Nonsteroidal Anti-inflammatory Drug Use After 3 Years of Aspirin Use and Colorectal Adenoma Risk: Observational Follow-up of a Randomized Study

Materials and Methods

Results

Characteristics of the subjects in the three aspirin treatment arms are summarized in Table 1. There were no statistically significant differences among the three arms with regard to demographic and lifestyle factors at AFPPS study entry, including age, sex, race, body mass index, first-degree family history of colorectal cancer, alcohol use, cigarette smoking status, and the lifetime number of adenomas before randomization. The three treatment groups were also similar with regard to the percentage of subjects who were randomly assigned to folate treatment, the duration of aspirin treatment, or the use of nonprotocol NSAIDs during the randomized treatment period. However, as suggested by previously published findings (1), subjects who were randomly assigned to receive low-dose aspirin were less likely than subjects in the other two treatment arms to have an adenoma detected during the 3-year treatment period (P = .080). Posttreatment follow-up characteristics were also comparable across the three randomized aspirin groups. Of the 850 subjects included in this analysis, 764 (90%) had one colonoscopy and 86 (10%) had two or more colonoscopies during the observational follow-up after the end of aspirin treatment. The average time from the end of the study treatment to the first posttreatment endoscopy was 45.8 months (SD = 14.4 months). During this time, the median number of study questionnaires completed by subjects was 9 (interquartile range = 4–12; 17 subjects did not complete any questionnaires). The response rate for the study questionnaires (the number completed as a proportion of the number expected) was approximately 85%. The most commonly used NSAIDs were the same in the three treatment groups: aspirin and aspirin-based drugs (used by 59% of subjects), ibuprofen (used by 46%), naproxen (used by 20%), rofecoxib (used by 8%), and celecoxib (used by 8%).

In general, subjects who had a posttreatment colonoscopy were similar to the 138 subjects who did not have a posttreatment colonoscopy with regard to demographic and lifestyle factors. However, subjects with a posttreatment colonoscopy were more likely to have had adenomas during the treatment phase of the study (47% vs 22%, P < .001) and advanced lesions (11% vs 5%, P = .03) detected during the 3-year aspirin treatment period.

Among the 833 subjects who completed posttreatment questionnaires, 589 (71%) took NSAIDs at most intermittently and were classified as sporadic users (average use <2 days per week), 96 (12%) were moderate users (average use 2 to <4 days per week), and 148 (18%) were frequent users (average use ≥4 days per week); 165 subjects (20%) reported no NSAID use at all (Table 1). Frequent NSAID users were similar to subjects with less frequent NSAID use with regard to most demographic, lifestyle, and clinical factors (data not shown). However, frequent NSAID users were statistically significantly older (P = .01) and had a higher body mass index (P = .03) than the sporadic or moderate NSAID users.

The effects of aspirin use on adenoma risk during the treatment phase of the trial are summarized in Table 2. As reported previously (1), aspirin treatment moderately reduced the risk of colorectal adenomas; the adjusted relative risk for all adenomas for the two aspirin groups combined vs the placebo group was 0.90 (95% confidence interval [CI] = 0.77 to 1.05). When we considered each aspirin group independently, a borderline statistically significant risk reduction was observed for the low-dose aspirin group (adjusted relative risk [RR] = 0.84, 95% CI = 0.70 to 1.00).

The protective effect of randomization to one of the aspirin groups persisted during the posttreatment period for subjects who continued to use aspirin or other NSAIDs, and there was a statistically significant trend of decreasing risk with increasing frequency of later NSAID use (P_trend = .02) (Table 2). However, subjects randomly assigned to either active aspirin arm who used NSAIDs only sporadically during observational follow-up were no longer at a reduced risk of adenomas compared with placebo subjects who also were sporadic NSAID users (adjusted RR = 1.00, 95% CI = 0.80 to 1.24), nor were they at increased risk. There was no association between NSAID use during the posttreatment period and the risk of adenomas among subjects who had been randomly assigned to placebo (Table 2). We obtained similar results when we evaluated the risk of small tubular adenomas (data not shown).

The protective effect of low-dose aspirin against colorectal adenomas during the treatment phase persisted only among subjects who continued to take NSAIDs during the observational follow-up phase (P_trend with NSAID use = .03) (Table 2). Subjects randomly assigned to 81 mg of aspirin who later used NSAIDs at least 4 days per week had a markedly reduced risk of all adenomas compared with placebo subjects who were at most sporadic NSAID users (39.9% vs 26.8%; adjusted RR = 0.62, 95% CI = 0.39 to 0.98) (Table 2). The reduction in unadjusted absolute risk was 13.1 percentage points (95% CI = −0.3 to 26.5 percentage points) (P = .07). We obtained similar results when we restricted the reference category to placebo subjects who took no NSAIDs after the end of randomized treatment (data not shown).

High-dose aspirin did not confer a statistically significant reduction in adenoma risk during the randomized treatment period of the trial (Table 2). In this analysis, subjects who were randomly assigned to high-dose aspirin and who later were frequent users of NSAIDs had a nonstatistically significantly lower risk for all adenomas than placebo subjects who were at most sporadic NSAID users (30.6% vs 39.9%; adjusted RR = 0.72, 95% CI = 0.46 to 1.12; P_trend with NSAID use = .08). The corresponding reduction in unadjusted absolute risk was 9.3 percentage points (95% CI = −5.3 to 23.8 percentage points) (P = .23).

The association between randomized aspirin treatment with later NSAID use and the risk of colorectal adenoma in the posttreatment observational period was not statistically significantly modified by randomized folate treatment assignment (P = .89). Stratification by adenoma occurrence during the randomized aspirin treatment period of the trial did not modify the posttreatment results statistically significantly either. Among subjects who were diagnosed with at least one adenoma during study aspirin treatment, randomization to low-dose aspirin with later frequent NSAID use was associated with a 50% reduction in the risk of adenomas (adjusted RR vs placebo subjects with sporadic NSAID use = 0.50, 95% CI = 0.25 to 0.99). Among those without adenomas during the randomized treatment period, the reduction in risk was 22% (adjusted RR vs placebo subjects with sporadic NSAID use = 0.78, 95% CI = 0.41 to 1.47). The multiplicative interaction between the frequency of NSAID use during observational follow-up and the detection of adenomas during the treatment period in the low-dose aspirin group was not statistically significant (P = .60). The results for high-dose aspirin were similar to those for low-dose aspirin (data not shown).

Findings regarding the risk of advanced lesions during the randomized treatment phase are summarized in Table 3. During the randomized treatment period, aspirin (81 or 325 mg) conferred a protective effect against advanced lesions compared with placebo (RR = 0.73, 95% CI = 0.50 to 1.09), and the effect was more pronounced, although not statistically significant, in the low-dose aspirin group (adjusted RR = 0.66, 95% CI = 0.41 to 1.07). A comparable analysis for the posttreatment follow-up period was hampered by the small number of advanced lesions detected. Therefore, we conducted this analysis by combining the moderate and frequent NSAID users into a single category of NSAID users (Table 3). None of the relative risks was statistically significant. However, the lower limit of the confidence limits for each exposure group included reductions in risk as large as 50% in comparison with the reference category, and each point estimate suggested reduced risks, except for subjects randomized to high-dose aspirin who subsequently were NSAID users.

All analyses were repeated with aspirin and nonaspirin NSAIDs considered separately. The findings were similar to those shown above (data not shown).

Discussion

In this analysis of adenoma occurrence after the end of randomized aspirin treatment, we found that continued use of aspirin and/or other NSAIDs extended the chemopreventive effects of aspirin that were seen during the treatment period. There was an apparent strengthening of the chemopreventive effect associated with NSAID use of at least 4 days per week on average. There was no evidence of tachyphylaxis or tolerance, nor was there an indication of a rebound in the incidence of adenomas after the end of randomized treatment. Although our numbers for advanced lesions were too small to draw confident conclusions, low-dose aspirin and subsequent NSAID use also seemed to yield a persistent reduction in risk. These findings imply that aspirin does not lose its antineoplastic properties over time and that these effects persist and may even be accentuated with prolonged NSAID use.

During the randomized treatment phase of the study, low-dose aspirin had a chemoprotective effect against adenomas, whereas high-dose aspirin did not (1). This finding was unexpected. In this secondary analysis, both doses of aspirin were associated with a decreased risk for all adenomas among subjects who frequently used NSAIDs after the end of randomized aspirin treatment, although the risk reduction was not statistically significant among subjects who had been randomly assigned to high-dose aspirin. These results seem to support the hypothesis that chance was responsible for the dose–response pattern observed in the analysis of adenoma occurrence during randomized aspirin treatment (1). For advanced lesions, small numbers of endpoints hampered the analysis, and no firm conclusions can be drawn regarding the delayed effects of aspirin treatment.

A recent observational study reported that the use of two standard tablets of aspirin per week (equivalent to a dose of 650 mg of aspirin) was associated with a statistically significant reduction in the risk of colorectal cancer, with a trend of decreasing risk with increasing number of tablets taken per week (9). A more recent analysis of data from two large randomized trials with aspirin concluded that a daily dose of 300 mg or more of aspirin taken for 1–7 years could prevent colorectal cancer after a latency period of about 10 years (10). Our observational follow-up data suggest a trend of reduced risk of colorectal adenomas with increased frequency of NSAID use after 3 years of daily aspirin use. The implications of this secondary prevention of adenomas on the incidence of colorectal cancer are not clear, but our findings are consistent with the two studies mentioned above, given that most cancers in the colon and rectum evolve from the same types of precancerous neoplastic lesions that we studied (11). It is reasonable to argue that by reducing the incidence of colorectal adenomas, long-term use of NSAIDs could reduce the future risk of colorectal cancer.

In this analysis, we did not find an increased risk of adenomas after the end of randomized treatment. Indeed, for advanced adenomas, the chemopreventive potency of aspirin seemed to persist even in the absence of later NSAID use, suggesting a carryover effect of the aspirin treatment, at least for a few years. Such an effect might occur if, for example, aspirin has a longer latent period of effect for advanced adenomas than for small tubular adenomas. However, as mentioned above, the numbers of advanced adenomas are too small and our statistical power is too limited to rule out chance as a possible explanation for this result.

It is unclear why we found no reduction of adenoma risk associated with later NSAID use among subjects who had been randomly assigned to placebo. Previous studies have described inverse associations after relatively short exposures of 1–3 years (3,12). In this analysis, it is possible that longer and/or more intense use of these drugs would be required to see such an effect in placebo subjects because of the 3 years of NSAID abstinence during the randomized treatment period of our study.

There is extensive evidence supporting the antineoplastic properties of aspirin and other NSAIDs (2–4,13–15). Regarding colorectal adenomas and colorectal cancer, experimental studies (16–21) have reported a protective effect of aspirin. More importantly, results of observational studies (5,22–27) and randomized clinical trials (1,10,12,28,29) have consistently supported the inverse association of aspirin with adenoma risk. The proposed mechanisms underlying the anticarcinogenic activity of aspirin and other NSAIDs may be related to their ability to inhibit the activity of COX enzymes. These enzymes are responsible for the synthesis of prostaglandins and other metabolites that are associated with inflammatory processes and carcinogenesis (19,30–35). Some experimental studies (32,36–42) have described anticarcinogenic mechanisms independent of COX enzymes. Indeed, aspirin in the doses we used has relatively little effect on the activity of COX-2 (43), the enzyme isoform that has been most frequently associated with neoplastic processes, a finding that supports the idea that COX-2 inhibition may not be necessary for the chemopreventive effects of NSAIDs.

Our study has several strengths. Compliance with the study regimen was excellent: More than 85% of subjects reported taking study pills at least 90% of the time, even during the year before the year 3 colonoscopy that marked the end of the randomized aspirin treatment period (1). We had a large sample size: Of the 1121 subjects originally randomly assigned to placebo or aspirin, 1007 agreed to continue in the follow-up phase and 850 had at least one colonoscopy. Furthermore, all biopsy specimens were examined in a blinded fashion by a single study pathologist in a uniform endpoint review (1,8).

This study also has a number of limitations. Most importantly, because in this analysis, the posttreatment follow-up period was observational, our findings are susceptible to bias and confounding. For example, posttreatment NSAID use was based on self-report and thus, our categorization of subjects according to reported use may not have been entirely accurate. In addition, frequent NSAID users were older than the other subjects and tended to have a higher body mass index. Although our estimates were adjusted for these factors, we could not deal similarly with other potentially confounding factors that we did not identify. Subjects in this analysis were a higher risk subset of an already selected population that met the eligibility requirements for the randomized trial because the participants with neoplastic lesions detected during the treatment period were more likely to have a colonoscopy during the posttreatment phase than those who were free of adenomas. Thus, our study subjects are not representative of the general population. Furthermore, small numbers of advanced lesions limited our statistical power and prevented us from drawing firm conclusions regarding the association of NSAIDs with this endpoint.

To evaluate response trends, we relied on the reported frequency of use because our questionnaires did not capture dose information, and this is another important weakness of our analysis. Furthermore, there was no uniform interval between the end of the randomized study and the first posttreatment colonoscopy. However, our statistical models were adjusted for this varying time at risk, and it is unlikely that it affected our results. It is possible that the particular characteristics of the NSAIDs taken after the end of randomized treatment could have played a role in our findings. For example, specific types of NSAIDs and/or higher doses could have been responsible for the apparent strengthening of the chemopreventive effect of aspirin during the subsequent follow-up interval.

In summary, our study provides further evidence of the potential chemopreventive effect of NSAIDs against colorectal adenomas in patients with a history of these polyps. Our findings suggest that prolonged and frequent use of these agents can have strong beneficial effects. We did not find either loss of effectiveness with continued use or increased risk following discontinuation of use. This study raises the possibility that NSAIDs taken 4–7 days per week might be sufficient to prevent adenomas and, by implication, colorectal cancer. Because our results imply that the strength of the inverse association between NSAID use and adenoma risk may increase with frequency of use and because more frequent use is also likely to cause more adverse events, further investigation is important to understand ways to maximize the possible benefits of these drugs while minimizing their potential harms.

Funding

This work was supported in part by funding from the National Cancer Institute, National Institutes of Health (CA 046927 and CA 098286) to Dartmouth College (J.A.B. was the principal investigator).