Colorectal cancer is the second leading cause of cancer death in BC after lung cancer. The lifetime risk of dying of colorectal cancer is about 3%. Most, if not all, colorectal cancers are believed to develop from adenomatous polyps[1] over a period of about 10 years. Screening for colorectal cancer offers the opportunity to reduce mortality by the earlier detection of invasive disease and by the removal of precancerous polyps, thus preventing the development of colorectal cancer.

Three methods are currently considered for colorectal cancer screening. Fecal occult blood (FOB) has been proven effective in clinical trials at reducing colorectal cancer mortality.[2-4] FOB is not very effective at detecting polyps and preventing cancer,[5] but can detect cancers at earlier, more treatable stages and with better survival than at later stages. Direct visualization using flexible sigmoidoscopy can only reach as far as the splenic flexure, which includes about 60% of cancers, although 70% are identified due to colonoscopy follow-up of polyps within the viewing range.[6] Trials are currently ongoing for sigmoidoscopy in a screening program.[7] Colonoscopy is the most reliable method of detecting polyps and cancers. It is used to investigate and treat persons with positive results of other screening tests, as well as for polyp removal. However, colonoscopies have associated complications (including death), and there may be insufficient resources for the volume of colonoscopies required for a screening program in BC. Moreover, the effectiveness of colonoscopy screening has never been tested in a randomized controlled trial. Sigmoidoscopy and colonoscopy offer a better potential to prevent colorectal cancer through the identification and treatment of adenomatous polyps.

Several studies have examined the cost-effectiveness of colorectal cancer screening programs.[1,8] All three screening methods are sufficiently cost-effective to be considered for public health programs. Which method is most cost-effective depends on organizational and cost assumptions. The final choice may be determined by feasibility.

This article presents a computer model for the development of colorectal cancer in BC and uses it to estimate the effectiveness of screening strategies using three methods: colonoscopy, flexible sigmoidoscopy, and (nonrehydrated) fecal occult blood.

Method
Figure 1 [9] illustrates the assumed colorectal cancer process and was based on a model developed by Frazier and colleagues.[8] It assumes the disease process begins with the development of low-risk polyps in normal tissue. Low-risk polyps progress at defined rates into high-risk polyps, which in turn progress to asymptomatic localized cancers. These in turn progress to symptomatic localized cancers or to asymptomatic regional cancers, which in turn progress to symptomatic regional cancers or to distant cancers. Death from colorectal cancer can occur only after the disease has become symptomatic. Individuals die from causes other than colorectal cancer at any time according to population rates.

A computer simulation of the process of colorectal cancer development from normal tissue to colorectal cancer was programmed using TreeAge software.[9] The parameter values assumed by the model are given in the Appendix. Individuals move between states, corresponding to the disease process, according to annual progression rates. Progression rates are not affected by the length of time an individual is in a disease state (Markov property). The disease process can only remain in its current state or advance; it cannot regress.

Progression rates were calibrated to the observed BC incidence, mortality, and survival rates for BC males and females for the period 1995 to 1999 using data obtained from the BC Cancer Registry. Figure 2 [10] shows the obtained agreement between model-predicted and observed incidence and mortality rates. Most of the assumed model parameter values are within the range of values considered by Frazier and colleagues,[8] with the exception of the annual cancer mortality rates. In order to provide a better fit to observed mortality, the present model—in contrast to the Frazier model—assumes a proportion of cases are cured and applies the annual cancer mortality rate to the uncured cases only. Screen-detected cancers are assumed to incur a 5-percentage-point benefit in cure and mortality rates following observations by Hardcastle and colleagues.[4]

The sensitivities and specificities of the screening tests used in the model are presented in Table 1. The following management protocols for positive screens were used:

• Individuals with positive FOB or sigmoidoscopy results have the results verified by colonoscopies, which, if negative, (i.e., false-positive screens) exempt the individual from screening for at least 10 years.

• Individuals with low-risk polyps (confirmed by colonoscopy) have follow-up colonoscopies in 5 years, which, if negative, exempt the individual from screening for 10 years.

• Individuals with high-risk polyps (confirmed by colonoscopy) have follow-up colonoscopies in 3 years and every 5 years while results are negative or low risk.

• Individuals with cancer (confirmed by colonoscopy) have follow-up colonoscopies in the next and every 3 years thereafter.

Complication rates of endoscopy used in the model are given in Table 2 [10] and are based on reported rates from endoscopy services in general patient populations and not from screening programs of low-risk individuals in high-volume centres. The rates used are likely conservative and overestimate the complication rates associated with population-based screening.

To assess the effect of different screening strategies, a hypothetical population was followed from age 50 to age 89. Life expectancy, cancer incidence and mortality, and the number of screens and follow-up colonoscopies required under various screening scenarios were output from the program. This comparison assumed that there was 100% compliance for screening, regardless of screening modality, and 90% compliance with recommended follow-up.

Results
The computer simulation was run under a variety of screening scenarios, including one-time colonoscopy at ages 50, 55, 60, 65, and 70, as well as two-time colonoscopies at ages 50 and 60; 55 and 65; 60 and 70; 65 and 75; 50 and 65; 55 and 70; and 60 and 75. One- and two-time sigmoidoscopies were tested at the same ages. In addition, three-time sigmoidoscopies were tested at ages 50, 55, and 60; 55, 60, and 65; 60, 65, and 70; and 65, 70, and 75. FOB was tested annually and biennially between ages 50 and 74 and ages 55 and 74. We report here results for those ages that gave the best life expectancy results for that modality and frequency of use (Table 3). In general FOB was less effective at preventing cancers but detected more at earlier stages, and was quite effective at preventing colorectal cancer death.

We used the model to determine the amount of resources required to implement a screening program for the screening strategies presented in Table 3 . We calculated the number of screens and colonoscopies required in BC in 2004 for each screening program if it were adopted by 80% of the age-eligible population. The results are presented in Table 4. As anticipated FOB identified fewest polyps but still involved a significant requirement for colonoscopy.

Discussion
Colonoscopy is the most effective screening modality, but it also has the highest predicted number of complications. In terms of preventing death, annual FOB has approximately the effectiveness of a single sigmoidoscopy and about 75% the effectiveness of a single colonoscopy or two or more sigmoidoscopies. Two or three sigmoidoscopies are as effective as one colonoscopy and have a requirement for colonoscopy comparable to FOB, which is probably feasible given current resources.

The fact that FOB is the only method that has been proven effective by clinical trials should not disqualify colonoscopy and sigmoidoscopy from consideration. Although the effectiveness of these methods at particular screening frequencies has not been determined in randomized trials, there cannot be any doubt that they are effective at some screening frequency. This is because colonoscopies were used to validate positive FOB results in the studies that demonstrated the effectiveness of FOB.

Currently about 37 000 colonoscopies are performed each year in BC, some of which are for screening purposes. Even with a program that screened with colonoscopy, only 80% of people 55 years of age in 2004 would require 46 000 colonoscopies (Table 4), which exceeds current total capacity. If colonoscopy screening were offered to all BC residents aged between 50 and 74, there would be a latent demand for up to 1 million colonoscopies (the size of the BC population in that age range), which is clearly unfeasible. In contrast, FOB screening can be offered to the same population without placing such a huge demand on endoscopy services. But even with low compliance rates, substantial increases in endoscopy and pathology capacity would be required to accommodate any screening program.

The model used in the present computer simulation has made certain assumptions about compliance, disease prevalence, and transition probabilities between disease states. Although the assumed values have been calibrated to the observed patterns of incidence, mortality, and stage-specific survival in BC, some unobserved aspects of the model have a degree of indeterminacy that could affect the outcomes. In particular, it would be helpful to have estimates of compliance as well as of low- and high-risk polyp prevalences at various ages. A pilot project may be useful as a first step in the implementation of a population screening program for colorectal cancer, to explore efficiency measures, compliance, and the prevalence of polyps in the BC population.

Competing interests
None declared.

Table 1. Sensitivity and specificity of screening tests.

*As far as the splenic flexure and otherwise zero.

 Table 2. Complication rates of endoscopy.

*This value inferred assuming the same ratio between hemorrhages and perforations as for colonoscopy

 Table 3. Summary of effects of screening 1000 individuals followed from age 50 to death for selected screening strategies, assuming 100% compliance.

 Table 4. Number of screening tests and follow-up colonoscopies required and number of identified low-risk and high-risk polyps that would be identified from screening of 80% of the target BC population in 2004.

Appendix. Parameter values used in the model.

Norman Phillips MSc, MA and Andrew J. Coldman, PhD

Mr Phillips is a biostatistical analyst with the BC Cancer Agency (BCCA) and Dr Coldman is head of Population and Preventive Oncology at BCCA.