Prevention of Colorectal Cancer With Low-dose Aspirin
Prevention of Colorectal Cancer With Low-dose Aspirin
The primary end points of this analysis address the following:
The secondary end points of this analysis address the following:
To address these issues, we simulated primary prevention with low-dose aspirin and secondary prevention with either colonoscopy or sigmoidoscopy screening, as well as the possible interaction between the two, in a theoretical cohort of 100 000 male and female American citizens from 50 to 100 years of age generated by a Markov model (supplementary figure 1). Age-/size-/site-specific prevalences of non-advanced and advanced adenomas, as well as of hyperplastic polyps, were matched with estimates from autopsy and endoscopic data in order to compute the costs related with polypectomy and follow-up when an endoscopic screening was simulated (online appendices 1 and 2). In detail, endoscopic screening was simulated to be repeated every 10 years between 50 and 80 years of age, with postpolypectomy surveillance differing according to polyp size and histology. Age- and site-related CRC incidence and mortality were integrally assumed from SEER (Surveillance, Epidemiology and End Results) database for the natural history cohort. Overall and site-specific reduction of CRC incidence and mortality by primary prevention with aspirin and/or endoscopic screening were extracted from the available literature. Natural attrition by the annual age-specific death rate of the US population was also simulated.
Primary prevention was simulated as the daily administration of 75 mg of aspirin between 50 and 80 years of age (this dose was chosen since it is as effective as higher doses). According to the recent pooling of randomised trials, CRC incidence and mortality were assumed to be reduced by 38% and 52%, respectively, by aspirin prevention. This prevention appeared to be limited to a 65% and 76% reduction in incidence and mortality of proximal CRC and, to a lesser extent, to a 42% and 53% reduction for rectal cancer, respectively, while no protection for distal CRC was found. This site-specific pattern was adopted in the reference case scenario. Because of the delayed effect of aspirin on CRC prevention, aspirin efficacy was simulated to begin only after 5 years of daily administration. Aspirin efficacy has been shown to last for several years after treatment cessation. Since we assumed a 10-year duration for the efficacy of endoscopic screening, we preferred to simulate the same duration for the post-treatment effect of aspirin prevention. The mechanism of action of aspirin on colorectal carcinogenesis is yet to be clarified. Consequently, we assumed independence between the (adenomatous) polyp and CRC compartments, so that any aspirin-related reduction of CRC incidence or mortality was not associated with a corresponding decrease in age- and size-specific polyp prevalences in the reference case scenario, while this possibility was explored in the sensitivity analysis. Since we integrally applied the estimates of CRC prevention shown by the pooling of cardiovascular trial to our simulated cohort of subjects undertaking CRC screening, we implicitly assumed the same compliance with aspirin treatment between the two conditions (ie, cardiovascular and CRC primary prevention). Previous studies on prevention of postpolypectomy adenomatous recurrence with aspirin showed indeed a very similar compliance with aspirin treatment to that achieved in the trials on cardiovascular prevention.
Efficacy of the simulated screening with colonoscopy was adopted from a recent population-based case–control study, in which screening colonoscopy was shown to prevent 56% and 84% of proximal and distal (including the rectum) CRC, respectively. Since proximal CRC protection is likely to be operator dependent, and no practical reduction of CRC incidence and mortality in the first 10 years following a negative colonoscopy was shown in other studies, a progressive reduction of proximal CRC prevention by colonoscopy has been simulated in the sensitivity analysis. Efficacy of sigmoidoscopy screening has been integrally adopted from a recent randomised trial on flexible sigmoidoscopy. In detail, we assumed a 50% reduction in distal CRC incidence, while the degree of reduction in distal CRC mortality was calibrated in order to match the 43% overall CRC mortality prevention reported from this study (ie, the study did not separately reported data on CRC mortality for proximal and distal CRC). When the addition of aspirin on endoscopic screening (or vice versa) was simulated, we applied the CRC prevention rates of aspirin over the residual CRC risk remaining in the population after assuming the initial efficacy of sigmoidoscopy or colonoscopy screening.
Complications were simulated for all the strategies. Endoscopy complications and related death rates were adopted from a recent US-based survey. Upper gastrointestinal bleeding (UGB), haemorrhagic stroke and related mortality with low-dose aspirin were estimated from the literature, as detailed in online appendix 1.
Reimbursement data for direct costs of endoscopy and related complications, as well as for stage-specific CRC treatment, were based on Medicare data. One-year wholesale cost for a daily administration of 75 mg (81 mg) aspirin was estimated to be US$3 at the Indiana University Medical Center pharmacy, Indianapolis, Indiana, USA. The cost of aspirin-related complications, namely, UGB and haemorrhagic stroke, was also included (see appendix 1). All costs were adjusted to 2010 US dollars using the Medical Consumer Price Index.
The clinical effectiveness of screening is measured in terms of life-years gained through prevention or downstaging of all the included diseases. In the natural history and screening models, the life-years lost by the age-dependent proportion of patients dying prematurely of CRC or aspirin-related complications are accumulated for each cycle during the entire expected lifetime. The number of life-years gained by screening corresponds to the difference in life-years lost from CRC between a Markov model with and one without screening. Future costs and future life-years saved were discounted using an annual rate of 3%. Strategies that were more costly and less effective were ruled out by simple dominance. Strategies that were more costly and less effective than a combination of other strategies were ruled out by weak dominance. The relative performance of the remaining strategies was measured using the incremental cost-effectiveness ratio (ICER), defined as the additional cost of a specific strategy, divided by its additional clinical benefit, compared with the next least expensive strategy. An ICER of US$50 000 per life-year gained was used as willingness-to-pay threshold to differentiate an efficient procedure from an inefficient one. Since the cost-effectiveness of the addition of aspirin over endoscopy is independent from the initial adherence to endoscopy screening, being equally applied to the two strategies, we assumed a 100% adherence to initial endoscopic screening in order to simplify the interpretation of the model outcomes. However, we simulated different rates of initial adherence in online appendix 3.
One- and two-way sensitivity analyses were performed for all the variables of the model, with the results being reported for those most relevant (see online appendix 4). To estimate the distribution of expected costs and efficacies of the screening strategies dependent on the uncertainty in the input parameters, we used Monte Carlo simulation to repeatedly sample from the distributions assigned to all the uncertain parameters shown in online appendix 1. In detail, β distributions were chosen for accuracy and adherence parameters, triangular distributions for costs and lognormal for the natural history transition rates. The model was simulated by using Excel spreadsheets (Microsoft Corp., Redmond, Washington, USA) and @risk 5.0 (Palisade Corp., Ithaca, New York, USA). All the input assumptions and corresponding ranges have been reported in online appendix 1.
Methods
The primary end points of this analysis address the following:
Is the addition of low-dose aspirin to colonoscopy screening cost-effective, when assuming a suboptimal colonoscopy-related proximal CRC protection?
Is the addition of low-dose aspirin to sigmoidoscopy screening cost-effective, when assuming no sigmoidoscopy-related proximal CRC protection?
What is the minimum level of colonoscopy-related proximal CRC protection at which aspirin addition is not cost-effective?
The secondary end points of this analysis address the following:
Is primary prevention with low-dose aspirin cost-effective as compared to no screening or endoscopic screening?
Is sigmoidoscopy or colonoscopy screening cost-effective when added to patients who have already taken aspirin for cardiovascular prevention?
To address these issues, we simulated primary prevention with low-dose aspirin and secondary prevention with either colonoscopy or sigmoidoscopy screening, as well as the possible interaction between the two, in a theoretical cohort of 100 000 male and female American citizens from 50 to 100 years of age generated by a Markov model (supplementary figure 1). Age-/size-/site-specific prevalences of non-advanced and advanced adenomas, as well as of hyperplastic polyps, were matched with estimates from autopsy and endoscopic data in order to compute the costs related with polypectomy and follow-up when an endoscopic screening was simulated (online appendices 1 and 2). In detail, endoscopic screening was simulated to be repeated every 10 years between 50 and 80 years of age, with postpolypectomy surveillance differing according to polyp size and histology. Age- and site-related CRC incidence and mortality were integrally assumed from SEER (Surveillance, Epidemiology and End Results) database for the natural history cohort. Overall and site-specific reduction of CRC incidence and mortality by primary prevention with aspirin and/or endoscopic screening were extracted from the available literature. Natural attrition by the annual age-specific death rate of the US population was also simulated.
Primary prevention was simulated as the daily administration of 75 mg of aspirin between 50 and 80 years of age (this dose was chosen since it is as effective as higher doses). According to the recent pooling of randomised trials, CRC incidence and mortality were assumed to be reduced by 38% and 52%, respectively, by aspirin prevention. This prevention appeared to be limited to a 65% and 76% reduction in incidence and mortality of proximal CRC and, to a lesser extent, to a 42% and 53% reduction for rectal cancer, respectively, while no protection for distal CRC was found. This site-specific pattern was adopted in the reference case scenario. Because of the delayed effect of aspirin on CRC prevention, aspirin efficacy was simulated to begin only after 5 years of daily administration. Aspirin efficacy has been shown to last for several years after treatment cessation. Since we assumed a 10-year duration for the efficacy of endoscopic screening, we preferred to simulate the same duration for the post-treatment effect of aspirin prevention. The mechanism of action of aspirin on colorectal carcinogenesis is yet to be clarified. Consequently, we assumed independence between the (adenomatous) polyp and CRC compartments, so that any aspirin-related reduction of CRC incidence or mortality was not associated with a corresponding decrease in age- and size-specific polyp prevalences in the reference case scenario, while this possibility was explored in the sensitivity analysis. Since we integrally applied the estimates of CRC prevention shown by the pooling of cardiovascular trial to our simulated cohort of subjects undertaking CRC screening, we implicitly assumed the same compliance with aspirin treatment between the two conditions (ie, cardiovascular and CRC primary prevention). Previous studies on prevention of postpolypectomy adenomatous recurrence with aspirin showed indeed a very similar compliance with aspirin treatment to that achieved in the trials on cardiovascular prevention.
Efficacy of the simulated screening with colonoscopy was adopted from a recent population-based case–control study, in which screening colonoscopy was shown to prevent 56% and 84% of proximal and distal (including the rectum) CRC, respectively. Since proximal CRC protection is likely to be operator dependent, and no practical reduction of CRC incidence and mortality in the first 10 years following a negative colonoscopy was shown in other studies, a progressive reduction of proximal CRC prevention by colonoscopy has been simulated in the sensitivity analysis. Efficacy of sigmoidoscopy screening has been integrally adopted from a recent randomised trial on flexible sigmoidoscopy. In detail, we assumed a 50% reduction in distal CRC incidence, while the degree of reduction in distal CRC mortality was calibrated in order to match the 43% overall CRC mortality prevention reported from this study (ie, the study did not separately reported data on CRC mortality for proximal and distal CRC). When the addition of aspirin on endoscopic screening (or vice versa) was simulated, we applied the CRC prevention rates of aspirin over the residual CRC risk remaining in the population after assuming the initial efficacy of sigmoidoscopy or colonoscopy screening.
Complications were simulated for all the strategies. Endoscopy complications and related death rates were adopted from a recent US-based survey. Upper gastrointestinal bleeding (UGB), haemorrhagic stroke and related mortality with low-dose aspirin were estimated from the literature, as detailed in online appendix 1.
Costs
Reimbursement data for direct costs of endoscopy and related complications, as well as for stage-specific CRC treatment, were based on Medicare data. One-year wholesale cost for a daily administration of 75 mg (81 mg) aspirin was estimated to be US$3 at the Indiana University Medical Center pharmacy, Indianapolis, Indiana, USA. The cost of aspirin-related complications, namely, UGB and haemorrhagic stroke, was also included (see appendix 1). All costs were adjusted to 2010 US dollars using the Medical Consumer Price Index.
Cost-effectiveness Analysis
The clinical effectiveness of screening is measured in terms of life-years gained through prevention or downstaging of all the included diseases. In the natural history and screening models, the life-years lost by the age-dependent proportion of patients dying prematurely of CRC or aspirin-related complications are accumulated for each cycle during the entire expected lifetime. The number of life-years gained by screening corresponds to the difference in life-years lost from CRC between a Markov model with and one without screening. Future costs and future life-years saved were discounted using an annual rate of 3%. Strategies that were more costly and less effective were ruled out by simple dominance. Strategies that were more costly and less effective than a combination of other strategies were ruled out by weak dominance. The relative performance of the remaining strategies was measured using the incremental cost-effectiveness ratio (ICER), defined as the additional cost of a specific strategy, divided by its additional clinical benefit, compared with the next least expensive strategy. An ICER of US$50 000 per life-year gained was used as willingness-to-pay threshold to differentiate an efficient procedure from an inefficient one. Since the cost-effectiveness of the addition of aspirin over endoscopy is independent from the initial adherence to endoscopy screening, being equally applied to the two strategies, we assumed a 100% adherence to initial endoscopic screening in order to simplify the interpretation of the model outcomes. However, we simulated different rates of initial adherence in online appendix 3.
Sensitivity Analysis
One- and two-way sensitivity analyses were performed for all the variables of the model, with the results being reported for those most relevant (see online appendix 4). To estimate the distribution of expected costs and efficacies of the screening strategies dependent on the uncertainty in the input parameters, we used Monte Carlo simulation to repeatedly sample from the distributions assigned to all the uncertain parameters shown in online appendix 1. In detail, β distributions were chosen for accuracy and adherence parameters, triangular distributions for costs and lognormal for the natural history transition rates. The model was simulated by using Excel spreadsheets (Microsoft Corp., Redmond, Washington, USA) and @risk 5.0 (Palisade Corp., Ithaca, New York, USA). All the input assumptions and corresponding ranges have been reported in online appendix 1.
