Objective To evaluate the association between statin use and risk of biliary tract cancers (BTC).
Design This is a nested case–control study conducted in the UK Clinical Practice Research Datalink. We included cases diagnosed with incident primary BTCs, including cancers of the gall bladder, bile duct (ie, both intrahepatic and extrahepatic cholangiocarcinoma), ampulla of Vater and mixed type, between 1990 and 2017. For each case, we selected five controls who did not develop BTCs at the time of case diagnosis, matched by sex, year of birth, calendar time and years of enrolment in the general practice using incidence density sampling. Exposures were defined as two or more prescription records of statins 1 year prior to BTC diagnosis or control selection. ORs and 95% CIs for associations between statins and BTC overall and by subtypes were estimated using conditional logistic regression, adjusted for relevant confounders.
Results We included 3118 BTC cases and 15 519 cancer-free controls. Current statin use versus non-use was associated with a reduced risk of all BTCs combined (adjusted OR=0.88, 95% CI 0.79 to 0.98). The reduced risks were most pronounced among long-term users, as indicated by increasing number of prescriptions (ptrend=0.016) and cumulative dose of statins (ptrend=0.008). The magnitude of association was similar for statin use and risk of individual types of BTCs. The reduced risk of BTCs associated with a record of current statin use versus non-use was more pronounced among persons with diabetes (adjusted OR=0.72, 95% CI 0.57 to 0.91). Among non-diabetics, the adjusted OR for current statin use versus non-use was 0.91 (95% CI 0.81 to 1.03, pheterogeneity=0.007).
Conclusion Compared with non-use of statins, current statin use is associated with 12% lower risk of BTCs; no association found with former statin use. If replicated, particularly in countries with a high incidence of BTCs, our findings could pave the way for evaluating the value of statins for BTC chemoprevention.
- biliary duct carcinoma
- gall bladder cancer
Statistics from Altmetric.com
Significance of this study
What is already known on this subject?
Statins are commonly used cholesterol-lowering medications that have demonstrated effectiveness in the primary and secondary prevention of cardiovascular disease.
Several epidemiological studies have shown that statins are associated with a lower risk of gallstone disease (gallstones or a history of cholecystectomy), but associations with biliary tract cancers (BTC) have not been well studied.
What are the new findings?
There was significantly reduced risk of BTCs among current users of statins, as compared with non-users (OR 0.88).
The reduced risks were most pronounced among long-term users, as indicated by increasing number of prescriptions (p for trend=0.016) and cumulative dose of statins (p for trend=0.008).
How might it impact on clinical practice in the foreseeable future?
Current or long-term statin use may reduce risk of BTCs.
If replicated in other settings, the value of statins for BTC chemoprevention should be evaluated.
Biliary tract cancers (BTC) comprise cancers of the gall bladder, bile duct (ie, both intrahepatic and extrahepatic cholangiocarcinoma) and ampulla of Vater. BTCs are rare but highly fatal diseases, with a median overall survival of ~12 months.1 Although the aetiology of BTCs is largely unknown, the epidemiological and molecular characteristics suggest that they are distinct disease entities.1 2 For example, gall bladder cancer (GBC) is more common in women than men, while other BTCs are more common in men.3 Similarly, cirrhosis and infection with hepatitis B and C viruses are associated with intrahepatic cholangiocarcinoma, while their associations with GBC and ampulla of Vater cancer (AVC) are less clear.1 4
Despite these differences, BTCs share common risk factors. For example, disorder of lipid regulation (ie, hyperlipidaemia) has been associated with an increased risk of cancer at all sites in the biliary tract.5 Hyperlipidaemia is thought to interact with inflammation and proinflammatory immune response to influence the risk of BTCs.6 7 For GBC, local inflammation due to gallstones plays a central role,8 while for cholangiocarcinoma, both intrahepatic and extrahepatic sclerosing cholangitis play a major role.1 Inflammation from biliary adenomas, which are seen in patients with familial adenomatous polyposis,9 is thought to contribute to the 100-fold higher risk of AVC in that population.10 Finally, other causes of inflammation, such as bacterial infection and obesity/metabolic syndrome, may play a role in the aetiology of BTC.1
Given these strong ties with dyslipidaemia and inflammation, statins (3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors) could potentially reduce the risk of BTCs. Statins are commonly used cholesterol-lowering medications that have demonstrated effectiveness in the primary and secondary prevention of cardiovascular disease.11 Several studies have shown that statins are associated with a lower risk of gallstone disease (gallstones or a history of cholecystectomy),12–15 but associations with BTCs have not been well studied. To further our understanding of the associations between statins and BTCs, we thus conducted a nested case–control study within a large database of primary care electronic medical records—the UK Clinical Practice Research Datalink (CPRD).16
This study is based on data from the CPRD GOLD database November 2017 release, obtained from the UK Medicines and Healthcare Products Regulatory Agency database, reused with the permission of the Health and Social Care Information Centre. All rights reserved. The interpretation and conclusions contained in this study are those of the authors alone.
Study population and design
The UK CPRD was established in 1987 and contains information on ~8.5% (13.3 million) of the UK population. All data are provided by general practitioners; the CPRD population is representative of the general UK population.17 Detailed information includes demographic data, physician contacts, tests conducted and medications prescribed. Diagnoses, physical findings, symptoms and administrative events, such as referrals to specialists, are recorded using Read codes, the standard clinical terminology system used in general practice in the UK.18
In this nested case–control study, we included all persons whose most recent active registration at a CPRD practice overlapped with the study period from 1990 (3 years after the establishment of the database) through 2017. Exit date was defined as the date of cancer diagnosis/selection, death, leaving the practice or last collection of data from the practice, whichever occurred first. Eligible cases included persons aged 21–100 years with a first cancer diagnosis (except non-melanoma skin cancer) of BTC (Read codes provided in online supplementary table 1) identified in CPRD during 1990–2017. To ascertain statin use, we excluded the 1 year period prior to cancer diagnosis to minimise the possibility of reverse causation (ie, incipient cancer leading to statin use). All cases were required to have at least 2 years of recorded activity in the CPRD prior to the date of cancer diagnosis.
Five controls per case were randomly selected using incidence density sampling. Controls were individually matched on sex, year of birth (±3 years), diagnosis year (±3 years) and number of years in the general practice and in the CPRD prior to diagnosis/selection date. All controls were required to be alive, cancer free (except for non-melanoma skin cancer) and have at least 2 years of recorded activity in the CPRD prior to the diagnosis date of their matched case. Selected controls with exit date before entry were excluded (n=55).
Ascertainment of statins and covariates
Records of medications prescribed more than 1 year prior to diagnosis/selection were obtained from the electronic patient prescription records. Drugs of interest included statins, other lipid-lowering agents (ie, bile acid sequestrant, cholesterol absorption inhibitor, fibrate, nicotinic acid derivative, omega-3 fatty acid and bisphenol antioxidant), treatment for type 2 diabetes (ie, metformin, insulin, acetohexamide, tolbutamide, chlorpropamide, glibenclamide, glibornuride, gliclazide, rosiglitazone, tolazamide and tolbutamide), aspirin and cortisone, which were included because of their potential effect on metabolism. Statin use was defined as having two or more records of statin prescriptions, and non-use of statins was defined as having 0 or one prescription record.19 Former statin use was defined as use ending more than 2 years prior to the diagnosis/selection date. Use ending within 2 years prior to the diagnosis/selection date was considered as current use. Based on the distribution among controls, statin use was classified by time since first recorded statin use to 1 year prior to the diagnosis/selection (0–2, 3–5, 6–8 and 9+ years), number of prescriptions (2–15, 16–35, 36–65 and 66+, representing approximate duration of use) and cumulative dose (quartiles), which was defined as number of pills multiplied by the dose per pill. This grouping strategy permitted examination of the trend for the association of increasing statin use with risk for BTCs. Finally, specific commonly used statins (atorvastatin, simvastatin, pravastatin and rosuvastatin) were examined individually.
Because body mass index (BMI), smoking, alcohol drinking, diabetes mellitus, gallstones, cirrhosis, hepatitis virus infection, chronic heart disease and dyslipidaemia may be associated with both statin use and BTCs or BTCs at specific subsites, we identified persons who had these conditions more than 1 year prior to the diagnosis/selection. We also assessed healthy user effect, healthy adherer effect and functional status more than 1 year prior to the diagnosis/selection, because these factors have been suggested as important confounders in observational studies (online supplementary materials).20
We used conditional logistic regression to estimate ORs for the association of statins with BTCs overall and with individual types (ie, GBC, cholangiocarcinoma and AVC). The regression model included adjustment for known BTC risk factors: BMI, smoking status, alcohol drinking status and a history of diabetes.21 GBC-specific analyses excluded 177 controls with a history of cholecystectomy because those controls were not at risk for developing GBC, and history of gallstones was additionally included in the model. Cholangiocarcinoma-specific analyses additionally included history of cirrhosis in the model.
Other potential confounders were selected if they changed the adjusted OR for BTC and statin use status by more than 10%. We first considered a fully adjusted model that included all potential confounders: a history of hepatitis virus infection, chronic heart disease, dyslipidaemia, high exercise, healthy diet, cancer screening, influenza vaccination, pneumonia, dementia, comorbidity and use of treatment for type 2 diabetes, aspirin, cortisone or other lipid-lowering agents. However, additional adjustment for those variables resulted in only marginal changes (<10%) in the ORs (online supplementary table 2); such adjustments were therefore not included in the final model. We also selected covariates by mutually entering all potential confounders into a stepwise logistic regression model, with p<0.15 as the model entry criterion and p<0.05 for a variable to remain in the model. Likewise, adjustment for variables selected resulted in OR changes of less than 10% (online supplementary table 2). We thus only present results from the minimally adjusted model. We tested for linear trends across statin use categories with the Wald test using categorical values of the number of prescriptions and cumulative dose with 1 degree of freedom.
We evaluated potential effect modification by matching variables (ie, sex, age at 1 year prior to the diagnosis/selection (≤70 vs >70 years; the age cut-off was chosen to create two equally large age groups among the control subjects, among whom the median age was 72 years) and calendar year at diagnosis/selection (≤2005 vs >2005 because 10 mg of simvastatin became available over the counter in August 2004)22). Likelihood ratio tests were used to compare models with and without an interaction term between statin use and each potential modifier. Because statin use has been associated with increased risk of diabetes,23–25 we conducted a joint analysis of diabetes and statin use initiated before or after diabetes diagnosis. We used patients with diabetes who did not use statins as the reference group for a model focused on the potential protective effect of statins on BTCs among patients with diabetes; this model was additionally adjusted for treatment for type 2 diabetes. We then reran the model using non-diabetic patients who did not use statins as the reference group to focus on the effect of former or current statin use among non-diabetics.
In sensitivity analyses, we studied only incident statin users, defined as people who had their first recorded use of statins 5 years or longer after entering CPRD. Studying a cohort of incident users allows us to obtain results more robust to bias from treatment assignment, a common limitation to study medication use in an observational setting, as opposed to in a randomised controlled trial.26 We also conducted analyses excluding 145 cases with carcinoma in situ of the biliary system and their matched controls, excluding subjects with non-biliary cancer in situ (ie, all carcinomas in situ except that of the biliary system) since diagnosis of in situ cancer may have led to lifestyle changes, or excluding subjects with missing/unknown information on smoking, alcohol drinking and BMI.
Statistical analyses were performed using R Statistical Software (Foundation for Statistical Computing, Vienna, Austria). All statistical tests were two sided, and p<0.05 was considered statistically significant.
Table 1 shows the distribution of demographic characteristics and other potential risk factors. A total of 3118 BTC cases (708 GBC, 1678 cholangiocarcinoma, 228 AVC, 500 mixed, 4 unknown) and 15 519 cancer-free controls were included (table 1). By design, cases and controls were individually matched on sex, age, calendar year and years in the CPRD participating practice. Cases were more likely to be obese; be current smokers; drink alcohol; have a history of chronic diseases, including diabetes, gallstones and cirrhosis; be infected with hepatitis viruses; have treatment for type 2 diabetes; and use aspirin or cortisone. Controls were more likely to be diagnosed with dementia.
Table 2 shows adjusted ORs for statin use and risk of all BTCs combined. Although former statin use was not associated with BTC risk, current use versus non-use was associated with a reduced risk of BTCs (adjusted OR=0.88, 95% CI 0.79 to 0.98, table 2). Individuals who started using statins 9 or more years before 1 year prior to the diagnosis/selection had reduced risk of BTCs (OR=0.85, 95% CI 0.72 to 1.01), but the trend was not consistent across increasing categories of time since first statin use (ptrend=0.128). Although there was some evidence of a dose-response trend with increasing number of prescriptions (ptrend=0.016) and cumulative dose of statins (ptrend=0.008), the trend was not consistent across all categories,; only the comparison between highest quartile (Q4) and non-use of statins was statistically significant. For example, compared with non-use of statins, the adjusted OR of the highest quartile (Q4) for cumulative dose was 0.81 (95% CI 0.68 to 0.96). We did not observe heterogeneity by type of statin but had limited statistical power for individual types of statins (online supplementary table 3).
The reduced risk associated with statin use was similar for individual types of BTCs (table 3). For example, the adjusted ORs for current statin use versus non-use and risk of GBC, cholangiocarcinoma and AVC were 0.86 (95% CI 0.67 to 1.09), 0.87 (95% CI 0.75 to 1.01) and 0.89 (95% CI 0.59 to 1.34), respectively. There were no clear dose–response relationships between risk of individual types of BTCs and number of prescriptions and cumulative dose, but the ORs were of similar magnitude to those for BTCs overall.
The observed associations were similar for men and women, for individuals ≤70 and >70 years of age at 1 year prior to the diagnosis/selection and for those with a diagnosis/selection year of ≤2005 and >2005 (all pheterogeneity>0.05, online supplementary table 4). In the joint analysis by history of diabetes, we found that statin use was associated with a lower risk of BTCs among patients with diabetes, with an adjusted OR for current statin use versus non-use of 0.72 (95% CI 0.57 to 0.91). Among non-diabetics, the adjusted OR for current statin use versus non-use was 0.91 (95% CI 0.81 to 1.03, pheterogeneity=0.007). The reduced risk of BTCs was more pronounced among participants with statins prescribed before diabetes was diagnosed (adjusted OR=0.59, 95% CI 0.44 to 0.81) than among those with statins prescribed after diabetes was diagnosed (0.84 (95% CI 0.66 to 1.08), pheterogeneity=0.03).
The sensitivity analyses remained agreed that current statin use is associated with reduced BTCs among incident statin users (OR=0.81, 95% CI 0.70 to 0.94), excluding persons with biliary carcinoma in situ and their controls (0.88, 95% CI 0.79 to 0.98), and among cases and controls without a history of cancer in situ except carcinoma in situ of biliary system (0.89, 95% CI 0.79 to 0.99) or without missing/unknown status for smoking, alcohol drinking and BMI (0.90, 95% CI 0.80 to 1.02).
In this large study, we observed a reduced risk of BTCs among current statin users compared with non-users. Reduced risk of BTCs was also more pronounced among subjects with the highest number of prescriptions and cumulative dose of statins, suggesting that long-term statin users may have a lower risk of developing BTCs. The risk estimate was consistent across individual types of BTCs, age groups, sex and calendar time. Adjustment for various risk factors did not substantively alter the results. Finally, we also observed a risk reduction among persons with diabetes. If replicated in other studies, these findings suggest that statins may have a role as chemopreventive agents.
Comparison to other studies
To our knowledge, this study is the first to address the association between statin use and risk of cancer across the biliary tract. Our findings are in line with the one previous study focusing on cholangiocarcinoma in Taiwan, which reported that the adjusted OR for statin use versus non-use for cholangiocarcinoma was 0.80 (95% CI 0.71 to 0.90).27 The results are also consistent with studies that have demonstrated a decreased risk of gallstone disease,12–15 a risk factor for BTCs,28 especially GBC.1 2 Statins could potentially reduce risk of BTCs through suppression of biliary cholesterol secretion and saturation and anti-inflammatory and endothelial effects.29 30 The reduction in hepatic and gall bladder bile cholesterol concentrations could alter bile acid composition and interfere with gallstone production or even promote gallstone dissolution.31 In addition, both experimental and clinical studies provide evidence that statins have anti-inflammatory effects.29 30 Statins reduce circulating C-reactive protein and proinflammatory cytokines and chemokines.32 33 Statins also favourably affect important pathways regulating nitric oxide bioavailability, which is critical for maintaining endothelium homeostasis via statins’ vasodilatory, anti-inflammatory and overall antiatherogenic effects. In addition, statins have been shown to regulate expression of many inflammatory genes.30 Associations between statin use and lower risk of cancers in the digestive system related to inflammation, including cancers of the colorectum,34 35 liver19 36 37 and oesophagus,38–40 have been reported in epidemiological studies. However, we did not observe a joint effect of use of both statins and aspirin (data not shown), as has been reported for oesophageal cancer.39 40
Our findings of reduced BTC risk among persons with diabetes using statins should be interpreted with caution. Some studies suggest that statins may increase the risk of diabetes, although results are inconsistent.23 24 Experimental data support the hypothesis that statins may cause diabetes by altering glucose homeostasis through both impaired insulin secretion and diminished insulin sensitivity.41 Following this hypothesis, statins could increase the risk of BTCs by increasing risk of diabetes. However, in the present study, we observed a reduced risk of BTCs among persons with diabetes who used statins before their diabetes was diagnosed, as compared with those with diabetes who did not use statins. Lipophilic (simvastatin and atorvastatin) and hydrophilic (pravastatin and rosuvastatin) statins have been observed to have different diabetogenic effects,42 but we lacked the power to investigate whether the observed reduced risk of BTCs was driven by specific types of statins. Additional studies are needed to clarify the associations between statins, diabetes and BTCs, but overall our results suggest that statins reduce the risk of BTCs both in patients with diabetes and those without.
Our study has several strengths. First, it was conducted using a large, well-established, validated, longitudinal primary care database with long-term follow-up. This database is known for accuracy of diagnoses, including cancer, and completeness of pharmacy data.43–46 Second, all information on diseases and drug exposures in the CPRD is recorded prior to BTC diagnosis and in the absence of a study hypothesis, reducing the risk of recall bias. Third, we could evaluate the impact of various important confounders (BMI, smoking, alcohol drinking, a history of diabetes, hepatitis virus infection, chronic heart disease, dyslipidaemia and use of treatment for type 2 diabetes, aspirin, cortisone or other lipid-lowering agents); such information is usually not accessible in registry-based studies. Although we did not adjust for socioeconomic status (SES), SES has not been associated with statin prescribing in England.47 In addition, cases and controls were matched on general practice, and the universal healthcare coverage in the UK may, to some degree, control for SES. Finally, we conducted a large number of sensitivity analyses that further support the robustness of our findings.
Our study has a few limitations. First, although our sample size was sufficient for all BTCs combined, power was limited for individual types of BTCs and individual statins. Second, compliance rates among persons taking statins could not be evaluated, leading to potential exposure misclassification. Third, although we adjusted for several factors, these conditions could be inconsistently and incompletely recorded by general practitioners. For example, our adjustment for confounding by healthy user effect, healthy adherer effect and functional status or cognitive impairment, which are not usually considered in observational studies,20 did not change the magnitude of the association between statin use and risk of BTCs. This result could be explained by the fact that residual confounding is possible even after our adjustment. Fourth, previous validation studies have reported that cancer diagnoses recorded by general practitioners in CPRD are accurate and complete.43 45 46 However, the diagnosis of BTCs in the present study was not verified against pathology records, and some misclassification cannot be ruled out. We therefore believe that further studies conducted in databases with validated cancer diagnosis are warranted. Fifth, use of statins may not be accurately ascertained in CPRD due to introduction of over-the-counter 10 mg simvastatin in 2004.22 We do not have evidence to support that the misclassification occurred differentially based on case-control status. In the case that the misclassification of statin use is non-differential by case–control status, statins may actually have an even more significant risk reduction on BTCs than currently observed in this study. Finally, we cannot rule out a potential role of chance for false positive findings, given the number of tests evaluated in this study.
In conclusion, this large observational study provides evidence that current or long-term statin use is associated with decreased risk of BTCs. Our findings should be replicated in other populations, particularly in high-risk areas, such as Chile, to confirm our observations and further establish the potential value of statins for BTC chemoprevention.
Contributors ZL was involved in the study design, analysis and interpretation of data and writing the manuscript. RA, KAM, LAA, HTT, YZ, SMM and SMG were involved in conception of the study, interpretation of data and revision of the manuscript. BZ was involved in data analysis, interpretation of data and revision of the manuscript. JK was involved in conception of the study, study design, interpretation of data and writing the manuscript. All authors revised the article critically for important intellectual content and gave final approval of the version to be published.
Funding Intramural Research Program of the National Institutes of Health, National Cancer Institute, Division of Cancer Epidemiology and Genetics.
Disclaimer The funders had no role in the conduct of this research. The views expressed in this article are those of the authors and not necessarily those of the US Food and Drug Administration.
Competing interests HTT is currently an employee of the US Food and Drug Administration.
Patient consent Not required.
Ethics approval Independent Scientific Advisory Committee of the CPRD (proposal number 17_160.R).
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.