Commentary: Aspirin and colorectal cancer an epidemiological success story

Colorectal cancer is the second most common cancer in developed countries, with a lifetime risk of 5%, and about one million new cases worldwide each year.1 Treatment is often ineffective and population screening by regular colonoscopy is expensive.1 Prevention is therefore the ideal. Although there has been longstanding interest in the possibility that aspirin might reduce the risk of colorectal cancer, this effect has only recently been proven by long-term follow up of two large randomized trials of aspirin from the late 1970s and early '80s.2,,3 Randomization to ≥300 mg aspirin daily prevented up to 75% of colorectal cancers after a latency of approximately 10 years.4 There is still uncertainty about the effect of lower and less frequent doses of aspirin and further research is required, but having established a causal link between aspirin use and a reduced risk of colorectal cancer, it is timely to review the origins of the hypothesis and the important role played by observational epidemiological studies. There was evidence from cell culture studies and animal studies in the 1960s and 1970s to suggest that aspirin might reduce the rate of tumour growth and metastasis in general.5–7 The potential for benefit from adjuvant aspirin in patients with colorectal cancer was studied in a small randomized controlled trial in the late 1970s and early 1980s, but no effect was seen.8 There were also concerns around this time that aspirin might increase carcinogenesis, particularly in the stomach and kidney. However, very few, if any, useful data were available on the effects of aspirin on risk of cancer from epidemiological studies in man. In 1988, Kune and colleagues9 published the first report of an inverse association between the use of aspirin and the risk of colorectal cancer from the Melbourne Colorectal Cancer Study, a population based case–control study. The 1988 paper investigated associations between colorectal cancer and a variety of chronic illnesses, operations and medications in 715 patients with incident, histologically confirmed colorectal adenoma-carcinoma and 727 age- and sex-matched controls. The stated aim of the study was ‘to examine some current hypothesis of colorectal cancer risk, partly to examine previously described associations between colorectal cancer and other cancers and partly as an exploratory step to stimulate the creation of new hypotheses of colorectal cancer aetiology’. The authors did not mention any specific a priori hypotheses about any effect of aspirin on colorectal cancer, but data were collected on medication exposure including aspirin and other non-steroidal anti-inflammatory drugs (NSAIDs), including frequency (daily, weekly, don't know) and duration of use. Kune and colleagues9 found that patients with colorectal cancer were less likely than controls to have used aspirin-containing medications in the past [relative risk 0.53, 95% confidence interval (CI) 0.40–0.71]. This association remained significant after adjustment for comorbidities and was consistent for both colon and rectal cancers and for men and women. A similar association was found for use of other NSAIDs but was confined to colon cancers only. The authors examined at least 37 different potential associations with colorectal cancer and reported a number of other significant associations in addition to aspirin use. Hypertension, heart disease, stroke, chronic chest disease, chronic arthritis and use of vitamin supplements were less common amongst patients with colorectal cancer, whereas haemorrhoids and large bowel polyps were more common amongst cases. Prior aspirin use remained inversely associated with colorectal cancer in a multivariate model that included all of the statistically significant univariate associations as well as previously identified dietary factors,10 whereas the associations for stroke and chronic chest disease were greatly reduced. The authors concluded that ‘this finding whatever the mechanism may be, has potential significance in colorectal cancer chemoprevention and merits early confirmation’. Further analyses stratified by aspirin dose, length of exposure, regularity and frequency of use would have been useful, but were not published in the 1988 paper, or in any subsequent publication from the Melbourne study. Nevertheless, the study stimulated considerable interest among researchers in aspirin as an anti-cancer agent and numerous other epidemiological studies were published over the subsequent two decades. Data on the association between aspirin use and colorectal cancer are now available from 11 cohort studies and 19 case–control studies.4 There are several lessons that can be drawn from the Melbourne Colorectal Cancer Study and the subsequent two decades of work on the link between use of aspirin and the risk of colorectal cancer. The first lesson is the need for long-term follow up of cohort studies and randomized controlled trials if all of the important benefits and harms of treatments are to be determined. Most colorectal cancers develop from adenomas—the so-called adenoma-carcinoma sequence.1 Several small randomized trials have shown that aspirin11–13 and cyclo-oxygenase-2 enzyme (COX-2) inhibitors14–16 reduce the recurrence of adenomas by up to 40% in patients with previous adenomas or colorectal cancer, but with only 2–3 years follow up, these trials were unable to determine any effect on colorectal cancer. Moreover, a reduction in risk of cancer could not simply be assumed to be an inevitable consequence of the effect of aspirin on recurrence of adenomas. The likelihood of malignant transformation of adenomas that develop despite aspirin or COX-2 inhibitors versus those that are prevented was uncertain, but was important because although up to 40% of people in developed countries have one or more colorectal adenomas by age 60 years, 10% of these adenomas progress to cancer.17 Moreover, it could not be assumed that secondary prevention of adenomas by short-term treatment with aspirin or COX-2 inhibitors would be maintained on long-term treatment, nor that the same effect would necessarily be seen in primary prevention. Two large randomized trials of aspirin in primary prevention of vascular disease showed no effect on colorectal cancer during 10 years of follow up,18,,19 and the results of short-term follow up in cohort studies were similarly unpromising.4 However, a latency of more than 10 years would be expected given that the delay between the initiation of development of an adenoma, the point at which aspirin is believed to act, and presentation of colorectal cancer is estimated to be 10–15 years.20,,21 Only after 10 years follow up in the UK-TIA Aspirin Trial and the British Doctors Aspirin Trial did the substantial reduction in rates of colorectal cancer appear.1 If aspirin or other interventions are effective in reducing the risk of other cancers then similar periods of follow up may be necessary to detect any signals. Even longer follow up might well be required to confirm or exclude effects on slow-growing tumours, such as prostate cancer. It is also quite possible that there are long-term benefits or harms of other widely used current treatments that are unrecognized because of short-term trials and a lack of post-trial follow up. Yet, although post-trial follow up for certain registered diseases and for deaths is possible in many countries by linkage with national registries, over-regulation by ethics and data protection authorities often makes such research difficult or even impossible. The second lesson is simply a reminder that observational epidemiological studies remain a very important way of identifying causal associations between risk factors and disease. Observational studies can, of course, reliably identify powerful causal associations (e.g. smoking and lung cancer, cholesterol and coronary heart disease, blood pressure and stroke, radiation exposure and cancer, sleeping position and sudden infant death and male circumcision and incidence of HIV infection). However, there has been much discussion about the difficulties in reliably determining moderate effects, particularly for ‘behavioural’ risk factors that involve an element of choice, such as diet or use of vitamins, hormone replacement therapy or aspirin. Yet, such ‘behavioural’ risk factors have been reliably identified in the past (e.g. folate supplementation in pregnancy and risk of neural tube defects, hormone replacement therapy and breast cancer) and the unreliability of some observations has stemmed at least partly from neglect of appropriate methodology.22–24 The importance of rigorous methodology in the reliable detection of causal associations in observational studies was illustrated by our recent systematic review of case–control studies of aspirin use and colorectal cancer, including the Melbourne Colorectal Cancer Study. Overall, there was evidence of significantly lower use of aspirin or NSAID in cases than in controls (pooled OR = 0.80, 95% CI 0.73–0.87, P < 0.0001, Figure 1), but with substantial heterogeneity between studies (P < 0.0001).4 Associations tended to be much stronger in smaller studies, with a highly significantly asymmetrical funnel plot (Figure 1), which could be misinterpreted as evidence of publication bias and overestimation of any true effect. However, on closer scrutiny the asymmetrical funnel plot appeared to be due to more discriminating definitions of use of aspirin or NSAID in smaller studies, with a strong inverse relation (weighted regression: r2 = 0.53, P = 0.0005) between the percentage of the control group defined as users and the relative use of aspirin in cases vs controls.4 Any use of aspirin and/or NSAID in cases of colorectal cancer vs age-and sex-matched controls in 19 case–control studies. The Melbourne Colorectal Cancer Study is highlighted in bold font.9 (Data derived from Flossmann and Rothwell4) Fourteen studies stratified analyses by the extent of use of aspirin or NSAID, and when the analysis of all 19 studies was based on the maximum use reported (most regular and/or longest duration) the association with colorectal cancer was much stronger and less heterogeneous (Figure 2), and was no longer related to the percentage of controls defined as users (r2 = 0.10, P = 0.18). In eight studies where it was possible to look specifically at irregular or occasional use of aspirin or NSAID, there was no association with colorectal cancer (OR = 1.01, 0.93–1.09, P = 0.87, heterogeneity—P = 0.05). Those studies that stratified analyses by both regularity of use and duration of use reported 50–70% reductions in relative risk of colorectal cancer associated with use of medium to high dose aspirin for over 10 years.4 The potential advantages of smaller more rigorous observational studies over larger simpler studies have been outlined previously.22 Maximum use of aspirin and/or NSAID in cases of colorectal cancer vs age-and sex-matched controls in 19 case–control studies. The Melbourne Colorectal Cancer Study is highlighted in bold font.9 (Data derived from Flossmann and Rothwell4) Having obtained randomized evidence that regular aspirin does indeed prevent a substantial proportion of colorectal cancers, and having shown that the observational studies were consistent with the randomized studies, we were able to use data from previous observational studies to address clinically important outstanding questions. The observational data convincingly demonstrated no difference in effect of aspirin and other NSAIDs, no difference in relation to age, sex, race or family history, no difference in relation to the site or aggressiveness of the cancer and no fall off in apparent effect with use for ≥20 years. However, a consistent association was only seen with use of ≥300 mg aspirin daily, with diminished and inconsistent results for lower or less frequent doses. The power of the observational data to explore these important issues was considerable given that the cohort studies included 1 136 110 individuals with over 6000 colorectal cancers during follow up and the 19 case–control studies included 20 815 cases of colorectal cancer.4 In contrast, the pooled randomized comparison was based on a total of only 215 colorectal cancers during follow up.4 In conclusion, the first report in 1988 of an inverse association between the use of aspirin and the risk of colorectal cancer from the Melbourne Colorectal Cancer Study has proved to be reliable and highly influential. The somewhat ‘scattergun’ approach of collecting data on many different risk factors and exposures was intellectually valid and experience has shown that this approach is often more productive than focussed hypothesis driven studies of single ‘flavour of the month’ risk factors—a lesson that has some resonance with the current preference for genome scans over candidate gene studies. Finally, the aspirin and colorectal cancer story also shows that where basic laboratory observations raise the possibility of effects of specific exposures on the development of disease in tissue culture or animal models, the best approach to ‘translational research’ is often to jump straight to population-based observational epidemiology.