DRAFT: Not for Citation
We conducted a cost-effectiveness analysis of CT colonography in comparison with the currently recommended CRC screening tests of colonoscopy, flexible sigmoidoscopy, and FOBT (guaiac Hemoccult II and SENSA, and FIT) in response to a request by AHRQ and CMS for a National Coverage Determination. The analysis is based on a cohort of previously unscreened 65-year-old individuals followed over their lifetimes and is conducted from both the CMS payer perspective and a modified societal perspective. We evaluated two recent large-scale CT colonography studies as our base case with referral to optical colonoscopy for a CT colonoscopy-detected lesion of 6 mm or larger diameter and with repeat screening with CT colonography every 5 years. Sensitivity analyses were conducted for referral of individuals with only larger lesions (10 mm or larger) and for longer repeat screening intervals (10 years) as well as for worse case test parameters. Even though the life-years gained by 5-yearly CT colonography with a 6 mm referral for optical colonoscopy were roughly comparable to those from colonoscopy screening every 10 years, the overall costs of both base case CT colonography strategies were higher than all of the other screening strategies considered and were dominated. However if CT colonography reimbursement costs were relatively lower than that of colonoscopy, or CT colonography adherence was differentially higher than for other CRC screening tests, including colonoscopy, then screening with CT colonography would be a cost-effective alternative.
At first it may seem surprising that CT colonography, based on the best evidence available to date, was not cost-effective when compared with the other CRC screening tests since the CT colonography sensitivity for the larger adenomas and CRC is comparable to that of optical colonoscopy and the cost for CT colonography was less that of optical colonoscopy. However, the strategy of CT colonography screening is not a single test but a two-step procedure with those with 6 mm or larger polyps referred to optical colonoscopy. In addition, repeat screening is every 5 years rather than every 10 years as for colonoscopy. Consequently the aim of this analysis was also to explore the conditions under which CT colonography (or for that matter any other new test) could be considered cost-effective compared with the existing screening tests. We therefore conducted threshold analyses to determine what a CT colonography would have to cost in order for one of the CT colonography strategies to lie on the efficient frontier (i.e., be a non-dominated strategy). CT colonography screening could be cost-effective (i.e., be a non-dominated strategy) at a cost of $108 to $205 per scan depending on the simulation model used and the test characteristics of CT colonography. If the cost per test were $179 to $237, CTC would provide additional years of life at the same cost per year as colonoscopy (with CMS reimbursement of approximately $500 for colonoscopy without polypectomy and $650 for colonoscopy with polypectomy).
We conducted sensitivity analyses to address the question of whether with increased adherence CT colonography would be on the efficient frontier. For this analysis we assumed that adherence was 50% for the currently-recommend tests and that there was increased adherence with the CT colonography test strategies among unscreened individuals. If screening adherence were higher with CT colonography compared with other screening tests, CT colonography screening could be included among the efficient strategies at the base-case cost estimate of $488
We assumed that all in the cohort of 65-year-old individuals were previously unscreened. In reality, many subjects entering the Medicare program will have had CRC screening before age 65. Of those with prior screening, only those without adenomas detected are still eligible for average-risk screening. Adenoma patients should undergo more frequent surveillance with colonoscopy (Winawer 2006) than those with no neoplasia. This means that on average the eligible population for average-risk screening entering Medicare will be at lower risk than an unscreened population. Accordingly we may have overestimated the life-years gained from screening. However, this holds for all tests and strategies and is therefore not expected to significantly influence our results, because the relative performance of one test over the other remains the same. We assessed the potential effect of the assumption of an unscreened 65-year-old population by determining threshold costs for CTC screening when screening a 50-year-old cohort from age 50 onwards; the results did not change substantially.
As reported in the DNA stool test report to CMS, (Zauber 2007) an important finding from our analysis is that the currently recommended CRC screening tests provide good value for the resources spent. Hemoccult II, the test proven in randomized controlled trials to reduce CRC mortality by 15-33%, with a $4.54 CMS reimbursement, is cost-saving relative to no screening. Other FOBTs as well as flexible sigmoidoscopy and colonoscopy provided additional life-years gained over Hemoccult II, often with reasonable costs. Our favorable cost-effectiveness result for the CRC screening strategies is likely due to the increasing costs of CRC-related care and the costs of the screening tests not increasing at the same rate or even lower than previously reported. In this analysis all the costs come from the same source: Medicare reimbursement. The costs for treating CRC stage III and IV and incurable CRC have been increasing since the introduction of newer therapies. The reason that the SimCRC and CRC-SPIN models found more cost-saving strategies than the MISCAN model is likely due to the fact that they find a great reduction in cancer incidence with CRC screening because of their longer dwell times.
CRC screening guidelines from the Multi-Society Task Force were published in 1997 for currently available tests but the authors also considered how to evaluate new screening tests as well. The guidelines state that a newer test could be substituted for a currently recommended test (or added to the recommendations) if evidence were available to demonstrate that the new test had: (1) a comparable performance for sensitivity and specificity in detecting cancer or adenomatous polyps at comparable stages, (2) was equally acceptable to patients, and (3) had comparable or lower complication rates and costs (Winawer 1997). We address each of these issues below.
The two well-designed studies used as our base cases demonstrate that CT colonography has comparable sensitivity to detect adenomas 10 mm or larger and CRCs as optical colonoscopy but slightly lower sensitivity to detect adenomas of size 6-9 mm. Furthermore adenomas of size <6 mm are not reported at all for CT colonography (Zalis 2005). The natural history of adenomas <6 mm is not well known (2008a, Butterly 2006, O'Brien 1990). The risk of high-grade dysplasia or invasive CRC is lower in these smaller adenomas than those ≥6 mm but the smaller lesions are also the most common. Repeat CT colonography screening at 5-year intervals with referral to optical colonoscopy for those lesions of larger size is one way to offset the optical colonoscopy screening strategy of removing all polyps.
The specificity of CT colonography varied for the two base cases, with the DoD study having higher sensitivity but lower specificity than the NCTC. Lack of specificity is also a factor in optical colonoscopy which detects and removes hyperplastic and other polyps as well as the adenomas less than 6 mm in size. In the analyses we assumed 90% specificity for optical colonoscopy to take into account the detection and removal of non-adenomas in optical colonoscopy screening.
The evidence to date has primarily been for a one-point-in-time assessment of CT colonography. Information on programmatic use of CT colonography (i.e., repeated screening) is not yet available. Future studies are needed to assess repeat screenings and the impact of a programmatic utilization of CT colonography.
The evidence shows that there is a strong learning curve for CT colonography and that readers must have standardized rigorous training and proper technique to obtain the good test parameters observed in the well-designed trials. Quality measures for CT colonography are in development (McFarland 2008). New techniques or modifications of older techniques must be evaluated as to their test performance characteristics.
Additional techniques are demonstrated for optical colonoscopy to detect flat adenomas (Soetikno 2008) and the clinical importance of flat adenomas has been discussed (Lieberman 2008b). The CT colonography literature has also discussed detection of flat lesions (Fidler 2002, Park 2007). Additional techniques to detect flat adenomas have not been included in the modeling for this report.
The currently-recommended CRC screening tests all require considerably more patient involvement than screening tests for other diseases. The individual undergoing screening must complete a cleansing bowel prep for colonoscopy, flexible sigmoidoscopy as well as for CT colonography, restrict their diet for Hemoccult II, colonoscopy, and CT colonography; and restrict NSAID use with Hemoccult II; have contact with the stool for any of the FOBTs; and go to a medical setting for colonoscopy, flexible sigmoidoscopy, or CT colonography. Colonoscopy procedures have a small but real risk of perforations and due to sedation, require an escort to and from the procedure. Although CT colonography is non-invasive it does require a cathartic bowel preparation just as for optical colonoscopy, as well as stool tagging. In addition, a positive CT colonography requires referral for optical colonoscopy as is the case for other two-step procedures. Whether same-day CT colonography and optical colonoscopy for those with a positive CT colonography is possible in the general medical practice is not yet known although there is discussion of this as a practice model (Pickhardt 2006b). If not, then the referred patient must undergo two cathartic preparations. The patient impression is often that CT colonography is 'virtual' and non-invasive. It is not known whether the adherence to optical colonoscopy referral for those with positive CT colonography will be as high or higher as those with positive findings on other CRC screening tests. Although non-cathartic preparations have been developed for CT colonography (Callstrom 2001, Iannaccone 2004) they involve both dietary restriction over a number of days and ingestion of various oral contrast agent (Pickhardt 2007b). Consequently, the non-cathartic preparations are not 'prepless'. Also same-day optical colonoscopy cannot be performed in those with non-cathartic preparations if the CT colonography is positive for lesions of size 6 mm or larger.
There is a low level of radiation exposure with CT colonography. The long-terms effects of cumulative exposure to radiation that would be associated with interval screening with CT colonography are unknown. In addition, concern for radiation risk on part of patient or physician could affect willingness to adhere to CTC screening.
In addition to findings within the colorectal tract, CT colonography may identify extracolonic findings (Hara 2000, Pickhardt 2008a). The extent to which these finding may lead to early diagnosis of a potentially lethal disease, or just a false-positive finding resulting in extra work-up and additional exposure to radiation is also not well established (USPSTF 2008; Whitlock 2008).
Patient-stated preference for CT colonography relative to other CRC screening tests has been investigated in those who have had CT colonography. Pickhardt conducted a survey of patient preferences for repeat CT colonography versus repeat optical colonoscopy in his DoD study (2003) and demonstrated a slight preference for CT colonography. Gleucker (2003) addressed patient preferences for those having CT colonography and colonoscopy versus those with CT colonography and double contrast barium enema; CT colonography was preferred. Further studies of patient preference for CT colonography versus optical colonoscopy for the initial screen and of the willingness to have optical colonoscopy if CT colonography is positive are needed, especially among subjects who have been unwilling to perform any of the current CRC screening tests (Levin 2008).
Although there are these potential problems in obtaining high adherence for CT colonography, if adherence for CT colonography could be achieved at only slightly higher levels (10% to25% over current CRC screening levels of 50%) our sensitivity analysis on adherence suggests that CT colonography would become cost-effective.
There are perforation complications associated with CT colonography but at a lower rate and with less substantial level of complications as colonoscopic complications (Whitlock 2008). There is radiation exposure with CT colonography but at a low level. The harm of low-level radiation has been difficult to assess. Furthermore followup of extracolonic findings detected on CT colonography does contribute to a higher cumulative dose of radiation exposure that should be taken into account (Brenner 2007, Levin 2008). Risk may be small, but certainly not negligible.
CT colonography is associated with exposure to radiation, which we did not consider in the current analysis. Brenner (2007) estimated that the excess cancer risk from a pair of CT colonography scans using typical current scanner techniques is about 0.14% for a 50-year old and half that for a 70-year old. This estimate is controversial, because it was based on simulation calibrated to atomic bomb survivors. Multiple CT colonography screens will increase the radiation dose proportionally and most likely also the radiation risks. We found that CT colonography is only compatible to colonoscopy screening if offered seven times (every 5 years between ages 50 and 80), potentially leading to an excess cancer risk of approximately 0.47%. This will lead to life-years lost due to CT colonography which are not negligible compared to the life-years gained. We did not take these excess cancer cases into account, because there is good evidence that radiation dose with CT colonography can be reduced by at least a factor of 5 (and perhaps as much as 10), while still maintaining sensitivity and specificity for polyps larger than approximately 5 mm (Brenner 2005). With these dose reductions, excess risk of cancer from CTC becomes negligible.
CT colonography generally costs less than optical colonoscopy on a per scan basis but the overall screening strategy for CT colonography screening is more expensive than other screening strategies in general as demonstrated here given comparable adherence.
All analyses were conducted by three separate microsimulation modeling groups of the NCI-sponsored modeling consortium, CISNET, using independently developed models but with common inputs. The comparability of the findings of the three modeling groups strengthens the credibility of our results and can be viewed as a sensitivity analysis on the underlying natural history assumptions. All three models have been calibrated to CRC incidence rates from a pre-screening era. All the models have been extensively validated against clinical trial data on Hemoccult II screening. The models do differ in the dwell time from adenoma to clinically detectable CRC. The MISCAN model assumes a shorter dwell time compared with the SimCRC and CRC-SPIN models. Based on this difference in dwell time, the MISCAN model estimates fewer life-years saved from removing adenomas as a result of screening than the SimCRC and CRC-SPIN models, and estimates a greater benefit for shorter rescreening intervals for adenoma-sensitive tests than does the other two models. The fact that all three models come to similar conclusions with respect to cost-effectiveness and threshold costs of CT colonography screening shows the robustness of the results for uncertainties in the duration of the adenoma-carcinoma sequence.
The distribution of dwell time from adenoma to carcinoma is not known with certainty. The uncertainty on dwell time affects the assessment of all the screening tests, including CT colonography. In particular it affects the tests with respect to detection of adenomas.
This report is the first cost-effectiveness analysis using the new estimates of test performance from the DoD and NCTC trials in the 65-year-old-age group. Other cost-effectiveness analyses based on test performance of earlier CT colonography technology or in a 50-year-old cohort include Sonnenberg (1999), Ladabaum (2004), Vijan (2007), Pickhardt (2007c, 2008b) and Scherer (2008).
The models simulate the progression from adenoma to CRC by increasing the size of the adenomas over time. Because adenoma size, villous component, and high-grade dysplasia are highly correlated (O'Brien 1990), the size representation indirectly represents histology and high grade. However, the models do not separately simulate the step from adenoma with low-grade dysplasia to an adenoma with high-grade dysplasia. We also did not allow for de novo cancers (cancers that arise without a prior adenoma state). Lastly, we assumed that SEER incidence data prior to the time of active CRC screening in the US is a good representation of the cancer incidence expected today in an unscreened population. However, because there has been a small net improvement in CRC lifestyle risk factors for CRC over time (Knudsen 2004, 2005), estimates of CRC incidence may be overestimated. The impact of overestimating CRC incidence is that all CRC screening benefits are also overestimated, though we would not expect significant differences in the relative benefit across strategies.
In the current analysis, we assumed conditional independence of repeat screenings. Consequently we assumed that there were no systematic false-negative results for adenomas and cancers. This is likely a reasonable assumption for FOBT and FIT testing because bleeding of a lesion is assumed to be a random event, so that if a test misses a lesion the first time, then it has approximately the same probability of catching a bleed on the next screen. This assumption may be less reasonable for optical endoscopy, as certain lesions may be more difficult to find (e.g., in a fold) but is a reasonable assumption for CT colonography which can detect lesions on folds (Pickhardt 2004).
In this analysis, we included the current recommendations for average-risk CRC screening as the comparator strategies. We did not consider alternative screening intervals for the currently recommended screening tests. We also made the assumptions that screening would stop at age 80 and that individuals would remain on a surveillance schedule for their lifetime, which may not be realistic assumptions for what occurs in practice.
In our sensitivity analysis of screening adherence we assumed that individuals would be either fully adherent with a screening strategy or never screened. This is an oversimplification of what occurs in practice, but is closer to reality than an assumption that individuals show up randomly to their scheduled screens. A recent study by Coups et al. (2007) of data from the 2000 National Health Interview Survey found that almost 40% of the US population aged 50 and older were adherent with CRC screening guidelines and only 13% were screened but not according to guidelines (the remaining group was never screened).
The costs of the screening tests, as well as the costs of complications associated with screening (primarily colonoscopy), were based on 2007 Medicare reimbursement rates. To the extent that these rates change differentially in the future (e.g., a decrease in the reimbursement rate for colonoscopy) our results will change.
Costs for CRC treatment were for the period 1998 to 2003. In this period use of the expensive biological therapies cetuximab and bevacizumab was limited (Schrag 2004). We would expect that inclusion of these costs as later data become available would make the cost-effectiveness more favorable overall. CRC screening can have two potentially beneficial effects: 1) primary prevention of CRC through detection and removal of adenomas that might have eventually become cancer, and 2) early detection of CRC, when it is in an earlier stage that is more amenable to treatment. In general, those strategies that are associated with a higher reduction in cancer incidence (i.e., act largely through primary prevention rather than early detection,) will have a greater net savings.
With the exception of the Warren, Klabunde, and Brown upcoming manuscript (Klabunde 2007), there are few data specifically on colonoscopy complications in the Medicare population. For example, the Warren analysis reports hospitalization for dehydration following colonoscopy. This complication was not cited in the general population studies across ages. Complications rates are generally lower in organized screening programs, which often focus on the age group of 50 to 65 for CRC screening. Consequently a program to track complications in Medicare beneficiaries who receive CRC screening would be of value to assess the magnitude of risk for this age group.