|Cost-effectiveness analysis of strategies for colorectal cancer screening in Japan
|Shimbo T, Glick H A, Eisenberg J M
This is a critical abstract of an economic evaluation that meets the criteria for inclusion on NHS EED. Each abstract contains a brief summary of the methods, the results and conclusions followed by a detailed critical assessment on the reliability of the study and the conclusions drawn.
Several screening strategies for colorectal cancer (CRC) were examined. These were based on the combination of biochemical faecal occult blood testing (F), immunological faecal occult blood testing (IF), barium enema (B), colonoscopy (C) and sigmoidoscopy (S). The strategies were as follows.
Strategy 1 was F-F-B-C. F was performed annually, followed by second F if the first results was positive. The patient underwent a B if the second test result was positive and C if the B result was positive.
Strategy 2 was IF2y-C. IF was performed every 2 years and, if positive, was followed by C.
Strategy 3 was F-C. F was performed annually and, if positive, was followed by C.
Strategy 4 was IF-B-C. IF was performed annually and, if positive, was followed by a B. If the B was suggestive of cancer beyond the range of S, the patients underwent C, while if the B was suspicious within range, then S was conducted.
Strategy 5 was IF-B/S-C. This was similar to strategy 4, but both a B and S were performed after a positive IF result. If S revealed a lesion, then examination was stopped. If S was negative but the B was positive, C was performed.
Strategy 6 was IF-C. IF was performed annually and, if positive, was followed by C.
Strategy 7 was IF/S3y-C. IF was performed annually and, if positive, was followed by C. In addition, S was performed every 3 years and, if positive, was followed by C.
Strategy 8 was no screening.
Economic study type
The study population comprised a hypothetical cohort of asymptomatic 40-year-old Japanese men and women. Other starting ages were also considered.
The setting appears to have been primary care. The economic study was conducted in Japan.
Dates to which data relate
The effectiveness data were derived from studies published between 1975 and 1990. No dates for the resource use data were explicitly reported. The costs were, in part, obtained from studies published in 1989 and 1990. The price year was not reported.
Source of effectiveness data
The effectiveness evidence was derived from a synthesis of completed studies and authors' assumptions.
A state-transition model was constructed to simulate the natural history of CRC in a cohort of 100,000 asymptomatic men and women in Japan. The model was populated mainly with published data specific to the Japanese setting. Individuals were followed for 35 years. The cycle length appears to have been one year. Six different health states were considered. These were cancer-free and polyp-free, polyp remains benign, polyps transforms into cancer, CRC, follow-up after cancer treatment, and death.
Outcomes assessed in the review
The outcomes derived from the literature were the sensitivity and specificity of the alternative screening tests, the complication rates, and other probability values used in the decision model.
Study designs and other criteria for inclusion in the review
It was not stated whether a formal review of the literature was undertaken and the design of the primary studies was unclear. The stage- and age-specific distribution of undetected cancers and estimated detection rates were derived from observational data in a Japanese regional cancer registry. The probabilities of death were derived from Japanese life tables. Only those studies that specified clinical data were considered.
Sources searched to identify primary studies
Criteria used to ensure the validity of primary studies
Methods used to judge relevance and validity, and for extracting data
Number of primary studies included
Forty primary studies provided the evidence.
Methods of combining primary studies
The primary studies appear to have been combined using narrative methods. The 95% confidence interval (CI) was derived from the sum of total cases using the method of Blyth and Still. Ranges (minimum and maximum) were also reported when the data were derived from more than two studies.
Investigation of differences between primary studies
Results of the review
The specificity of F was 0.886 (95% CI: 0.869 - 0.901).
The sensitivity of F was 0.167 (95% CI: 0.142 - 0.196) for polyps, 0.263 (95% CI: 0.140 - 0.438) for stage A cancer, and 0.588 (95% CI: 0.539 - 0.635) for stage B/C cancer.
The specificity of IF was 0.991 (95% CI: 0.975 - 0.997).
The sensitivity of IF was 0.037 (95% CI: 0.022 - 0.079) for polyps, 0.481 (95% CI: 0.342 - 0.622) for stage A cancer, and 0.843 (95% CI: 0.763 - 0.898) for stage B/C cancer.
The sensitivity of S was 0.9 for polyps and 0.957 for cancer.
The specificity of a B was 0.978 (95% CI: 0.941 - 0.993).
The sensitivity of a B was 0.887 (95% CI: 0.847 - 0.918) for polyps and 0.917 (95% CI: 0.899 - 0.931) for cancer.
The sensitivity of C was 0.9 (95% CI: 0.875 - 0.921) for polyps and 0.957 (95% CI: 0.843 - 0.993) for cancer.
The rate of major bleeding was 0 with S and B, 0.000085 (95% CI: 0.000027 - 0.00023) with diagnostic C, and 0.00662 (95% CI: 0.005502 - 0.007959) with polypectomy.
The rate of perforation was 0.000143 (95% CI: 0.000007 - 0.000927) with S, 0.000168 (95% CI: 0.000088 - 0.000310) with a B, 0.00197 (95% CI: 0.00168 - 0.002311) with diagnostic C, and 0.00377 (95% CI: 0.003048 - 0.004652) with polypectomy.
Fifty per cent of cancer was derived from polyps, and a polyp took 7 years to transform into CRC.
Of all polyps, S detected 60.6%.
The case-fatality rate was set at 0.818 for perforation due to B and 0.0966 for perforation due to C.
Methods used to derive estimates of effectiveness
The authors made some assumptions that were used in the decision model.
Estimates of effectiveness and key assumptions
The authors assumed the following:
97% of polyps remained benign;
the probability of detecting both benign and malignant polyps in the no screening model was 5 polyps per 1,000 persons per year (which estimates 50% more observed cases of polyps than the incidence of CRC);
the probability of detecting both types of polyps in the screening models was the same as the sensitivity of screening examinations;
the probability of polyps that could be resected was 50%;
individuals with resected polyps re-entered the polyp-free and cancer-free health state in the following year;
undetected cancer remained in Duke's Stage A for 2 years, in Stage B for one year, and in Stage C for one year, at which point its presence became evident clinically if it were still undetected;
the sensitivity of S was the same as that of C within the range of S; and
the specificity of S and C were 1;
In addition, the sensitivity and specificity of each combination for detecting CRC was calculated assuming that the test characteristics of each examination were independent.
Measure of benefits used in the economic analysis
The summary benefit measure was the number of years of life saved (YOLS). These were obtained from modelling. An annual discount rate of 5% was applied, but the undiscounted results were also reported. The number of perforations was also estimated and reported.
A 5% discount rate was applied as the costs were incurred over a long timeframe. The unit costs were presented, but information on the quantities of resources used was less clear. The health services included in the economic evaluation were F, IF, S, B, C, bowel preparation, biopsy, complication (major bleeding and perforation), pre-hospital work, initial treatment of cancer or polypectomy, outpatient clinic visits, and terminal care. The cost/resource boundary of the payer was adopted. Resource use was mainly estimated from authors' assumptions and some published data. The costs were mainly based on charges, which might not have reflected the true costs, but which represented the costs from the perspective of the payer. The cost of C was derived from experts' opinions, while other costs were estimated from 1989 and 1990 rates. The price year was not reported.
Statistical analysis of costs
The costs were treated deterministically.
The indirect costs were not considered.
Japanese yen (Y). The exchange rate was Y135 = 1 US dollar ($).
Univariate sensitivity analyses were carried out to assess the robustness of the cost-effectiveness ratios to variations in a number of variables. The variables considered included the sensitivity and specificity of screening, the probability of cancer development from adenomatous polyps, the probability of polyp detection and treatment, and the interval during which polyps transformed to cancer. Also investigated were survival rates after cancer treatment, cost estimates, and discount rates. The ranges used in the analysis were derived from CIs obtained from the literature. The ranges of the cost estimates were not reported. Different ages of initiation and compliance rates were also considered.
Estimated benefits used in the economic analysis
The estimated discounted (undiscounted) YOLS per 100,000 persons screened with strategies 1 to 7 over no screening were:
strategy 1, 92 (405);
strategy 2, 1,073 (4,378);
strategy 3, 1,331 (5,506);
strategy 4, 1,485 (6,088);
strategy 5, 1,592 (6,481);
strategy 6, 1,610 (6,590); and
strategy 7, 2,217 (9,057).
The number of perforations per 100,000 persons screened with each strategy over no screening was 83 with strategy 1, 60 with strategy 2, 307 with strategy 3, 74 with strategy 4, 82 with strategy 5, 105 with strategy 6 and 430 with strategy 7.
The estimated total average costs per patient with strategies 1 to 8 were:
strategy 1, Y38,670;
strategy 2; Y36,090;
strategy 3, Y72,660;
strategy 4, Y50,740;
strategy 5, Y53,170;
strategy 6, Y49,850;
strategy 7, Y134,390; and
strategy 8, Y21,430.
Synthesis of costs and benefits
An incremental cost-effectiveness ratio was calculated to combine the costs and benefits of the screening strategies over no screening.
The incremental cost per YOLS with strategies 1 to 7 over no screening was:
strategy 1, Y18,683,000 ($138,400);
strategy 2, Y1,366,000 ($10,100);
strategy 3, Y3,850,000 ($28,500);
strategy 4, Y1,974,000 ($14,600);
strategy 5, Y2,006,000 ($14,900);
strategy 6, Y1,756,000 ($13,100); and
strategy 7, Y5,096,000 ($37,700).
F-F-B-C was dominated by IF2y-C, which had the best cost-effectiveness ratio among all strategies. IF-C dominated both IF-B-C and IF-B/S-C. F-C was dominated by the three strategies using annual IF. The incremental cost per YOLS from IF-C to IF/S3y-C was very high (Y13,939,000; $103,000). Therefore, the IF-C strategy was considered the preferred strategy.
The sensitivity analysis showed that, with a younger starting age, more effectiveness would have been obtained but at higher costs. Initiating screening at age 45 had the best cost-effectiveness at Y1,680,000 ($12,400) per YOLS, while the incremental cost of changing the initial screening from 45 to 40 years was greater although still relatively low (Y2,269,000; $16,800). Variations in model assumptions did not produce substantial changes and IF-C remained the preferred option. Discounting made screening less favourable, but the cost-effectiveness of IF-C remained low. The dropout rate had a significant impact on the initiation age. With a 10% dropout rate, screening at age 40 was dominated by screening at age 45. However, when 15 to 20% dropout rates were considered, screening at age 50 dominated earlier initiation ages.
The strategy of performing immunological faecal occult blood testing (IF) every 2 years, followed by colonoscopy (C) after positive results, was the most cost-effective strategy for screening for colorectal cancer (CRC) in Japan. The analysis also showed that screening should be started at age 45, or even at age 40.
CRD COMMENTARY - Selection of comparators
The rationale for the choice of the comparators was reported and all the relevant combinations of screening strategies were considered in the analysis. The no screening option was also considered, which was appropriate since it may reflect the current approach in some settings. You should decide whether they are valid comparators in your own setting.
Validity of estimate of measure of effectiveness
The analysis of effectiveness used evidence mainly derived from published studies. However, it was unclear whether a systematic review of the literature had been undertaken to identify relevant studies. No information on the primary sources, the sample size and study design was reported. Therefore, the quality of the evidence used in the model was unclear. Similarly, there was limited information on the methods used to combine the primary estimates. Other estimates were derived from authors' assumptions. Most of the key model inputs were varied in the sensitivity analysis to test the robustness of the authors' conclusions.
Validity of estimate of measure of benefit
The summary benefit measure was appropriate as it captured the impact of the interventions on the patients' health. However, quality of life issues were not considered since the analysis focused on survival. Discounting was carried out, as recommended, and the impact of variations in the discount rates was investigated. Undiscounted results were also reported. The use of YOLS makes comparisons with the benefits of other health care interventions feasible.
Validity of estimate of costs
The authors explicitly stated the perspective adopted in the study. As such, it appears that all the costs relevant to the payer have been considered in the analysis. Charges were used as proxies for the costs, although the authors acknowledged that true costs would have been more appropriate. However, it was also noted that charges relevant to the payer were used. The unit costs were presented, but the information on resource use was unclear. The source of the cost data was given for all items. The costs were obtained in 1989 and 1990, but the price year was not reported. This makes reflation exercises in other settings difficult. The costs were also presented in US dollars. No statistical analyses of the costs were carried out, but sensitivity analyses were carried out on key cost estimates.
The authors made some comparisons of their findings with those from other studies and stated that their results were comparable with those reported in the literature. Some differences with other studies were also highlighted. The issue of the generalisability of the study results to other settings was not explicitly addressed, but extensive sensitivity analyses were carried out and these demonstrated the robustness of the base-case results. The authors noted that some data came from studies conducted outside of Japan (particularly from the UK) since local data were not available. This could have introduced some uncertainty into the analysis.
Implications of the study
The study results showed that IF-C at age 45 could be considered the screening option of choice for the detection of CRC in the general population. However, when choosing between the three IF options, the frequency of complications (higher with IF-C) should be considered, particularly when considering the indirect and intangible costs of treating complications.
Shimbo T, Glick H A, Eisenberg J M. Cost-effectiveness analysis of strategies for colorectal cancer screening in Japan. International Journal of Technology Assessment in Health Care 1994; 10(3): 359-375
Other publications of related interest
Eddy DM. Screening for colorecteral cancer in a high risk population. Annals of Internal Medicine 1990;13:373-84.
Okubo I, Glick HA, Frumkin H, et al. Cost-effectiveness analysis of mass screening for breast cancer in Japan. Cancer 1991;67:2021-9.
Subject indexing assigned by NLM
Adult; Colorectal Neoplasms /diagnosis /economics; Cost-Benefit Analysis; Female; Humans; Japan; Male; Mass Screening /economics; Sensitivity and Specificity
Date bibliographic record published
Date abstract record published