Systematic review: potential preventive effects of statins against oesophageal adenocarcinoma

Authors


  • This uncommissioned systematic review was subject to full peer-review.

Correspondence to:

Dr L. Alexandre, Norwich Medical School, University of East Anglia, Norwich, NR4 7TJ, UK.

E-mail: Leo.Alexandre@uea.ac.uk

Summary

Background

The incidence of oesophageal adenocarcinoma (OAC) has risen dramatically in recent decades, and its prognosis remains extremely poor. There is emerging evidence that statins may prevent OAC.

Aim

To systematically review both the experimental and epidemiological evidence to determine whether statins reduce the risk of developing OAC.

Methods

Relevant laboratory and epidemiological studies were identified by systematically searching the PUBMED and EMBASE electronic databases for data on statins and oesophageal cancer (OC). The evidence was assessed according to the nine Bradford Hill criteria (BHC) of causality. Pooled effect sizes (ES) were calculated for the risk of OC with prior statin use.

Results

Many of the BHC were supported including: ‘plausible biological mechanisms’, ‘coherence’, ‘strong associations’, ‘consistency’, ‘biological gradient’, ‘analogy’ and ‘temporality’. Three experimental studies reported that statins inhibited proliferation, induced apoptosis and may limit metastatic potential in OAC cell lines. Fixed effects meta-analysis of two prospective studies in Barrett's oesophagus cohorts, involving 1382 participants, showed an ES of 0.53 (95% CI = 0.36–0.78, P = 0.001, I2 = 0%) for risk of OAC with prior statin use. Meta-analysis of three prospective studies in general population cohorts, involving 35 214 participants, showed an ES of 0.86 (95% CI = 0.78–0.94, = 0.001, I2 = 0%) for risk of OC with prior statin use. The most important criterion, ‘experiment’, is as yet unfulfilled as to date there are no clinical trials which investigate this hypothesis.

Conclusion

There is some evidence that statins may protect against the development of OAC, although to be conclusive, data from randomised clinical trials are required.

Introduction

Oesophageal cancer (OC) is an important global healthcare problem and is associated with a poor prognosis. In 2008 worldwide there were approximately 482 000 new cases and 400 000 associated deaths,[1] making OC the fifth and eighth commonest cause of death from cancer in men and women respectively. The incidence of the two main histological subtypes, namely oesophageal adenocarcinoma (OAC) and oesophageal squamous cell carcinoma (OSCC), significantly differs. In recent decades, in western countries, the incidence of OAC has risen dramatically,[2-5] with at least a threefold increase in men and women since the 1970s in England and the US.[6, 7] Conversely, in many countries, the incidence of OSCC has declined or remained stable.[2-4, 7, 8]

Patients with OAC commonly present when the disease is at an advanced stage and palliative therapies are the only treatment option. However, of those suitable for potentially curative treatment, the outcomes are often still poor, with a recurrence rate post-oesophagectomy of 46% within 12 months[9] and a 5-year survival of only 23%.[10] Barrett's oesophagus (BO), the precursor lesion to OAC, is a histologically and endoscopically defined entity affecting the distal oesophagus, where the native stratified squamous epithelium is replaced by metaplastic columnar epithelium.[11] Unfortunately, screening and surveillance of BO has not led to a reduction in mortality from OAC.[12]

Associated risk factors for OAC include BO,[13] increasing age,[14] male gender,[14] Caucasian race,[3] gastro-oesophageal reflux disease,[14-16] obesity,[16, 17] tobacco[18] and specific medications which relax the lower oesophageal sphincter.[19-23] Conversely, the use of non-steroidal anti-inflammatory drugs (NSAIDs) and aspirin,[24] are inversely associated. Despite these known associations, the aetiology of OAC is incompletely understood and other factors may be involved. Statins are commonly used to lower serum cholesterol in the primary and secondary prevention of cardiovascular disease, and demonstrate a number of promising anticancer effects. The aim of this systematic review is to assess the current experimental and epidemiological evidence, according to the Bradford Hill criteria (BHC) of causality, as to whether or not statins may reduce the risk of developing OAC, including the carcinogenic progression from BO to OAC. Further work to clarify any existing findings and the need for new studies will be considered. Determining whether statins prevent carcinogenesis is important to establish if they should be included in aetiological models of OAC and if they may have a clinical application for cancer prevention.

Methods

Eligibility criteria, search strategy and data synthesis

Experimental in vitro studies examining the effects of statins on OAC or BO cell lines, and observational studies examining the association between statin use and the development of OC were selected. Observational studies of either cohort or case-control design were included that had cases with a primary diagnosis of OC, or OAC specifically, detailed prior exposure information for the use of statins, and participants were either selected from the general population or patients with known BO.

Two reviewers (LA and ARH) conducted the literature search and appraised the eligibility and validity of the studies. Full-text papers and abstracts written in English were identified from the PUBMED and EMBASE electronic databases, and the Cochrane Central Register of Controlled Trials until December 2011, using the search terms: ‘Oesophageal neoplasms’, ‘oesophageal cancer’, ‘oesophageal adenocarcinoma’, ‘Barrett Oesophagus’, ‘Hydroxymethylglutaryl-CoA Reductase Inhibitors’ and ‘statin’. The reference lists of all selected articles were searched to identify any other relevant papers. Using these criteria, the initial search yielded 29 articles, of which three experimental[25-27] and eight prospective cohort studies[28-35] were deemed relevant for full-text review. No relevant published case-control studies or clinical trials were identified. Two of the included observational studies[30, 31] were larger follow-up investigations of previous work.[28, 34]

Two reviewers (LA and ARH), independently and as a team, extracted data from the epidemiological studies (Table 1). Measures of effect size pertaining to the risk of OC with any prior statin use were extracted. For the meta-analysis, stata version 11 (StataCorp LP, College Station, TX, USA) was used to calculate the pooled effect size (ES) using the inverse variance method, fixed effects model, based on measures of effect size and CIs presented in the included epidemiological studies. We assumed that relative risk, hazard ratio and odds ratio measured the same effect to combine these into a single measure. Given the disease is relatively uncommon it seems appropriate that the odds ratio can be interpreted as a relative risk. Risk estimates from BO cohorts were analysed separately to those from general population cohorts. Statistical heterogeneity was assessed using the I2 statistic.

Table 1. Characteristics of observational studies reporting the risk of OC with statins in BO or general population cohorts
StudyParticipantsDrug use, definition and their ascertainmentRisk estimates of statins (vs. controls) for OC or OAC, and the co-variates adjusted forLimitations
Barrett's oesophagus cohorts

Kastelein 2011.[34]

Endoscopy units in 15 centres.

Netherlands.

Participants recruited between 11/03– 12/04.

786 BO.

26 HGD.

12 OAC.

All statins.

Statin user defined as statin use during the study period.

Patient interview, questionnaire, pharmacy records.

Use of statins ≥1 month HR = 0.46 (95% CI = 0.21–0.99).

Use of statins ≤5 years HR = 0.51 (95% CI = 0.18–1.47).

Use of statins ≥5 years HR = 0.49 (95% CI = 0.18–1.29).

Use of both statins and NSAIDS HR = 0.22 (95% CI = 0.06–0.85).

Adjusted for: age, gender, BO length, baseline histology and use of other medications.

Limited categories to account for duration-response relationship.

No adjustment for BMI or smoking.

Follow-up bias.

Nguyen 2010.[32]

VA computerized databases.

United States.

Study period 2000–2004.

11823 BO.

116 OAC.

696 matched controls.

All statins.

Statin use defined as at least one statin prescription during the study period.

Pharmacy Benefit Management database.

At least one prescription of a statin IDR = 0.56 (95% CI = 0.36–0.87).

Use of statins >0–12 months IDR = 0.63 (95% CI = 0.38–1.06).

Use of statins >12 months IDR = 0.52 (95% CI = 0.30–0.91).

Adjusted for: race, out-patient encounters, non cancer disease comorbidity index, priority level and use of other medications.

Limited external validity – veterans, 97% male, older.

Not adjusted for BMI, alcohol, smoking.

General population cohorts

Bhutta 2011.[35]

GPRD.

United Kingdom.

Study period 1/1/2000–31/12/08.

4242 OC.

17233 controls.

All statins.

Statin use defined as prescription of statins for at least 10 months in the year preceding diagnosis of OC.

Read codes within GPRD.

Use of statins OR = 0.84 (95% CI = 0.73–0.95).

Lipophilic statins OR = 0.86 (95% CI = 0.75–0.98).

Hydrophilic statins OR = 0.71 (95% CI = 0.51–0.98).

Adjusted for: diabetes mellitus, BMI, smoking, aspirin, NSAIDs, Proton pump inhibitors, bronchodilators, theophylline and calcium channel blockers.

Data for OAC and OSCC combined.

Vinogradova 2011[33] QRESEARCH database.

United Kingdom.

10 year study period 1/1/98 –1/7/08.

3159 OC.

13041 controls.

All statins.

Current statin use defined as at least two prescriptions that started over a 5-year period, at least 1 year prior to diagnosis.

Read codes within QRESEARCH database.

Use of statins OR = 0.88 (95% CI = 0.77–1.01).

Adjusted for: Townsend score, BMI, smoking status, myocardial infarction, coronary heart disease, diabetes, hypertension, stroke, rheumatoid arthritis, use of NSAIDs, COX-2 inhibitors and aspirin.

Data for OAC and OSCC combined.

Hippisley-Cox 2010[31]QRESEARCH database.

United Kingdom.

Study period 1/1/02–30/6/08.

1809 OC.

2004 692 participants overall.

All statins.

New users of statins defined as a statin prescription that started during the study period.

Read codes within QRESEARCH database.

Use of statins in women HR = 0.68 (95% CI = 0.52–0.88).

Use of statins in men HR = 0.78 (95% CI = 0.66–0.91).

Adjusted for: age, BMI, Townsend score, smoking status, type 2 diabetes.

Data for OAC and OSCC combined.

Long-term statin exposure not measured.

No adjustment for NSAIDS or aspirin.

Kaye 2004[29]

GPRD.

United Kingdom.

Study period 1990–2002.

100 OC.

430 controls.

All statins.

Current statin use defined as a statin prescription that started at least 1 year prior to diagnosis.

Read codes within GPRD.

Use of statins OR = 0.8 (95% CI = 0.3–1.8).

Adjusted for: smoking, BMI, and average number of GP visits.

Data for OAC and OSCC combined.

Small number of cases (100). No adjustment for NSAIDS or aspirin.

Bradford Hill criteria

The evidence was appraised according to the BHC,[36] a systematic method to assess if statins are protective in the absence of randomised controlled trials. The BHC are nine guidelines to assess whether a relationship between an exposure and the risk of developing a disease is causal.[36] The absence or presence of any or all nine criteria can neither absolutely confirm nor refute causality. They were proposed by the English epidemiologist, Sir Austin Bradford Hill in 1965, and with relevance to statins and OAC include the following:

  1. Plausible biological mechanisms – experimental studies which demonstrate specific anticancer mechanisms for statins in OAC.
  2. Consistency – multiple studies in different populations showing the same effect of statins.
  3. Strength – the greater the effect size of an association between dependent (OAC) and independent variables (statins), the more likely it is to be causative.
  4. Temporality – the use of statins precedes the diagnosis of OAC.
  5. Biological gradient or dose-response – the higher the dose of the statin, or longer the duration of use, the lower the risk of OAC.
  6. Analogy – absence of another confounding variable related to statin use which provides the true explanation for the association.
  7. Specificity – the presence of a clearly defined risk factor (statin use) that determines a single disease outcome (OAC), in the absence of confounding.
  8. Coherence – a logical connection between the laboratory and epidemiological evidence.
  9. Experiment (reversibility) – interventional studies, i.e. randomised controlled trials (RCTs), which demonstrate that those taking statins are at a lower risk of developing OAC.

Plausible biological mechanisms and coherence

Emerging experimental work suggests that statins may prevent OAC through their effects on cell proliferation, apoptosis and metastatic potential (Figure 1, adapted from Konstantinopoulos, 2007[37]). Statins inhibit the enzyme, 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase, the rate-limiting step in the mevalonate pathway, required for cholesterol synthesis.[38] Additional intermediates of this enzymatic pathway include farnesyl pyrophosphate (farnesyl-PP) and geranylgeranyl pyrophosphate (geranylgeranyl-PP), which are responsible for the prenylation (a post-translational modification) and therefore cell-membrane localisation of a number of guanosine-triphosphate-bound proteins, including Ras, Rac and Rho.[39] These are important in regulating cell signalling cascades involved in cell survival and proliferation. Therefore, by inhibiting HMG-CoA reductase, statins could exert potential anticancer effects by reducing cell survival and turnover, and stimulating apoptosis. The biological anticancer effects of statins have also been reported in other cancer cell lines in addition to OAC, including colorectal,[40-43] hepatocellular,[44, 45] pancreatic,[46-48] prostate,[49] glioblastoma,[50] thyroid,[51] breast,[52] lung,[53, 54] osteosarcoma[55] and myeloma.[56] In colorectal cancer cell lines, where most in vitro experimental evidence exists, statins enhanced apoptosis, reduced cell proliferation, inhibited angiogenesis and decreased metastatic potential.[40-43]

Figure 1.

The potential anticancer mechanisms of statins on OAC.

The cellular effects of statins on OAC cell lines have been examined in three in vitro studies, and all reported anti-proliferative and pro-apoptotic effects.[25-27] The first[25] observed that three statins (simvastatin, lovastatin and pravastatin) exhibited a drug class effect and inhibited proliferation and induced apoptosis in the Barrett's OAC cell line, OE33, in a dose-dependent manner. This action was mediated through inhibition of Ras farnesylation, but not geranylgeranylation, leading to a downstream reduction of two protein kinases, extracellular signal-regulated protein kinase and protein kinase B/Akt. Both kinases are responsible for cell survival and growth signal transduction in BO and OAC cell lines, and their reduction could inhibit carcinogenesis. Furthermore, expression of the pro-apoptotic proteins, Bax and Bad, was increased by statins in these cell lines, although additional experimental work is required to determine the pathway through which statins stimulate these pathways. Concomitant treatment with each of the three statins and a selective COX-2 inhibitor resulted in an additive reduction in viable cell counts, an observation that may be due to the differing mechanisms of action of these two drugs in OAC cell lines. The second study[26] demonstrated that simvastatin attenuated cell viability and proliferation and induced apoptosis in FLO-1, a human OAC cell line, in a dose-dependent manner. Simvastatin also induced a pronounced reduction in activated nuclear factor kappa beta and intracellular adhesion molecule-1 (ICAM-1). The latter is a critical adhesion molecule involved in transendothelial tumour cell migration,[57] which promotes metastases.[58, 59] Atorvastatin had less potent effects than simvastatin on cell growth and survival, while pravastatin had no such effects. The third study[27] examined a single OAC cell line, OE-19, and reported that in the presence of tumour necrosis factor-alpha (TNF-α), simvastatin reduced cyclooxygenase-2 mRNA expression, increased Bax and reduced the anti-apoptotic protein, Bcl-2.

Therefore, while the BHC, plausible biological mechanisms, are fulfilled for the anticancer effects of statins mediated on OAC, additional experimental work is required to confirm these effects in metaplastic and dysplastic Barrett's cell lines, and in an organotypic model to determine whether or not statins inhibit malignant progression through the metaplasia-dysplasia-carcinoma sequence. There is evidence of a second BHC, namely coherence, in that there is a logical connection between the experimental work and the epidemiological studies, which are now discussed.

Epidemiological criteria – strength, consistency, temporality and analogy

To support the experimental work on anticancer mechanisms, epidemiological studies are required where patients prescribed statins are shown to be at a lower risk of developing OAC. Cohort studies are the preferred observational methodology, compared to case-control studies, to investigate associations between statin use and OAC. In a cohort study, recall bias for statin use is reduced as the data are collected before the development of cancer, and selection bias is limited as both future cases and controls develop from the same baseline population. Furthermore, a cohort study allows a thorough reporting of drug use at repeated intervals before diagnosis and allows for the accurate assessment of temporal relationships.

The association between statins and the progression from BO to either high-grade dysplasia (HGD) or OAC has been investigated by two prospective cohort studies.[30, 32] In the first study, a multi-centre investigation from the Netherlands, 570 patients (72% men) were recruited with BO, of whom 26 (4.6%) later developed HGD and 12 (2.1%) OAC, after a median follow-up of 4.5 years.[32] Statin use for greater than 1 month was associated with a statistically significant inverse risk for neoplastic progression (HR = 0.46, 95% CI = 0.21–0.99, = 0.048), although this was only observed in men over 60 years of age. Important covariates, namely body mass index (BMI) and smoking were not adjusted for, which could underestimate the true effect size for statins. These covariates are associated with statin use[28] and increase the risk of OAC.[60] The concomitant use of both statins and NSAIDs was associated with a greater risk reduction (HR = 0.22, 95% CI = 0.06–0.85) than single use of either agent,[32] highlighting a potential additive effect which is supported by the experimental work.[25] The authors reported that their cohort was representative of the known BO population.[32] Within a cohort of 11 823 patients with BO in the US Veterans Affairs Healthcare System, 116 cases (0.98%) of OAC were matched to 696 controls, in a nested case-control analysis, and reported a 44% reduction in risk of neoplastic progression with statin prescriptions [incidence density ratio (IDR) = 0.56, 95% CI = 0.36–0.86].[30] Again, the majority of relevant covariates, including BMI and smoking were not adjusted for in analyses. Participants were predominantly middle-aged to older Caucasian men (97% men), therefore limiting the external validity[30]; although this is the demographic group most at risk of malignant transformation from BO. Point estimates for the inverse association with OAC were similar for both NSAIDs and statins in both BO cohorts,[30, 32] at approximately 0.5.

The effect of statins on the risk of OC in the general population has been investigated in four prospective cohort studies, all from the UK.[28, 29, 31, 33] All these reported inverse associations between statins and OC, although to date, none according to the histological subtype of cancer. Therefore, the protective effect in OAC specifically in general population cohorts is unknown, although in the UK, OAC is more prevalent than OSCC: two thirds of oesophageal tumours diagnosed between 2007 and 2009 were adenocarcinomas.[61] In the first UK study in 2004, in the General Practice Research Database (GPRD), the world's largest computerised database of healthcare data from 630 primary care practices throughout the UK, the risks of 13 different cancers according to statin use were investigated, and 100 incident cases of OC were identified during follow-up.[29] The relative risk (RR) of OC for current users of statins was 0.8 (95% CI = 0.3–1.8). Body mass index and smoking were adjusted for in the risk estimates, although, the concomitant use of NSAIDs or aspirin was not. A second prospective UK cohort study, in the QResearch database,[28] identified 1809 cases of OC from a population of 2 million, reported an inverse association between statin use and the development of OC in both men [hazard ratio (HR) = 0.78, 95% CI = 0.66–0.91] and in women (HR = 0.68, 95% CI = 0.52–0.88). The number needed to treat with statins to prevent one additional case of OC in men was 1266, and in women it was 1082. Importantly, many covariates including age, BMI, smoking, with the exception of concomitant relevant drug use, were adjusted for. A more recent, larger study in the QResearch database, of 3159 cases of OC, reported an inverse association with borderline significance, between statin use and OC [odds ratio (OR) = 0.88, 95% CI = 0.77–1.01, = 0.07].[31] The majority of relevant covariates were considered, including age, gender, BMI, smoking status and concomitant use of NSAIDs and aspirin. Finally, the largest epidemiological prospective cohort study, from the GPRD (currently reported in abstract form),[33] included 4242 participants with OC, who were matched with 17 233 controls, and reported a significant inverse association between statin use and the development of OC (OR = 0.84, 95% CI = 0.73–0.95). The inverse associations persisted when categorised into hydrophilic statins (OR = 0.71, 95% CI = 0.51–0.98) and lipophilic statins (OR = 0.86, 95% CI = 0.75–0.98). Known covariates were adjusted for, including BMI, smoking, use of NSAIDs, aspirin, proton pump inhibitors and drugs which relax the lower oesophageal sphincter. The results from the QRESEARCH database and GPRD are generalisable to the UK population, as these databases record information on large populations.

Two prospective cohort studies,[30, 32] involving 1382 participants, were included in the meta-analysis of risk of OAC in cohorts with BO with statin use. One study[34] was excluded from the meta-analysis as a larger follow-up study[30] involving additional participants was preferentially included. The pooled effect size was 0.53 (95% CI = 0.36–0.78, = 0.001) with minimal heterogeneity (I2 = 0%) (Figure 2a). Three prospective cohort studies,[29, 31, 33] involving 35 214 participants, were included in the meta-analysis of risk of OC among general population cohorts with statin use. One study[28] was excluded from the meta-analysis as a larger follow-up study[31] involving additional participants was preferentially included. Another study was excluded, as risk estimates were not adjusted for any confounders.[35] The pooled effect size was 0.86 (95% CI = 0.78–0.94, = 0.001) with minimal heterogeneity (I2 = 0%) (Figure 2b). There were too few studies to enable reliable publication bias testing in either meta-analysis.

Figure 2.

Fixed effects meta-analysis of statin use and risk of OAC in Barrett's oesophagus cohorts (a), and risk of OC in general population cohorts (b).

Several of the epidemiological BHC were fulfilled for the association between statin use and the development of OAC in both BO and general population cohorts, including strong effect sizes, consistency, temporality and analogy. All the included observational studies were of a cohort design, therefore the BHC of temporality was met as data on statin use were collected prior to the development of OC. The observational studies reported strong inverse associations, with a 47% reduction in risk of OAC in BO cohorts with statin use, and a 14% reduction in risk of OC in general population cohorts in pooled analyses. Reported reductions in risk were consistent between studies in BO cohorts (= 0.67, I2 = 0%) and general population cohorts (= 0.88, I2 = 0%). The criterion of analogy was supported, as known confounders could not explain the association in the general population as potential exposures such as BMI and smoking, which are independently linked to statin use and OAC, were mostly adjusted for. However, in the BO cohorts, such covariates need to be measured and adjusted for in future work. To date, there are no published observational studies that report associations between statins and specifically OAC in the general population. Further monitoring of these cohorts to allow acquisition of more cases will allow the effect individual statins to be assessed. This will be important if an RCT of statins in the prevention of OAC is contemplated, to appropriately select which statin to assess.

Biological gradient: dose and duration-response relationship

A causal protective effect for statin use, as proposed by Bradford Hill, would be supported if higher doses and/or longer prescription times of statins produced greater reductions in the risk of developing OAC than lower doses or shorter prescription times. In the first QRESEARCH study, a dose-response was reported between simvastatin use and the risk of OC in men only, when comparing non-users of statins (HR = 1.00) with low dose simvastatin (10–20 mg) (HR = 0.91, 95% CI = 0.73–1.12) and higher doses (40–80mg) (HR = 0.66, 95% CI = 0.48–0.91).[28] The lack of a dose-response in women may reflect the small sample size, as only 50 statin users were female patients with OC. Furthermore, there were insufficient participant numbers to allow meaningful dose-response calculations for the statins other than simvastatin. The reduction in risk of OC was apparent between 1 and 3 years after starting statins. After stopping the medications, the risk returned to normal within the first year in women, and in men within 1–3 years. No other studies have investigated dose-responses, although the effect of duration-response was reported in three other investigations.[30-32] In the larger QRESEARCH study on OC, no significant associations were detected for statin use across time periods between 1 and 10 years prior to diagnosis (= 0.87).[31] The Veterans Affairs study, in BO patients, reported a statistically significant difference (= 0.02 for trend) in the risk of developing of OAC, comparing non-users of statins (IDR = 1.00), statin use for less than 1 year (IDR = 0.63 95% CI = 0.38–1.06) and use for longer than 1 year (IDR = 0.52 95% CI = 0.30–0.91).[30] In the Dutch study, there was no significant difference (= 0.20) between less than 5 years' (HR = 0.51 95% CI = 0.18–1.47) and more than 5 years' use of statins (HR = 0.49 95% CI = 0.18–1.29)[32]; however, this study may have been underpowered to detect a significant difference if it existed as only 107 patients with BO, of whom 5 developed HGD or EAC, were statin users for at least 5 years. The effect of duration-response reported in the Veterans Affairs study is concordant with the first QRESEARCH study, which observed that the maximal protective effect of statins in OC was within 3 years. In summary, there is some evidence to support the BHC, biological gradient, which should be further addressed in future work with a detailed assessment over different time periods.

Specificity

The BHC, specificity, was not fulfilled, as statins exert pleiotropic effects in a number of cardiovascular, pulmonary, neurological, renovascular and rheumatological disorders; attributed to the modulation of the range of processes that the mevalonate pathway governs.[62] Statins may also exert chemoprotective and/or chemotherapeutic effects in a number of other malignancies, in addition to OAC.[63] According to Bradford Hill, while fulfilment of specificity is suggestive of causality, such as its particular utility in ascribing causality in infectious diseases, its absence does not disprove it: the numerous diseases attributed to smoking are an example.[36]

Experimental evidence

Bradford Hill stated that evidence supporting the criterion, ‘experiment’, was the most suggestive for a causal relationship. To fulfil this, RCTs are required which demonstrate that statins decrease the risk of developing OAC compared to placebo. To date none exist, although RCTs could investigate whether statins reduce the risk of OAC in either the general population or in BO cohorts. Trials conducted in the general population would be considerable undertakings as the number needed to treat to prevent an additional case of OC may be over 1000, using data from the QRESEARCH study, and such a trial may be more suited to examine multiple cancer endpoints, including OAC.[28] A more focussed approach may be a clinical trial in patients with BO to investigate if statins prevent the progression to OAC as the primary outcome. Statin use could lead to additional benefit to patients, as the age-adjusted mortality rates from ischaemic heart disease for men and women with BO are higher than the general population.[64] Future clinical trials could also possibly assess a combination of a statin with either NSAIDs (including aspirin), given their apparent additive protective effects in experimental and epidemiological work, or proton pump inhibitors. Such clinical trials of statins are required for the completeness of the BHC to assess if a true protective effect exists.

Conclusions

There is emerging experimental and epidemiological evidence to support a protective effect for statins in the development of OAC. Statins have several plausible mechanisms of action to explain their effects on cell proliferation, apoptosis and metastatic potential in OAC cell lines. The criterion of coherence is fulfilled as there is a logical link between the experimental and epidemiological work. Many of the epidemiological criteria are met as statins demonstrate consistent large inverse associations for the development of OC in both BO and general population cohorts, with some evidence of both dose and duration-response relationships. As drug exposure was measured prior to the development of OC, the criterion of temporality is fulfilled. There is no current known confounding variable to explain the association, which persists after correction for confounders, although not all studies have adjusted for co-variates. Further large prospective studies in BO cohorts investigating the risk of malignant progression in other populations, and importantly correcting for all known covariates are needed. To support the most important BHC, ‘experiment’, RCTs are required to investigate statins as chemopreventive agents in BO and/or the general population. Such work needs to progress to not only construct aetiological models of OAC, but to offer chemopreventive strategies and, therefore, potentially prevent the development of this aggressive cancer.

Acknowledgement

Declaration of personal and funding interests: None.

Ancillary