SLIDE 2 NOTES: Discrete-event simulation employs a computer modeling technique in which system evolution over time is represented by variables that change instantaneously at discrete points in time called “events” (e.g. adenoma incidence). The model moves forward through time by advancing from one event to the next. A discrete-event structure gives the modeler the ability to assign specific attributes to each simulated patient (e.g. the underlying risk of adenoma incidence).
The model is programmed in Insight, a FORTRAN-based programming simulation language.
SLIDE 3 NOTES: All colorectal cancers (CRCs) originate only from pre-existing adenomas (this is a fair assumption for a U.S. population where the development of CRC from flat lesions has rarely been described).
The major steps in neoplasm development include normal tissue, adenoma, asymptomatic CRC and symptomatic CRC.
The variables that define each step in neoplasm development include location, size and stage.
The underlying genetic risk of sporadic (non-familial) CRC principally controls the adenoma incidence rate. In this model (as in the actual US population it simulates), each person has a differential risk of developing colorectal cancer in their lifetime. We used that risk distribution to predict the adenoma incidence.
SLIDE NOTES 4: The health states that colorectal neoplasms can create include the no known adenoma, the known adenoma, the known CRC, the treated CRC and the death states.
The CRC dependent health states are defined by what was clinically known about the number and nature of any underlying colorectal neoplasms. Anything that alters clinical knowledge such as diagnostic tests or surgical interventions can lead to transitions between the adenoma and CRC states.
SLIDE 5 NOTES: Many assumptions went into building the model. The most important are listed in this slide. First, we assumed that all colorectal cancers originate from pre-existing adenomas. Recently we have talked to many experts in the field, and for the most part this idea is widely accepted. The real question is whether all pre-cancerous adenomas can be seen on optical colonoscopy. If not, then some cancers would appear to emerge de-novo, when in fact they have not been detected in their pre-cancerous state.
Our models assume that all adenomas progress to colorectal cancer, but some progress relatively quickly while others progress very slowly. The alternative hypothesis—that some never progress to cancer—is not inconsistent with our model, which assigns a such a long transition time to slow-growing polyps that they are very unlikely to progress in a person’s lifetime.
The dwelling times for these adenomas (the slow-growing ones) were fitted through an iterative process to match the autopsy data on adenoma prevalence and SEER on SEER data on CRC incidence. The fit resulted in a mean progression time for fast-progressing adenomas of 26 years; and for slow-progressing adenomas of 75 years.
The rate of progression from asymptomatic to symptomatic CRC is based on a previous mathematical analysis using the prevalence of malignant polyps to predict the mean latency period before CRC becomes symptomatic (4.8 years). Using this information and data on the relative prevalence of different stages of CRC at symptomatic diagnosis, distribution of times for the progression of CRC from local to regional, from local to distant, and from asymptomatic to symptomatic disease were fitted. CRC can be “discovered” within the model based on either the appearance of
symptoms or detection using a diagnostic test (e.g. colonoscopy). Stage-dependent survival following CRC diagnosis was taken directly from SEER data. Data for model calibration came from a variety of sources (Arminski and McLean, 1964; Blatt, 1961; Eide, 1986; Eide and Stalsberg, 1978; Hofstad et al., 1996; Hardcastle et al., 1996; Koretz, 1993; Kronborg et al., 1996; Mandel et al., 1993; Williams, et al., 1982; Vatn, 1982; Winawer et al., 1993).
SLIDE 6 NOTES: The National Polyp Study followed a cohort of 1400 people with identified adenomas on screening for 6 years following polypectomy. They found 5 asymptomatic cancers during follow-up (Winawer et al., 1993). Our model programmed with their protocol predicted that 3+3 asymptomatic cancers would be found.
SLIDE 7 NOTES: This chart shows the fit of the model adenoma prevalence with age to the autopsy data (Arminski and McLean, 1964; Blatt, 1961; Correa et al., 1977; Eide 1986; Eide and Stalsberg, 1978; Vatn and Stalsberg, 1982; Williams et al., 1982). This graph plots prevalence in terms of persons with adenomas/100 people against patient age. We fit the data separately for males and females. This chart displays the fit for males.
SLIDE 8 NOTES: This chart shows (via the line) the expected number of polyps in different age ranges for a simulated population of 100,000 patients based on SEER data for 1996 (SEER, 1996).
The bars are the number of polyps produced by the model in each of these age ranges. We calculated a total chi-square of 6.04 for the fit of 12 different 5-year age ranges between 30 and 95 years of age. The goodness-of-fit of the two distributions over 12 5-year blocks between age 30 and 95 was high (Chi-Square statistic=6.04).
We wish to point out that our model not only generates adenomas, but it also adds a background rate of non-adenomatous polyps (e.g., hyperplastic). The well-known existence of such polyps, whose prevalence has been estimated from screening colonoscopy studies, could be found during endoscopic or radiological screenings. The removal of such lesions alters costs and effectiveness of screening in two ways. First, the cost of removal and biopsy would be incurred. But, if such lesions are discovered on sigmoidoscopy or radiology, they might generate a full optical colonoscopy. That followup procedure could well find an adenoma by random happenstance.
SLIDE 9 NOTES: Our assumptions about follow up and surveillance are not shown here, but I want to point out that we assume that all polyps found on sigmoidoscopy are assumed to be referred for follow-up colonoscopy without biopsy. That assumption makes sigmoidoscopy a more costly strategy than it would be if only those polyps found to meet certain high-risk criteria were included.
In determining who is referred to followup for a positive test, and what happens to them on the followup examination, we interpret test specificity as reported in the literature to imply specificity for all polyps (adenomatous and non-adenomatous) in the case of endoscopic or radiologic screening modalities, but not in the case of FOBT tests (Allison et al., 1996; Allison et al., 1990). If a test is positive, it is considered a true positive if a polyp or cancer of any kind if found. If the polyp turns out to be hyperplastic, we do not consider the test to be a false-positive. This approach to the interpretation of specificity, which appears to differ from the interpretation of other models described at this workshop, has important implications for the outcome of cost-effectiveness analyses. We discuss this issue further later in the presentation.
SLIDE 10 NOTES: We based our adherence (compliance) assumptions on the Behavioral Risk Factor Surveillance System. In that survey, 70 percent of people report ever having had any CRC screening modality in their lifetime. In the population, of course, people can choose one or more screening modalities from the milieu that are available. Although people report different rates of adherence for different modalities, we do not know how they would behave if they were offered just one. In the modeling context, we must evaluate a specific screening strategy that excludes the full range of choice currently available in the community.
Among those patients adherent with any screening, (i.e., 70 percent) there were different sub-populations adherent at different maximum testing frequencies. This maximum testing frequency was programmed as a patient-specific characteristic. In practice, this created rates of effective adherence that varied between testing modalities and screening strategies (e.g. effective adherence for annual FOBT=35 percent).
SLIDE 11 NOTES: Patient compliance with post-polypectomy colonoscopic surveillance was set at 80 percent of the indicated population (Schoen et al., 2002). Simulated patients assigned to limited adherence with screening were more likely to be modeled as non-compliant with surveillance. Patients were assumed to be 100% adherent with surveillance after CRC.
SLIDE 12 NOTE: Procedure costs were based on Medicare reimbursement, while cancer treatment costs were derived from the experience of an HMO (Sonnenberg, et al., 2000; Taplin et al., 1995). Those costs are reported by in three phases—initial care, continuing care, and terminal care. For those patients who lived a short period of time after diagnosis, we assigned them randomly to initial care or terminal care.
SLIDE 13 NOTES: Because of its structure as a discrete-event simulation, we can—and in other contexts have—include quality of life adjustments (Ness et al., 2000). In fact, we can assign any number of specific attributes (e.g., sex, race, age, family history) to each simulated patient. The model can accommodate multiple interventions across the full range of prevention, screening and treatment. One could examine, for example, the overall cost-effectiveness of decrease cancer risk at the same time as screening individuals for cancer.
We can also model the prognostic capabilities of diagnostic testing modalities. Since each individual is assigned a unique risk of developing colorectal cancer, a patient with higher risk of developing colorectal cancer is likely to have more or larger colorectal adenomas, which then allows us to determine whether or not he or she should enter surveillance. That way, patients who end up in surveillance are usually a higher-risk group than people who are not.
SLIDE 14 NOTES: The following remarks relate to issues we discovered when we first received the results of the pre-workshop modeling exercise. (Note: Michael Pignone summarizes those results in another presentation.)
We noticed that, when all parameters were standardized to values provided by the organizers of the pre-workshop exercise (Run #6-see Pignone presentation), the Vanderbilt model’s lifetime costs appear to be much higher than the lifetime costs reported by all other models.
In piecing together the reasons for this seeming anomaly, we realized that our model explicitly recognizes that there exists a non-negligible prevalence of non-adenomatous polyps. Some screening tests detect such lesions without being able to differentiate between them and adenomas. In particular, radiology and sigmoidoscopy would identify them as polyps requiring follow up. When that is the case, recognizing their existence in a model generates additional follow-up procedures and additional costs. In addition, there may be changes in the effect on years of life lived for reasons described below.
SLIDE 16 NOTES: The left-hand chart shows the comparisons among models when Vanderbilt’s model assumed that non-adenomatous exist and will be detected by certain screening tests. The right-hand chart shows the results when the Vanderbilt model excludes non-adenomas. When we removed the non-adenomatous polyps from the model, the costs generated by the Vanderbilt model, especially for the strategies involving sigmoidoscopy were more in line with the costs recognized for the rest of the group.
SLIDE 18 NOTES: One would expect that eliminating non-adenomatous polyps from the model could affect years of life lived, first, because some individuals undergoing sigmoidoscopy or radiology would not be sent for followup colonoscopy. But, followup colonoscopy can detect adenomas serendipitously in the proximal colon. On the other hand, there would be a small increase in effectiveness because of a decrease in mortality from complications of followup colonoscopy. The attached charts show that there is, on balance, decreased effectiveness associated with the strategies that involve sigmoidoscopy when non-adenomatous polyps are excluded from the model.
SLIDE 19 NOTES: The chart in this slide shows that excluding non-adenomas has the largest incremental effect on lifetime costs for screening tests that identify lesions by their physical structure (radiology and sigmoidoscopy) and require followup colonoscopy for confirmation and removal. Still, the FOBT and colonoscopy tests were affected somewhat by removal of the non-adenomatous polyps from the model. That is because any positive FOBT would lead to a followup colonoscopy, where non-adenomatous lesions would be discovered serendipitously and removed. With colonoscopy, those lesions would be found during the screening test and removed
The small reductions in effectiveness from excluding non-adenomatous polyps are also concentrated in the strategies involving radiology or flexible sigmoidoscopy.
To conclude, the change in effectiveness is dwarfed by the change in cost that results from excluding non-adenomas from the universe of colorectal lesions in our model.
REFERENCES
Allison JE, Feldman R, Tekawa IS. 1990. Hemoccult screening in detecting colorectal neoplasm: sensitivity, specificity, and predictive value. Long-term follow-up in a large group practice setting. Ann Intern Med. 112:328–333.
Allison JE, Tekawa IS, Ransom LJ, Adrain AL. 1996. A comparison of fecal occult-blood tests for colorectal-cancer screening. N Engl J Med. 334:155–159.
Arminski TC, McLean DW. 1964. Incidence and distribution of adenomatous polyps of the colon and rectum based on 1,000 autopsy examinations. Dis Colon Rectum. 7:249–261.
Blatt LJ. 1961. Polyps of the colon and rectum: incidence and distribution. Dis Colon Rectum. 4:277–282.
Correa P, Strong JP, Reif A, Johnson WD. 1977. The epidemiology of colorectal polyps: prevalence in New Orleans and international comparisons. Cancer. 39:2258–2264.
Behavioral risk factor surveillance system online prevalence data, 1995–2001. http://apps.nccd.cdc.gov/brfss; 5/5/2003.
Eide TJ, Stalsberg H. 1978. Polyps of the large intestine in Northern Norway. Cancer. 42:2839–2848.
Eide TJ. 1986. The age-, sex-, and site-specific occurrence of adenomas and carcinomas of the large intestine within a defined population. Scand J Gastroenterol. 21:1083–1088.
Hardcastle JD, Chamberlain JO, Robinson MH, Moss SM, Amar SS, Balfour TW et al. 1996. Randomised controlled trial of faecal-occult-blood screening for colorectal cancer. Lancet. 348:1472–1477.
Hofstad B, Vatn MH, Andersen SN, Huitfeldt HS, Rognum T, Larsen S, Osnes M. 1996. Growth of colorectal polyps: redetection and evaluation of unresected polyps for a period of three years. Gut. 132:449–456.
Koretz RL. 1993. Malignant polyps: are they sheep in wolves’ clothing? Ann Intern Med. 118:63–68.
Kronborg O, Fenger C, Olsen J, Jorgensen OD, Sondergaard O. 1996. Randomised study of screening for colorectal cancer with faecal- occult-blood test. Lancet. 348:1467–1471.
Mandel JS, Bond JH, Church TR, Snover DC, Bradley GM, Schuman LM et al. 1993. Reducing mortality from colorectal cancer by screening for fecal occult blood. Minnesota Colon Cancer Control Study. N Engl J Med. 328:1365–1371.
Ness RM, Klein R, Holmes AM, Dittus R. 2000. At what age is one-time colonoscopic screening for colorectal cancer most cost-effective? Am J Gastroenterol. 95:1800–1811.
Schoen RE, Wissfeld JL, Trauth JM, Ling BS, Hayran M. 2002. A population-based, community estimate of total colon examination: the impact of compliance with screening for colorectal cancer. Am J Gastroenterol. 97:446–451.
SEER cancer incidence public-use database, 1973–93. 1996. Department of Health and Human Services: Bethesda, MD.
Sonnenberg A, Delco F, Inadomi JM. 2000. Cost-effectiveness of colonoscopy in screening for colorectal cancer. Ann Intern Med. 133:573–584.
Taplin SH, Barlow W, Urban N, Mandelson DJ, Ichikawa L, Nefcy P. 1995. Stage, age, comorbidity, and direct costs of colon, prostate, and breast cancer care. J Natl Cancer Inst. 87:417–426.
Vatn MH, Stalsberg H. 1982. The prevalence of polyps of the large intestine in Oslo: an autopsy study. Cancer. 49:819–825.
Williams AR, Balasooriya BAW, Day DW. 1982. Polyps and cancer of the large bowel: a necropsy study in Liverpool. Gut. 23:835–842.
Winawer SJ, Zauber AG, Ho MN, O’Brien MJ, Gottlieb LS, Sternberg SS et al. 1993. Prevention of colorectal cancer by colonoscopic polypectomy. The National Polyp Study Workgroup. N Engl J Med. 329:1977–1981.