Frazier AL, Colditz GA, Fuchs CS, Kuntz KM. Cost-effectiveness of Screening for Colorectal Cancer in the General Population. JAMA. 2000;284(15):1954-1961. doi:10.1001/jama.284.15.1954
Author Affiliations: Channing Laboratory, Department of Medicine, Brigham and Women's Hospital (Drs Frazier, Colditz, and Fuchs) and the Dana-Farber Cancer Institute (Drs Frazier and Fuchs), Harvard Medical School; and the Departments of Epidemiology (Dr Colditz) and Health Policy and Management (Dr Kuntz), Harvard School of Public Health, Boston, Mass.
Context A recent expert panel recommended that persons at average risk of colorectal
cancer (CRC) begin screening for CRC at age 50 years using 1 of several strategies.
However, many aspects of different CRC screening strategies remain uncertain.
Objective To assess the consequences, costs, and cost-effectiveness of CRC screening
in average-risk individuals.
Design Cost-effectiveness analysis from a societal perspective using a Markov
Subjects Hypothetical subjects representative of the 50-year-old US population
at average risk for CRC.
Setting Simulated clinical practice in the United States.
Main Outcome Measures Discounted lifetime costs, life expectancy, and incremental cost-effectiveness
(CE) ratio, compared used 22 different CRC screening strategies, including
those recommended by the expert panel.
Results In 1 base-case analysis, compliance was assumed to be 60% with the initial
screen and 80% with follow-up or surveillance colonoscopy. The most effective
strategy for white men was annual rehydrated fecal occult blood testing (FOBT)
plus sigmoidoscopy (followed by colonoscopy if either a low- or high-risk
polyp was found) every 5 years from age 50 to 85 years, which resulted in
a 60% reduction in cancer incidence and an 80% reduction in CRC mortality
compared with no screening, and an incremental CE ratio of $92,900 per year
of life gained compared with annual unrehydrated FOBT plus sigmoidoscopy every
5 years. In a base-case analysis in which compliance with screening and follow-up
is assumed to be 100%, screening more often than every 10 years was prohibitively
expensive; annual rehydrated FOBT plus sigmoidoscopy every 5 years had an
incremental CE ratio of $489,900 per life-year gained compared with the same
strategy every 10 years. Other strategies recommended by the expert panel
were either less effective or cost more per year of life gained than the alternatives.
Colonoscopy every 10 years was less effective than the combination of annual
FOBT plus sigmoidoscopy every 5 years. However, a single colonoscopy at age
55 years achieves nearly half of the reduction in CRC mortality obtainable
with colonoscopy every 10 years. Because of increased life expectancy among
white women and increased cancer mortality among blacks, CRC screening was
even more cost-effective in these groups than in white men.
Conclusions Screening for CRC, even in the setting of imperfect compliance, significantly
reduces CRC mortality at costs comparable to other cancer screening procedures.
However, compliance rates significantly affect the incremental CE ratios.
In this model of CRC, 60% compliance with an every 5-year schedule of screening
was roughly equivalent to 100% compliance with an every 10-year schedule.
Mathematical modeling used to inform clinical guidelines needs to take into
account expected compliance rates.
Colorectal cancer (CRC) is the second leading cause of cancer-related
mortality in the United States, resulting in approximately 56,600 deaths in
1999.1 Screening for CRC reduces mortality
through detection of malignancy at an earlier, more treatable stage as well
as by identification and removal of the precursor lesion, the adenomatous
polyp. A recent panel recommended that average-risk individuals begin screening
at the age of 50 years with one of the following strategies: annual fecal
occult blood testing (FOBT), flexible sigmoidoscopy (SIG) every 5 years, annual
FOBT plus SIG every 5 years, double-contrast barium enema (DCBE) every 5 to
10 years, or colonoscopy (COL) every 10 years.2
Colorectal cancer screening tests vary considerably in terms of their
performance characteristics, complication rates, acceptability, and cost.
Moreover, only the effectiveness of FOBT has been established by randomized
clinical trial.3 To inform the debate on the
health and economic impact of the various CRC screening strategies, we constructed
a decision-analytic model to evaluate the cost-effectiveness of CRC screening
in average-risk individuals. While 3 prior models have shown CRC screening
to be economically attractive,4- 6
in our analysis we specifically evaluate those strategies currently condoned
by expert panels,2,7- 10
address the ongoing debate about the use of rehydration for FOBT, evaluate
the impact of performing a follow-up COL after the detection of a small tubular
adenoma by SIG, and consider the impact of imperfect compliance with screening.
We developed a state-transition Markov model to simulate the evolution
from normal colonic epithelium to adenomatous polyp to malignancy. Superimposed
on the natural history of colorectal disease was a screening mechanism that
allowed for the detection and removal of polyps and the detection and treatment
of cancer. Persons representative of the 50-year-old US population, based
on sex and race, were placed into health states defined by the presence or
absence of a polyp (low risk or high risk, defined as either ≥1 cm or containing
villous histology) or cancer (localized, regional, or distant) in either the
distal or proximal colon. The 2 sides of the colon were modeled separately
because SIG can only visualize the distal colon. The population transitioned
through the different health states on an annual basis (ie, cycle length)
based on transition probabilities estimated from the literature.
The strategies evaluated include the 6 strategies endorsed by the expert
panel,2 1-time screens at 55 years of age,
use of rehydrated (RFOBT) vs unrehydrated FOBT (UFOBT), and different follow-up
regimens for SIG. Persons with a positive FOBT or DCBE test result were referred
for follow-up COL. For polyps detected during SIG or COL, a polypectomy was
performed during the procedure. The 2 different follow-up procedures after
a positive SIG were as follows: (1) only persons found to have a high-risk
polyp were referred for follow-up COL (SIG1) vs (2) persons found
to have any adenomatous polyp, regardless of size or histology, were referred
for follow-up COL (SIG2). Individuals diagnosed with a high-risk
polyp underwent surveillance COL every 3 years. Screening and/or surveillance
continued until 85 years of age.
Incremental analyses were performed by rank ordering the strategies
by increasing effectiveness after eliminating those that were more costly
and less effective than an alternative (ie, ruled out by simple dominance).
We then calculated the incremental cost-effectiveness (CE) ratio for each
strategy (additional cost divided by additional benefit) compared with the
next least expensive strategy. Strategies with a lower effectiveness and higher
CE ratio than another strategy were ruled out by weak dominance, eliminated
from the rank-ordered list, and the incremental CE ratios were then recalculated.11 Sensitivity analyses were performed to assess the
stability of the results to plausible ranges of uncertain parameters. Future
costs and life-years were discounted at an annual rate of 3%.11
The model was programmed in SMLTREE software (James Hollenberg, MD, Roslyn,
Natural History of Colorectal Polyps and Cancer.Table 1 shows selected parameter
We estimated the age- and sex-specific prevalence of adenomatous polyps using
a weighted logistic regression analysis of results from 6 autopsy studies.12- 17
For example, the prevalence of polyps for men aged 50 years was estimated
at 21%, of which 98% were low risk and 61% were in the distal portion of the
colon.12- 17,31- 33
The subsequent incidence of polyps among those polyp free at 50 years of age
was estimated to match the autopsy results for older age groups. We estimated
that 28% of persons with proximal CRC would have a "sentinel" polyp in the
distal portion of the colon that could trigger further workup after SIG.34
The probability of transformation from low-risk to high-risk polyp was
estimated from studies of small polyps left in situ and reexamined annually.18- 20 The probability that
a high-risk polyp would develop into localized cancer was derived from a study
of patients who refused resection of a high-risk polyp.21
There are no empirical data with which to estimate the rates of progression
through cancer stages or to determine the likelihood of presentation with
symptoms. Thus, we varied the estimates of cancer progression and symptom
detection across clinically plausible ranges so that the stage distribution
and CRC incidence predicted by our model was similar to those reported by
the Surveillance Epidemiology and End Results (SEER) program.22
Prevalence of CRC and stage distribution at 50 years of age was obtained
from SEER data.22 Distal vs proximal site of
the cancer was determined as a function of age using published Medicare data.35 SEER estimates of sex-, race-, and stage-specific
CRC mortality (L. Ries, MS, written communication, May 17, 1995) were applied
uniformly to all malignancies regardless of means of detection (by symptoms
or screen) or state of detection (diagnosed vs undiagnosed cancer).
The Effects of Screening. Where data were available, we estimated test sensitivity separately
for the detection of low-risk polyps, high-risk polyps, and cancer.2,23- 29
Mortality caused by the risk of perforation was assigned to the endoscopic
procedures (1.4 × 10−6 for SIG and 5.0 × 10−5 for COL).36,37
Recurrence rates after polypectomy were higher for individuals with
a history of a high-risk polyp diagnosis (25% in the first year and 7.5% a
year thereafter) compared with a prior diagnosis of low-risk polyp (18% in
the first year and 6% per year thereafter).38- 44
Only the transition from normal epithelium to low-risk polyp was increased
among those with a history of polyp.
The costs of CRC treatment by stage and time period (initial, continuing,
and terminal care) were obtained from a cost study from a large health maintenance
organization.30 These costs include the actual
costs of medical personnel and supplies to provide the service as well as
overhead costs, such as administration, charting, and automated information
systems. Test costs were obtained from the same health maintenance organization
(S. H. Taplin, Group Health Cooperative, Seattle, Wash, written communication,
September 30, 1996). All costs were updated to 1998 dollars using the medical
care component of the Consumer Price Index.45
Compliance rates of 50% to 70% have been obtained in the optimized setting
of clinical trials of CRC screening.25,46,47
Therefore, we estimated a realistic goal for compliance with CRC screening
to be 60% for initial tests and 80% for follow-up or surveillance COL. At
each particular screening event, we assumed that a random 60% of the population
underwent the initial screening test, independent of whether they were compliant
with past tests. Among persons referred for follow-up COL, we assumed that
a random 20% would not undergo this diagnostic test. We also show the results
for an initial compliance rate of 100%, which models results of screening
based on actual tests received by patients.
The health and economic outcomes of 22 CRC screening strategies for
white men at average risk are shown in Table 2 and Figure 1.
All screening strategies resulted in reductions in CRC incidence and mortality.
The least intensive strategy, screening once at 55 years of age with SIG,
reduced CRC incidence by 14% and mortality by 16%, whereas the most intensive
strategy, repeated screening from ages 50 to 85 years with RFOBT plus SIG
every 5 years, reduced CRC incidence by 60% and mortality by 80%.
Although every CRC screening strategy extended life expectancy, 7 strategies
were more effective for a lower cost per life-year saved than the alternatives.
Screening once at 55 years of age with SIG or with SIG every 10 years (regardless
of follow-up) resulted in incremental CE ratios of less than $17,000 per life-year
saved. Unrehydrated fecal occult blood testing plus SIG2 every
10 years had a CE ratio of $21,200 per life-year saved compared with SIG2 every 10 years, whereas UFOBT plus SIG2 every 5 years had
a CE ratio of $51,200 per life-year saved compared with UFOBT plus SIG2 every 10 years. Rehydrated fecal occult blood testing plus SIG2 every 5 years was the most effective CRC screening strategy, resulting
in an incremental CE ratio of $92,900 per life-year saved compared with UFOBT
plus SIG2 every 5 years. Other strategies recommended by the expert
panel were eliminated by either simple or extended dominance. In Figure 1, dominated strategies fall below
the lines connecting the nondominated alternatives.
We directly compared UFOBT and RFOBT and found an incremental CE ratio
of RFOBT compared with UFOBT of $45,000 per life-year saved. Rehydrated fecal
occult blood testing resulted in a 65% reduction in cancer mortality, whereas
annual UFOBT resulted in only a 55% reduction. However, RFOBT was dominated
as a stand-alone test because the combination of UFOBT plus SIG2
every 10 years was less expensive but more effective. The more aggressive
follow-up after a positive sigmoidoscopy (SIG2) was only modestly
more costly than the less aggressive strategy (SIG1) with a substantial
increase in effectiveness (direct comparisons of SIG2 vs SIG1 strategies yielded incremental CE ratios less than $12,000 per life-year
The estimated health and economic outcomes of CRC screening for white
women, black men, and black women are not shown due to constraints of space.
However, the relative ordering of the tests in terms of cost and effectiveness
did not change. Because of increased life expectancy (white women) or increased
cancer mortality (blacks), CRC screening was even more cost-effective among
these groups than for white men.
The relative ordering of the strategies did not substantially change
under the assumption of 100% compliance (Figure 1). However, COL once at 55 years of age replaced SIG every
10 years on the cost-effective frontier (illustrated by the lines in Figure 1). Although the DCBE strategies remained
dominated, they moved closer to the cost-effective frontier.
At 100% compliance, screening more frequently than every 10 years became
prohibitively expensive (RFOBT plus SIG2 every 5 years cost an
additional $489,900 per life-year saved compared with RFOBT plus SIG2 every 10 years.) However, the results of the model at 60% compliance
vs 100% compliance are less disparate than appears at first glance. Sixty
percent compliance with an every 5-year strategy translates into an actual
interval between screens of almost 10 years. Interestingly, therefore, 60%
compliance with an every 5-year strategy is equivalent to 100% compliance
with an every 10-year strategy.
Table 3 shows the variables
that caused the incremental CE ratio of RFOBT plus SIG2 every 5
years to shift by more than 10%. The cost-effectiveness of RFOBT plus SIG2 every 5 years was most influenced by compliance with follow-up COL,
test performance of UFOBT, mortality and cost associated with COL, prevalence
and recurrence rate of polyps, and progression from polyp to cancer.
Slight changes in the assumptions about UFOBT (ie, increase in specificity
of UFOBT from 97% to 99%; increase in sensitivity of UFOBT for cancer from
33% to 39%; decrease in cost of UFOBT from $38 to $32) caused it to become
an undominated strategy with a CE ratio less than $20,000 per life-year saved.
Rehydrated fecal occult blood testing strategies dominated all UFOBT strategies
when the specificity of RFOBT was greater than 93% (base-case estimate was
90%). Double-contrast barium enema remained a dominated strategy over a wide
range of values for both sensitivity and specificity. Even if the cost of
DCBE was less than $100 (base-case cost was $296), DCBE every 5 years remains
a dominated strategy. If the cost of COL was lowered by 18%, then screening
once at 55 years of age with COL was no longer dominated.
Colonoscopy is the most sensitive test, but also the most expensive.
If compliance with COL every 10 years was substantially greater than with
the other tests (70% vs 40%), then COL every 10 years was the most effective
strategy with an incremental CE ratio of $92,000 per life-year saved compared
with RFOBT plus SIG2 every 5 years. Colonoscopy every 5 years is
more effective than RFOBT plus SIG2 every 5 years, but the costs
are prohibitive. If the cost of COL was reduced by 23%, then COL every 10
years became a viable strategy as well.
Under certain scenarios, it was less costly to screen than it was not
to screen (ie, the costs of screening were less than the costs of averted
cancer treatment). If the cost of SIG was reduced by 15% ($240) or the cost
of cancer treatment was increased by 20%, then screening was cost saving.
If the incidence of polyps at 50 years of age was 20% higher than estimated
at baseline or the rate of progression from high-risk polyp to cancer was
20% faster than at baseline, then screening was cost saving.
We compared 22 strategies for CRC screening persons at average risk,
including all strategies currently recommended by expert panels.2,7- 10
Seven strategies were not ruled out by either simple or extended dominance.
The incremental CE ratios for these strategies ranged from $1200 per life-year
saved (SIG once at 55 years of age) to $92,900 per life-year saved (RFOBT
plus SIG every 5 years). However, factors such as local expertise, availability
of providers, and patient preferences should also be incorporated into the
choice of screening strategy in addition to estimates of cost-effectiveness.
The cost-effectiveness of CRC screening compares favorably with the
cost-effectiveness of other cancer screening strategies such as annual Papanicolaou
testing beginning at 20 years of age ($99,000 per life-year saved, updated
to 1998 dollars) and annual mammography for women ages 55 to 64 years ($132,000
per life-year saved, updated to 1998 dollars).48
Rarely did the cost of CRC screening exceed what we, as a society, are willing
to pay for other cancer screening tests, even under the most extreme assumptions
in the sensitivity analyses. Conversely, relatively minor changes in baseline
assumptions actually made certain less-intensive forms of screening cost saving.
Compliance with screening for CRC is currently quite low in the United
States. According to the 1997 Behavioral Risk Factor Surveillance System,
only 20% of respondents reported having had FOBT during the preceding year,
and only 30% reported having had a proctoscopy or SIG in the preceding 5 years.49 Given the low proportion of Americans who currently
comply with the recommended screening schedule, advising all Americans to
be screened at least once may be a reasonable starting point for national
policy. Among the 1-time screening alternatives, COL was the most effective
option with a lifetime reduction in CRC mortality of 31% and an incremental
CE ratio of $22,400 per life-year saved compared with 1-time SIG, assuming
60% compliance. A strategy of once-only SIG or COL at 60 years of age has
been evaluated for the Netherlands.50 Those
authors concluded that 40% to 70% of CRC could be prevented at costs comparable
to the existent breast cancer screening program.
At 60% compliance, the CE ratio for RFOBT plus SIG every 5 years falls
within the parameters of what is paid for other preventive services, but at
100% compliance, the cost is prohibitive. The reason for this difference is
that at only 60% compliance, total screening costs for the strategy are markedly
reduced. For instance, over a 10-year period, only 21% of the population will
actually have completed screening at baseline, 5 years, and 10 years, and
6% of the population will not have complied with any screening test at all.
Should guidelines be based on the assumption of perfect or imperfect compliance?
The gold standard for the establishment of clinical guidelines is the results
of a clinical trial. However, in a clinical trial, compliance is never 100%.
Therefore, we advocate that when using modeling to replicate a clinical situation,
compliance rates should be used that mirror the clinical experience.
Colonoscopic follow-up for everyone found to have an adenomatous polyp
at SIG, regardless of size or histology, typically dominated the more restricted
follow-up strategy. Two retrospective analyses have suggested that subsequent
risk of malignancy in patients with adenomas less than 1 cm is minimal.21,51 However, 2 recent reports52,53 have shown an elevated risk of advanced
proximal neoplasia among patients with distal polyps that are either small
(<10 cm) or of tubular histology. Furthermore, those 2 studies reported
that approximately 50% of patients found to have advanced proximal neoplasia
had no distal pathology. The accompanying editorial advocated the use of COL
rather than SIG as a screening test for CRC.54
In our model, COL every 10 years was more effective than SIG every 5 years;
however, it was slightly less effective than the combination of annual FOBT
plus SIG every 5 years. The addition of annual FOBT allows for the detection
of polyps that would develop in the 10-year interval between COL screens.
The mortality reduction of annual FOBT estimated by our model exceeds
the mortality reduction reported in the Minnesota Colon Cancer Control Study.25 The difference is due to the fact that in the clinical
trial, population ages ranged from 50 to 80 years and follow-up was for 10
to 15 years, whereas in the model, screening began for everyone at 50 years
of age, continued for 35 years, and persons were followed up for their lifetime.
When we configured our model to simulate the parameters of the clinical trial,
we obtained a 31% reduction in CRC mortality and a 7% reduction in cancer
incidence, similar to the results of the trial.25
Our analysis has several limitations, principally reflecting the uncertainty
about the natural history of colorectal disease. We assumed that all cancers
arose from polyps; we did not allow polyp progression to depend on time; and
we modeled the progression of polyps in the distal vs proximal colon as independent
events, although they are likely correlated. However, our model was calibrated
with the incidence and stage distribution of CRC based on SEER data.22
Another limitation of our model is that we assumed that the sensitivity
of FOBT was the same for initial and repeated tests. Although we were able
to simulate the results of the Minnesota trial, the longer-term effects of
this assumption could bias our results toward strategies with FOBT. We also
assumed that polypectomy would be performed at the time of an initial SIG
if a polyp were found, which is not universal practice in the United States.
Our model suggests, as do the results of 2 clinical trials,52,53
that the finding of a distal polyp warrants a follow-up COL, regardless of
the size or histology of the polyp. The advantage of removing the polyp at
the time of SIG is that the effectiveness of the SIG is not dependent on compliance
with the follow-up test. However, immediate polypectomy is not always possible
because of either the training of the practitioner or the preparation of the
patient. Under the assumption that all polypectomies are performed at COL,
we found that SIG every 5 to 10 years was dominated and COL once at 55 years
of age and annual UFOBT became reasonable strategies. However, the incremental
CE ratios of the strategies with FOBT plus SIG every 5 years did not change
We compared our results with those from 2 other mathematical models
designed to evaluate the cost-effectiveness of CRC screening in average-risk
we calculated incremental lifetime costs, life expectancy, and CE ratios for
UFOBT and UFOBT plus SIG every 5 years (Table 4). To make the results more comparable, we assumed 100% compliance
and a 5% annual discount rate in our model and updated the costs of the previously
published models to 1998 dollars. The life expectancy gains predicted by our
model were intermediate to those published previously, and our costs were
consistently higher. In a recent report, the MISCAN-COLON model predicted
that screening every 5 years with SIG compared with no screening would be
cost saving.5 With minor changes in our assumptions,
the results of our model concur that CRC screening could be cost saving.
In summary, screening for CRC is as cost-effective as other forms of
cancer screening. Among the screening strategies that we considered, RFOBT
plus SIG every 5 years was the most effective strategy, with an 80% reduction
in CRC mortality and an incremental CE ratio of $92,900 per life-year saved
compared with UFOBT plus SIG every 5 years. Although all other strategies
recommended by the expert panel (annual FOBT, SIG every 5 years, DCBE every
5 to 10 years, or COL every 10 years) were dominated, the choice of screening
strategy in clinical practice should be determined not just by cost-effectiveness
but also by provider competence and patient preferences. A 1-time screen at
55 years of age with COL can achieve a 30% to 50% reduction in CRC mortality,
depending on the level of compliance. Although further reductions in mortality
can be accomplished with repeated screening, significant progress in reducing
CRC mortality can be achieved with a single screen.