Against this background, the effects of HMG-CoA reductase inhibitors (statins) are particularly pertinent. On one hand they could be expected to be commonly prescribed to patients at risk of colon cancer, hence any effects on increasing CRC risk may be clinically important and on the other, there are data suggesting a potentially important protective effect of statins against CRC [8-12]. Laboratory studies have shown that statins induce apoptosis, inhibit proliferation and reduce invasion in multiple colon cancer cell lines [13,14]. This effect seems to be a class effect and has been documented with lovastatin, simvastatin, pravastatin and atorvastatin and is most plausibly due to the inhibition of the mevalonate synthetic pathway reducing the cellular availability of substrates required for the isoprenylation of small signaling G-proteins and the resultant inhibition of pro-carcinogenic and pro-survival pathways [13-15]. Other protective mechanisms have been suggested, including anti-inflammatory actions, reduction in circulating lipids and a beneficial effect on adipokine secretion profiles [16].

Clinical studies with statins and CRC have given variable results. Two separate studies showed a highly significant 50% reduction in CRC incidence in statin users [8,11]. A more recent observational study showed a 38% reduction in CRC incidence only in lovastatin users with a non-significant trend for greater protection with more than 3years therapy [17] and two separate meta-analyses suggested a protective effect of statins [9,18] Bardou at al confirmed that randomized controlled trials tended to show a small a non-significant-reduction in CRC incidence in statin users, whilst observational and case–control studies suggested a persistent but modest protective effect of statins [18]. These differences are thought to arise from the low absolute incidence of CRC in the randomized trials and relatively limited short term nature of statin trials, which were not primarily designed to examine cancer incidence. In addition to lovastatin protective effects have been reported with simvastatin and pravastatin [8,19]. However other studies, often with different methodologies have shown no effects [19-25]. The reasons for these discrepancies are not clear but probably involve differences in the study design, populations, specific statin use in different populations, age at initiation of statin use and the controls examined, duration of statin exposure and exposure to confounding factors. However, many of the recent observational studies had significant limitations with statin use being measured from prescriptions and an uninvestigated control group recruited from general practice databases. In several of the studies statin exposure was regarded as positive if as little as one prescription or 3months therapy was taken [24,25]. The maximal duration of follow up in cohort studies was usually less than 5years and this may not have been sufficient time for the effects of statins to become apparent as other studies with over 5years, or a mean of 9years follow up showed more protective effects [8,10]. A recent longer-term study showed that > years statin exposure, assessed using a questionnaire, was not associated with a reduced incidence of colon cancer [26]. Statins and other medications purchased over the counter were not included in many of the studies and neither BMI nor smoking data were available in all studies. The controls were not investigated with endoscopy so there was no certainty over these diagnoses and no information available to assess early CRC. One major review confirmed that the data on statins and colon cancer were conflicting and inconsistent and although a number of studies have provided no evidence of protection, the situation was sufficiently unclear that further clinical studies were warranted [19].

There is biological plausibility to our data: we detected both time- and dose-dependent effects against CRC development and experimental studies are supportive of CRC-protective effects of statins. In addition to in vitro cell line studies, statins have been shown to reduce polyp formation in Min^−/− mouse models [29] and pre-neoplastic and neoplastic lesions in animal models [30,31]. A single study has shown that statin use was associated with a reduced rate of post-polypectomy adenomatous polyp formation [32], although another study with shorter follow up failed to detect any protective effect of statins against polyp recurrence [33].

More recent studies have suggested that biological differences may underlie some of the variability in the apparent effects of statins. Upregulation of HMG-CoA has been reported in CRCs but one study has shown this to be more marked in left sided cancers [15]. Similarly HMG-CoA genotype seems to influence the protective effect of statins: a higher activity allele was associated with a protective effect, whereas a significant protective effects was not associated with a lower activity allele [11]. The polyp-cancer sequence is estimated to take 10–15years and it is not clear, where along this sequence the effect of statins may be most noticeable [33]. Experimental studies have shown that statins may become less effective in cell lines as more genetic and hence functional changes downstream of the putative cellular effect of statins accumulate [34,35]. If, for example, the effect of stains is more prominent at stages of polyp-development or progression, then the average age of initiation of statin therapy in any population is likely to contribute to differing effects of statins on CRC incidence between studies.

The main specific strengths of our study are the comprehensive drug history available and the fact that all patients underwent diagnostic colonoscopy. Many previous studies have used prescribing records and although these studies have many advantages, they may misreport actual drug exposure as not all prescribed medication is actually consumed. In addition in the United Kingdom aspirin, ibuprofen, diclofenac and simvastatin are available to purchase without prescription, thus we feel our records of drug exposure are both accurate and comprehensive. Asymptomatic colon cancer is not uncommon and our use of a control group, who all underwent colonoscopy will have minimized bias due to misdiagnosis which is a problem in uninvestigated cohorts. There are data from the USA showing that patients with co-morbidity undergo more frequent screening colonoscopies [36]. We only included cases and controls having their first ever diagnostic colonoscopy, so removing any potential bias from repeated colonoscopy or previous polyp removal. A potential disadvantage of our design studying symptomatic patients attending for diagnostic colonoscopy is that the control group may exhibit more health-seeking behavior and be more health conscious and hence introduce bias. However all had genuine symptoms for which colonoscopy would be indicated and as all were provided under the United Kingdom National Health Service, no direct financial or insurance incentives will have influenced decisions regarding colonoscopy. An ideal study would examine statin use in cases and controls from asymptomatic patients invited for colonoscopic colon cancer screening, rather than symptomatic patients; however this would not be possible as the United Kingdom national screening programme is based on initial faecal occult blood screening, followed by colonoscopy only in those testing positive. Interestingly the association of reduced CRC incidence in statin uses was also seen in the sub-group of patients that presented via the national screening program with positive faecal occult bloods, although the overall number of cases in this group was too small to draw definite conclusions (statin use: controls 14/26, 53%, cases 3/13, 23% OR 0.37 (0.06 – 1.17)).

The issue of bias inherent to the control group affecting the results must always be considered in case–control and cohort studies, in particular the possibility that statin users are generally more health-conscious and this explains their lower CRC risk. As with previous studies in this field, we cannot exclude this completely but we feel this is an unlikely explanation of the findings. It is possible that statin-users may have had more specific dietetic input and this lead to a positive change in diet which influenced cancer risk. The majority of statin prescribing was by the patients’ primary care physician for the primary prevention of vascular disease and as such was not governed by any specific protocol and both patient and clinician perception are likely to have influence the use of statins. The control group is comparable to a standard low-moderate CRC risk cohort undergoing diagnostic colonoscopy for any indication in the UK and comparable to the age-matched population as a whole. In terms of smoking, alcohol intake, BMI and type 2 diabetes as markers of health-related behaviors, the control group did not appear to be especially healthier than the cases and the effect of statins persisted after correction for all known confounding variables. The risk of colonic neoplasia is probably increased in both diabetes mellitus and established coronary artery disease [6,37], and this would have been expected to reduce any effects of statins observed in our study. Chronic statin use is associated with a very low incidence of gastrointestinal side effects and so it seems unlikely that statins are over-represented in the control group because they were responsible for symptoms initiating the referral for colonoscopy [38]. It could be argued that because of the proclivity for aspirin to cause gastrointestinal bleeding and anaemia, this could have led to over-representation of aspirin in the control group and indeed there was a slightly higher number of controls in which anaemia was the presenting condition. Whilst this is possible, our study showed a protective effect of aspirin entirely consistent with other studies that used different control groups [25], and the effect of statins persisted after correction for aspirin use. We also showed an apparent protective effect of metformin against CRC, again consistent with previous studies, suggesting our results have external validity [39].

This study did have limitations, some of which were inherent with the case control study design. Despite this study being sufficiently powered to produce statistically significant results, this was a relatively small cohort compared to some similar case control studies. Data collection was cross referenced using patient interviews and clinical notes but recall bias may have caused inaccuracy and difficulty in determining the extent of exposure to medications, although as most subjects were interviewed in the same situation pre-colonoscopy, this is unlikely to unduly influence cases or controls. Data were collected for many potential confounding factors that had been identified from the literature but other uncontrolled confounders could have affected the results. We did not attempt to address the issues regarding diet and cancer risk and further studies in the area would benefit from including this. Although we utilized both pre- and post-cancer diagnosis interviews to establish statin exposure, we do not feel this introduced any bias of recall. All drugs exposures were accurately cross-referenced to medical records (including pre-colonoscopy medical notes), statins are characteristically long-term therapies and once started are very rarely stopped and although patients were informed we were interested in risk factors for their diseases, they were unaware of which particular factors. Our results show no difference in statin use between those interviewed pre- and post-colonoscopy. However the issue of residual confounding by other unrecognized factors must still be considered. Similarly, although adequately powered for our pre-designed end-point, the relatively small size of our study may have contributed to the relatively large protective association that we have described.