIT; $570 (cost saving to $2,900) for sigmoidoscopy/FOBT; $2,700 ($300–$5,300) for colonoscopy; and cost-saving in all iterations for FIT, FOBT, and sigmoidoscopy.

Differential screening uptake affected the comparisons between strategies. FIT was no longer more effective than ^mSEPT9-2well when the FIT uptake rate fell below 70% relative to that of ^mSEPT9-2well, and it was no longer more effective than ^mSEPT9-3well when the FIT uptake rate fell below 85% relative to that of ^mSEPT9-3well. For FOBT, the analogous thresholds were 85% and 95%, respectively. The thresholds at which ^mSEPT9-3well yielded incremental costs/QALY gained less than $50,000 compared with FIT and FOBT were relative uptake rates of 60% with FIT and 70% with FOBT (Fig. 3).

Differential rates of per-cycle adherence among persons taking up screening also affected the comparisons between strategies. To illustrate, assuming 75% per-cycle adherence with ^mSEPT9-based screening, ^mSEPT9-3well yielded incremental costs/QALY gained of less than $50,000 compared with FIT when the per-cycle adherence with FIT was less than 20% (Fig. 4), and compared with FOBT when the per-cycle adherence rate with FOBT was less than 40%.

Table 2 shows the estimated reductions in colorectal cancer incidence and mortality in a population cohort reflecting current rates of screening uptake and adherence for the currently accepted tests, and colonoscopy follow-up after an abnormal screening test in the United States. If the ^mSEPT9 blood test were then introduced into the population and it induced some of those who are currently not screened to undergo screening, then this yielded incremental reductions in colorectal cancer incidence and mortality at very reasonable costs/QALY gained (Table 3). However, when ^mSEPT9-based screening was substituted for other alternatives among some of those currently screened, effectiveness decreased and costs increased, as illustrated for FIT, FOBT, and colonoscopy (Table 3). For every 10 persons switching from the current strategies (in proportion to current use; Table 3) to ^mSEPT9-3well, more than 6 additional persons would need to switch from no screening to ^mSEPT9-3well to effect a net gain in QALYs, and a total of more than 13 additional persons would need to switch from no screening to ^mSEPT9-3well in order for this gain to incur costs of less than $50,000/QALY gained. This shows how the net impact of an emerging screening test at the population level is likely to depend on the degree to which it induces nonscreened persons to undergo screening, and the degree to which it changes the fractions of the population taking up each specific strategy.

Our health economic evaluation of ^mSEPT9-based colorectal cancer screening compared with established screening strategies offer insights into the potential role of screening with a blood-based biomarker at the population level, and highlights principles applicable to colorectal cancer screening in general. In our analyses, ^mSEPT9-based screening was effective and cost-effective by traditional standards compared with no screening. However, with current test performance characteristics, screening every 2 years with ^mSEPT9 was estimated to be less effective and more costly than existing alternatives. At the most favorable extremes of the reported CIs for ^mSEPT9-3well test performance characteristics (26), which serves to illustrate any emerging blood-based biomarker, ^mSEPT9-3well screening would need to be carried out yearly at a cost less than $10 to be preferred over FIT by traditional standards, assuming comparable levels of uptake and adherence.

The potential role of screening based on ^mSEPT9 or comparable biomarkers depends on the setting in which it is offered, and population rates of screening uptake, adherence per screening cycle, and follow-up colonoscopy after an abnormal screening test. From the point of view of a health care system seeking to choose a single testing strategy, FIT emerged as the preferred alternative assuming perfect uptake, adherence and follow-up. Population studies suggest, however, that uptake, adherence and follow-up rates for fecal-based tests are disappointing (10–19). In organized programs, uptake rates have been up to 60% to 80% (37, 38). It is hypothesized that a blood-based test could lead to improved colorectal cancer screening uptake or adherence rates, but this remains to be proven in clinical studies.

In a health care system that accepts multiple colorectal cancer screening strategies, the potential effect of screening based on ^mSEPT9 or comparable biomarkers depends on the extent to which it increases overall screening uptake as opposed to being substituted for current alternatives, and the relative rates of adherence and follow-up compared with current alternatives. For example, if ^mSEPT9-based screening were taken up only by persons who are currently not screened at all, then net clinical benefits would be realized at acceptable costs. Alternatively, if a test such as ^mSEPT9-3well with current test performance characteristics were simply substituted for current alternatives without improvements in adherence or follow-up, then overall effectiveness would decrease and costs would increase. The real effect of screening with ^mSEPT9 or comparable biomarkers is likely to lie between these extremes.

In most previous modeling studies by us (27, 32) and others (39), colorectal cancer screening strategies have been estimated to be highly cost-effective, but usually not cost-saving. As the cost of colorectal cancer treatment continues to rise (40, 41), certain strategies may actually be cost-saving (42–44). In recent years in the United States, the costs of colorectal cancer care have been increasing (40, 41, 45), whereas Medicare reimbursement rates for screening tests including colonoscopy have been decreasing (46). These opposing trends in colorectal cancer care costs and screening costs are the principal factors behind our and other recent study results that FIT, FOBT, and sigmoidoscopy may be cost-saving in the United States. With this in mind, it is possible that screening with a relatively inexpensive blood-based biomarker could be cost-saving compared with no screening, but its cost-effectiveness against other screening alternatives will depend on the relative test performance characteristics, uptake, adherence, and cost.

Our study has several strengths. Our model is validated against prospective, randomized clinical trials of screening with FOBT (4, 5) and sigmoidoscopy (8, 29, 30), which provides us with some confidence in our model's predictions. The test performance characteristics for ^mSEPT9-based screening were obtained from a prospective study, and the CIs serve to illustrate a range of potential performance by emerging biomarkers. We examined in detail the effects of imperfect screening uptake, adherence, and follow-up, factors that others have also emphasized recently (47). Our results with perfect uptake, adherence and follow-up represent estimated efficacy under ideal circumstances, whereas our results with imperfect uptake, adherence, and follow-up represent estimated effectiveness under current clinical realities.

Our study has some limitations. The degree of independence between individual tests in a given strategy is not known, and our model assumes independence. For instance, although improvements in ^mSEPT9 test-performance with 3 wells compared with 2 wells suggest that additional testing provides incremental yield, the true degree of independence between tests from year to year is not known. However, the effect of assuming independence is likely to be most pronounced for strategies with yearly testing, including FOBT, which was the subject of one of our encouraging validation exercises (Supplementary Appendix). We did not model adenoma multiplicity or location, and the number of people with adenomas and colorectal cancer detected by size and stage were not validated in the same way that colorectal cancer incidence and mortality reduction were validated (Supplementary Appendix). Disutilities associated with screening are not included; considering modest disutilities for 1 to 2 days every 1 to 10 years does not affect the results substantially (results not shown). In the real world, screening uptake, adherence, and follow-up are complex at the population and individual level. Instead of pure subpopulations, there is likely a distribution of screening behaviors that may vary over time. The data to inform a model with such complex attributes are not currently available.

In conclusion, colorectal cancer screening with a blood-based biomarker such ^mSEPT9 is likely to be cost-effective compared with no screening, but with test performance characteristics and costs such as those of the current ^mSEPT9 tests, established alternatives are likely to be preferred under idealized circumstances of uptake and adherence. Screening uptake, adherence, and follow-up are as influential determinants of the real-world effectiveness and cost-effectiveness of colorectal cancer screening strategies as are the performance characteristics and costs of screening tests themselves. Colorectal cancer screening with blood-based biomarkers offers potential benefits over current methods, but to realize their full potential, blood-based biomarkers will need to achieve a level of performance that is acceptable to clinicians and patients relative to current technologies. As blood-based biomarkers become available clinically, the decision over whether to adopt them will require weighing their costs and possibly lower performance but higher adherence rates, against the alternative of investing in efforts to improve uptake of current strategies.

U. Ladabaum has a commercial research grant from Epigenomics. S.D. Ramsey has expert testimony in Epigenomics. No potential conflicts of interest were disclosed by the other authors.

The sponsor had no role in the conduct of this study. No future research funding by the sponsor is pending for any author.

Conception and design: U. Ladabaum, M. Wandell, S.D. Ramsey

Development of methodology: U. Ladabaum, M. Wandell, S.D. Ramsey

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): U. Ladabaum, M. Wandell

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): U. Ladabaum, M. Wandell, S.D. Ramsey

Writing, review, and/or revision of the manuscript: U. Ladabaum, J. Allen, S.D. Ramsey

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): U. Ladabaum

Study supervision: U. Ladabaum, M. Wandell

This work was supported by an unrestricted research grant from Epigenomics (to U. Ladabaum).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.