The U.S. Preventive Services Task Force (USPSTF) strongly recommends screening for men and women 50 years of age or older for CRC (“A” recommendation)
(3). Biannual fecal occult blood testing can reduce CRC-related death by 21%, and it has been reported that flexible sigmoidoscopy reduces death by 60% for lesions within reach of the instrument. Further, data suggest that sigmoidoscopy followed by colonoscopy when polyps are found could decrease CRC incidence by up to 80%
(4). Despite evidence of the effectiveness of several screening methods, adoption of routine CRC screening by eligible individuals, using any method, continues to be low in the United States
(5–8).
A CRC chemoprophylactic strategy may be used as a complement to or instead of a screening strategy. Several basic science, population-based, and experimental studies have suggested a protective effect of aspirin (ASA) and non-ASA nonsteroidal anti-inflammatory drugs (NSAIDs), including cyclooxygenase (COX)-2 inhibitors, on colorectal adenomas and CRC. However, 2 long-term, randomized, controlled trials, the Physicians' Health Study
(9) and the Women's Health Study
(10), did not show a beneficial effect of low-dose ASA on CRC incidence. Furthermore, these agents are not without harms. Clinically significant gastrointestinal hemorrhage can occur with all of these agents, although it is substantially lower with COX-2 inhibitors. More recently, interest has focused on a potentially prothrombotic effect of selective COX-2 inhibitors and nonnaproxen NSAIDs. In fact, during the conduct of our systematic review, 2 COX-2 inhibitors (rofecoxib and valdecoxib) were withdrawn from the U.S. market because of concerns about their cardiovascular toxicity, leaving only celecoxib remaining and uncertainty about the future of others, such as lumiracoxib and etoricoxib. These developments have resulted in uncertainty about the safety of COX-2 inhibitors and non-ASA NSAIDs when used long-term, such as in the setting of CRC prevention
(11).
At the request of the Agency for Healthcare Research and Quality, the Centers for Disease Control and Prevention (CDC), and the USPSTF, we conducted a systematic review to ascertain the effectiveness of non-ASA NSAIDs and COX-2 inhibitors in the chemoprevention of colorectal adenomas, CRC, and CRC-related death in average- to higher-risk individuals. We also examined the harms associated with these agents.
Methods
Data Sources
We developed the search strategy in MEDLINE and modified it for other databases. The search was limited to English-language reports of human studies. We searched the following databases: MEDLINE (1966 to December [week 4] 2006), EMBASE (1980 to the 14th week of 2005 publication years 2003 to 2005), Cochrane Central Register of Controlled Trials (CENTRAL), and Cochrane Library Issue 4, 2004. Beyond these dates, we surveyed several sources to ascertain additional potentially eligible studies. PubMed Cancer subset was searched for non-MEDLINE material.
Search terms were derived from the National Cancer Institute (NCI) Cancer topic searches for “colorectal cancer” and “adenomatous polyps.” We derived a comprehensive retrieval strategy from the indexing in MEDLINE and EMBASE, investigator-nominated terms, and previous reviews
(12–14).
We developed a search strategy in MEDLINE (2003 to the third week of December 2006) to detect recent systematic reviews that appeared to address the harms of non-ASA NSAIDs and COX-2 inhibitors. We implemented a weekly monitoring strategy to detect emerging information on cardiovascular harms associated with COX-2 inhibitors. We also monitored the U.S. Food and Drug Administration News Digest and Health Canada's Health Product Information mailing list for announcements related to COX-2 inhibitors and cardiovascular harms (monitoring dates, 14 January 2005 to 26 May 2005). Beyond these dates, we surveyed several sources to ascertain additional potentially eligible studies.
Study Selection
Citation records were screened to identify potentially relevant articles and retained records were assessed for relevance to identify articles meeting inclusion criteria. A third screening phase was included to discriminate between the different study designs. At each screening stage, 2 members of the review team selected articles for inclusion after an initial calibration exercise. Conflicts were resolved by consensus.
We considered randomized, controlled trials (RCTs); controlled, clinical trials; and observational studies (cohort and case–control studies) of the efficacy of non-ASA NSAIDs and COX-2 inhibitors for inclusion if they fulfilled the population and outcome criteria.
We considered studies for inclusion if participants were at average risk for CRC (that is, no known risk factors for colorectal adenoma or CRC, other than age). We also considered studies of higher-risk individuals with a personal or family history of colorectal adenoma or a family history of sporadic CRC. Included studies addressed the incidence of colorectal adenomas, CRC, or both and CRC-related death or overall death. We excluded studies of high-risk patients with familial adenomatous polyposis or hereditary nonpolyposis colon cancer syndromes (Lynch I or II) and secondary prevention studies of patients with a personal history of CRC.
We sought existing systematic reviews to address the gastrointestinal, cardiovascular, and renal harms associated with the use of non-ASA NSAIDs and COX-2 inhibitors considering the number of reviews already done on these topics.
Data Extraction and Quality Assessment
Several members of the team extracted data independently by using a Web-based system (SRS 4.0, TrialStat Corp., Ottawa, Ontario, Canada). We extracted data by using the PICOS (participant, intervention and exposure, comparator, outcome, and study design) approach.
We used predefined criteria from the USPSTF to assess the quality of included systematic reviews, clinical trials, and observational studies, which we rated as good, fair, or poor
(11). This scale relies on 4, 6, 7, and 7 criteria for systematic reviews, case–control studies, cohort studies, and RCTs, respectively. A good rating was given when all criteria were met; a fair rating when at least 80% were met and the study had no fatal flaws; and a poor rating when less than 80% of the criteria were met, when there was a fatal flaw, or both.
Data Synthesis and Analysis
We used an analytical framework to facilitate study grouping and subsequent data analysis in an effort to minimize clinical heterogeneity. We initially grouped studies by disorder (that is, colorectal adenoma or CRC), study design, study population, and medication exposure and subsequently subcategorized studies based on measures of dose effect, duration of exposure, and secondary outcomes (when reported). Definition of categories, such as “regular use,” can be found elsewhere
(11).
We summarized and presented harms data from the included systematic reviews as a qualitative synthesis.
We combined results numerically only if clinically and statistically appropriate. We chose relative risk as the effect measure. In case–control studies, a direct estimate of the relative risk is not possible. However, when event rates are low, as was the case in our review, the odds ratio provides a close approximation of the relative risk. In what follows, we simply refer to the relative risk. We assessed heterogeneity by using the I
2 statistic. We combined studies when I
2 was 50% or less
(15). We directly abstracted point estimates of the adjusted relative risks and their 95% CIs from the reports of primary studies. One source of heterogeneity may be study-to-study variation in the method of selecting confounders for which to adjust and the final set of confounders chosen. In Appendix Tables
1 and
2, we summarize these characteristics for each study. Further, a detailed discussion of the methodological considerations is presented in the USPSTF report
(11). We computed standard errors by dividing the CI width by 2 × 1.96. We conducted quantitative synthesis by using inverse variance weighting and a random-effects model
(17).
Role of the Funding Sources
The evidence synthesis on which this article was based was funded by the CDC, the Agency for Health Research and Quality, and the USPSTF. Its design, conduct, and reporting was based on specific directives from these agencies.
Discussion
Colorectal cancer is an important burden on the U.S. population. The use of NSAID chemoprophylaxis, alone or in combination with a recommended screening program, is 1 strategy to reduce the incidence of colorectal adenomas, CRC, and CRC-related death.
The results of our systematic review suggest that the use of non-ASA NSAIDs for CRC chemoprevention is effective at reducing the incidence of colorectal adenomas and CRC. Cyclooxygenase-2 inhibitors seem to be effective at reducing the incidence of colorectal adenoma in patients with previous adenomatous polyps. Higher doses and longer durations of use of non-ASA NSAIDs seem to be associated with greater protection from CRC and adenomas. We found the magnitude of the relative risk reduction for CRC incidence to be approximately 30% to 40% in the pooled analyses.
We found no observational data on the effect of COX-2 inhibitors on CRC incidence or CRC-related death, although a single cohort study showed no effect of the non-ASA NSAID ibuprofen on CRC death but demonstrated a small statistically significant increase in all-cause mortality
(21). Further, no RCT data exist on CRC incidence with the long-term use of COX-2 inhibitors or non-ASA NSAIDs that are similar to data from the ASA-based Physicians' and Women's Health studies
(9, 10).
The use of non-ASA NSAIDs and COX-2 inhibitors are each associated with important harms. Non-ASA NSAIDs are associated with an increased risk for ulcers and clinically important ulcer complications, such as hemorrhage, perforation, or pyloric obstruction. Cyclooxygenase-2 inhibitors are associated with fewer gastrointestinal symptoms, endoscopic ulcers, and clinically important ulcer complications than non-ASA NSAIDs. However, data from the APPROVe study
(18) demonstrated that over a 3-year period, COX-2 inhibitors were associated with a statistically significant increased risk for clinical ulcer complications compared with placebo
(18). Although these data are in keeping with improved gastrointestinal safety of COX-2 inhibitors over non-ASA NSAIDs, the gastrointestinal safety of COX-2 inhibitors is not equivalent to that seen with placebo, as has been suggested in the past. On the other hand, COX-2 inhibitors are associated with an increased risk for adverse cardiovascular outcomes
(56).
During the conduct of our systematic review, rofecoxib was withdrawn from the market because of the results of the polyp prevention APPROVe study
(78), which demonstrated an excess risk for cardiovascular events (16 per 1000 events) with the use of rofecoxib, confirming the suspicions reported by the VIGOR investigators
(76). Subsequently, celecoxib was also found to have an excess risk for cardiovascular events (13 to 21 per 1000 events) in another polyp prevention study (Adenoma Prevention with Celecoxib [APC])
(79). Valdecoxib was also withdrawn because of excess risk for cardiovascular events in 2 short-term cardiac surgery pain studies (Coronary Artery Bypass Graft [CABG] 1 and 2) and because of a rare dermatologic toxicity
(80, 81).
A systematic review of the cardiovascular harms of rofecoxib and non-ASA NSAIDs
(59) suggested a small cardiovascular protective effect of naproxen, although the included studies were heterogeneous. Naproxen's relatively long half-life of 14 hours makes a twice-daily dosing schedule theoretically capable of consistently blocking COX-1 and potentially providing some degree of cardioprotection. Clinical trial data of the quality comparable to data available for the COX-2 inhibitors is not available for non-ASA NSAIDs. However, a recent meta-analysis
(56) using an extensive set of RCT data derived from published and unpublished studies suggests that, as a group, COX-2 inhibitors are associated with an increased risk for adverse cardiovascular outcomes (predominantly MI) when compared with placebo or naproxen but not when compared with nonnaproxen, non-ASA NSAIDs. These data, and evidence from some population-based studies
(82–84), suggest that the increased risk for cardiovascular harms with COX-2 inhibitors is shared by nonnaproxen, non-ASA NSAIDs (higher doses of ibuprofen and diclofenac)
(56).
Although it is tempting to consider adding ASA to a COX-2 inhibitor for cardioprotection, there seems to be an attenuation of the gastrointestinal safety of COX-2 inhibitors with this strategy. However, it should be noted that these observations were derived from post hoc subgroup analyses.
Non-ASA NSAIDs and COX-2 inhibitors are used for longer durations for a variety of arthritic and inflammatory conditions
(12). Although their use for these conditions is more easily justified, it is much more difficult to make a case for their use for the chemoprevention of adenomas and CRC in average-risk individuals or even in individuals with a history of polyps. In light of the cardiovascular and gastrointestinal toxicity of these agents when used in a multiyear setting, the demonstration of the chemopreventive efficacy may be a “pyrrhic victory” as stated by Lynch
(85) in his editorial on the APPROVe trial. Furthermore, considering the newly identified risks for cardiovascular events associated with these agents, the cost-effectiveness of a chemopreventive strategy for CRC needs to be fully evaluated, particularly because a screening strategy alone appears to be effective
(4). In a simplified risk–benefit analysis, assuming that CRC incidence can be reduced by 50% with COX-2 inhibitor use, Psaty and Potter
(86) suggested that significantly more cardiovascular events would occur than cases of CRC prevented. However, the balance of benefits and risks in high-risk patients, such as those with familial adenomatous polyposis and nonpolyposis syndromes and those with a history of CRC, may be quite different from that detailed here for average- to higher-risk individuals. A role for COX-2 inhibitors continues to be evaluated in the setting of these high-risk patients
(85).
Although ASA seems to be an attractive candidate for CRC chemoprophylaxis, the apparent need for doses higher than that used for cardiovascular protection represents a crucial drawback
(48). Likewise, the improved gastrointestinal safety profile of COX-2 inhibitors over non-ASA NSAIDs made COX-2 inhibitors an attractive candidate until their cardiovascular toxicity came to light. Nonnaproxen, non-ASA NSAIDs seem to be the least attractive option because they are associated with both gastrointestinal and cardiovascular toxicity.
In conclusion, non-ASA NSAIDs seem to be effective at reducing the incidence of colorectal adenomas and CRC in observational studies. Good-quality RCT data suggest that COX-2 inhibitors are effective at reducing the incidence of colorectal adenomas in patients with previous adenomas. However, positive data on the reduction of death is lacking for both non-ASA NSAIDs and COX-2 inhibitors.
No quantitative data exist on the risk for gastrointestinal or cardiovascular harms associated with daily, multiyear use of non-ASA NSAIDs. Available data on COX-2 inhibitors suggest that absolute risk increases of over 1% for cardiovascular events and for clinically important gastrointestinal complications can be anticipated after only 2 to 3 years of use, and higher risks may accrue over longer periods. Furthermore, the cost-effectiveness of chemoprevention needs to be considered carefully and compared with other strategies, such as colorectal cancer screening alone. Therefore, the balance of benefits and risks does not appear to favor chemoprevention with non-ASA NSAIDs or COX-2 inhibitors in average-risk individuals or in those with a history of colorectal adenomas.
Aspirin and NSAIDs for the primary prevention of colorectal cancer - weighting the evidence
The authors of the review papers on aspirin (1) and nonsteroidal anti -inflammatory drugs (NSAIDs) (2) for the primary prevention of colorectal cancer (CRC) ignore important limitations of observational studies, raise concerns unlikely to be valid (1), and fail to include the analysis on NSAIDs from the Physicians' Health Study (PHS) that was published in the time period covered (3). The authors correctly discuss shortcomings of randomized trials showing no protection in men and women, i.e. low dose and short duration (1). They fail, however, to address shortcomings of observational studies on regular long-term drug use, i.e. unmeasured confounding, as an alternative explanation for the reduced risk of CRC observed in most of these studies.
Specifically, regular long-term drug use is associated with difficult to measure healthy characteristics. These healthy characteristics of long- term adherers may be associated with reduced risks independent of drug effects as evinced by reduced risks for many adverse outcomes in adherers to placebo. This can lead to paradoxical relations in observational studies (4). To reduce the magnitude of this problem, we excluded regular users of aspirin and NSAIDs from the PHS, thus studying only new regular users, and observed no reduction of CRC risk with aspirin (5) and NSAIDs (3).
Unfortunately, the referenced detailed assessment of the quality of each of the studies (1,2) was not available from AHRQ or the authors. However, the authors mentioned concerns about our aspirin analysis in the PHS because of contamination by intervention and a reduced standardized mortality ratio for CRC compared with the U.S. population (1). Randomized aspirin treatment does not threaten the validity of our post-trial study of self-selected aspirin use. Any carry-over of putative aspirin effects would bias the results towards observing a reduced CRC risk in post-trial aspirin users. Reduced standardized morbidity ratios (5) are ubiquitous in volunteer studies and generally do not bias measures of relative risk.
Finally, since only three cohort studies of NSAIDs and CRC are included in their analysis (2), the failure to include our observational study on NSAIDs and CRC in the PHS (3) largely reduces the value of the summary estimates. Taken together, we believe that the authors overstate the overall benefits from aspirin (1) and NSAIDs (2) on CRC. Putting greater weight on randomized evidence and observational studies with new- user designs would tip the harm to benefit balance even more towards harm for individuals at average risk for CRC.
1. Dube C, Rostom A, Lewin G, Tsertsvadze A, Barrowman N, Code C, Sampson M, Moher D. The use of aspirin for primary prevention of colorectal cancer: a systematic review prepared for the u.s. Preventive services task force. Ann Intern Med 2007;146:365-75.
2. Rostom A, Dube C, Lewin G, Tsertsvadze A, Barrowman N, Code C, Sampson M, Moher D. Nonsteroidal Anti-inflammatory Drugs and Cyclooxygenase-2 Inhibitors for Primary Prevention of Colorectal Cancer: A Systematic Review Prepared for the U.S. Preventive Services Task Force. Ann Intern Med 2007;146:376-89.
3. Stürmer T, Buring JE, Lee IM, Kurth T, Gaziano JM, Glynn RJ. Colorectal cancer after start of nonsteroidal anti-inflammatory drug use. Am J Med 2006;119:494-502.
4. Glynn RJ, Knight EL, Levin R, Avorn J. Paradoxical relations of drug treatment with mortality in older persons. Epidemiology 2001;12:682- 9.
5. Stürmer T, Glynn RJ, Lee IM, Manson JE, Buring JE, Hennekens CH. Aspirin use and colorectal cancer: post-trial follow-up data from the Physicians' Health Study. Ann Intern Med 1998;128:713-20.
Conflict of Interest:
None declared