|Home | About | Journals | Submit | Contact Us | Français|
Robert O. Morgan, PhD, The Michael E. DeBakey Veterans Affairs Medical Center, 2002 Holcombe Blvd., Houston, TX 77030, Phone: (713) 794-8635; Fax: (713) 748-7359; e-mail: rmorgan/at/bcm.tmc.edu.
Experimental studies indicate a potential cancer prevention effect for statins. Given the increasing prevalence of statin use, and the rising incidence of hepatocellular carcinoma (HCC), the potential association between statins and HCC is an important issue to examine.
We conducted a matched case-control study nested within a cohort of patients with diabetes. Cases comprised incident HCC as defined by those occurring at least 6 months following entry in the cohort. Controls were identified by incidence density sampling from patients who remained at risk at the date of the HCC diagnosis matched on age and gender. We identified filled statin prescriptions as well as several potential confounding conditions, medications as well as propensity score to use statins. Odds ratios (OR) as estimates of the relative risk for HCC associated with statin use and 95% CIs were obtained using conditional logistic regression.
We examined 1303 cases and 5212 controls. The mean age was 72 years and 99% were men. A significantly smaller proportion of cases (34.3%) had at least one filled prescriptions for statins than controls (53.1%). There were no significant associations between HCC and non statin cholesterol or triglyceride lowering medications. The unadjusted OR for any statin prescription was 0.46 (95% CI: 0.40–0.517) and the adjusted OR was 0.74 (0.64, 0.87). To reduce the potential confounding effect of existing liver disease, we ran the analyses in a subgroup of patients without recorded liver disease; the ORs were slightly attenuated but remained highly significant both for any statin prescription (0.63 (0.50–0.78).
Statin use is associated with a significant reduction in the risk of hepatocellular carcinoma among patients with diabetes.
Interest has risen in the potential cancer preventive effect of cholesterol-lowering 3-hydroxy-3-methylglutaryl coenzyme A (HMG CoA) reductase inhibitors, commonly known as statins. Some but not all epidemiological studies have reported a modest statin-related reduction in the risk of several malignancies, including colon, breast, pancreas, and prostate cancer 1–4. Several general mechanisms have been proposed for this possible effect, including the disruption of malignant cell growth by inhibition of downstream products of the mevalonate pathway, an important activator of a number of cellular proteins, including K-ras 4. Statins also inhibit the activation of the proteosome pathway, limiting the breakdown of both p21 and p27, thus allowing these molecules to exert their growth-inhibitory effects 3,5.
The association between statin use and the risk of hepatocellular carcinoma (HCC) is unclear. HCC is a highly fatal malignancy that has been increasing in several regions of the world, including the United States 6. Experimental as well as indirect human data suggest that statins exert a beneficial action, reducing the progression of HCC 6,7. One study reported that statins inhibit the proliferation of HCC by inducing apoptosis and G1/S cell cycle arrest 8. Pravastatin has been shown to reduce progression and metastatic diffusion of established HCC in a rat model, an action that could be linked to the decreased matrix metalloproteinases activity 9. Similar findings have been suggested in a clinical trial of 91 patients with advanced HCC in which the median survival of 18 months of those on pravastatin was twice that of controls (p = 0.006) 10. Lastly, there is also a suggestion that HCV RNA replication is disrupted in vitro by high concentrations of statins as well as in vivo studies 11–13. On the other hand, statins have been linked in rodent studies to hepatic adenomas and carcinomas 14,15.
The association between statins and cancer, including HCC, were examined in clinical trials of statins in which association with cancer was a secondary end point. In at least two trials, investigators found a statistically nonsignificant inverse association between statin use and total cancer incidence 16,17. However, a meta-analysis 18 of five trials of statins and cardiovascular disease reported no association between statin use and risk of incident cancers. The number of HCC cases in these trials was not specifically mentioned, and it was likely very small (if existent). Clearly, there is not enough evidence to support a clinical trial in patients at this time, especially since statins can be hepatotoxic 19; therefore, epidemiological evidence for a cancer-protective effect is required. To meet that need, we undertook an epidemiological study in a large cohort of diabetics, whose risk of HCC is higher than average, to characterize the relationship between statin use and HCC and other liver disease.
We conducted a nested case-control study within a large cohort of patients with diabetes mellitus identified during the calendar year (CY) at the VA 1997 to 2002, in the Department of Veterans Affairs (VA) national databases, including the Patient Treatment File (PTF), Outpatient Clinic (OPC) files, Pharmacy Benefits Management (PBM) files, and Beneficiary Identification Records Locator System (BIRLS) death files. The institutional review boards of Baylor College of Medicine and the Michael E. DeBakey VA Medical Center in Houston, TX, approved this study. The PTF contain inpatient records, including date of admission and discharge, and up to 10 discharge diagnoses (by International Classification of Diseases [ninth revision] clinical modification [ICD-9] codes) for all hospitalizations at any of the more than 150 VA hospitals in the United States. The OPC files contain similar data for all outpatient encounters at any VA facility. We identified patients in OPC files and PTF with ICD-9 codes 250.0–250.9, 357.2, 362.0, and 366.41 as having diabetes. We employed an additional measure to confirm the diabetes classification by searching national VA PBM files for CYs 1999–2002 for oral diabetes medications (HS502), insulin (HS501), or blood glucose monitoring supplies (in drug class DX900).
Each cohort member’s entry date was the earliest qualifying inpatient or outpatient event associated with diabetes. The VA diabetes cohort was linked with Medicare files (CYs 2000–2002) to maximize the probability of capturing health care encounters among patients who used both the VA and Medicare. Finally, we used the date of death recorded in the VA BIRLS death file. The PTF and BIRLS death files together reportedly capture 90%–95% of deaths in VA users 17,18
Cases were defined as patients with incident HCC identified by ICD-9 code 155.0 recorded in VA or Medicare files at least 6 months following the entry date into the diabetes cohort. We limited eligible cases to those diagnosed between January 1, 2001, and December 30, 2002, to allow having 2–4 years of prior exposure with complete PBM and Medicare data. Controls were identified by incidence density sampling from subjects who remained at risk of HCC at the date of the HCC diagnosis for the case (index date) 20. This entailed matching on age (±1 year), sex, and the date of entry into the diabetes cohort (±30 days). Controls could not have a diagnosis of HCC within 6 months of the matched case’s HCC index date. Controls were selected randomly with potential for replacement. That is, one subject could be a control for more than one case, and any one control could later (> 6 months) become a case if diagnosed with HCC. We excluded patients without any VA pharmacy use. For cases and controls with dual VA-Medicare enrollment, we included only patients with continuous enrollment in non–HMO Medicare (defined as >75% of the duration between diabetes cohort entry and HCC index dates). These patients are likely to have complete VA prescription information (which is the most likely source of prescription drugs, even in dual VA-Medicare users 21 and should have virtually all of their health care activities captured either in VA or in Medicare records.
We identified patients who filled outpatient prescriptions for statins in any VA pharmacy between January 1, 1999, and the index date for cases and controls. Statin prescriptions were limited to those available in the VA, by formulary or non-formulary request, including simvastatin, lovastatin, atorvastatin, fluvastatin, pravastatin, and cerivastatin. We collected the dates of filled prescriptions, the daily dose, the number of days supplied, and the number of pills per prescription. We collected similar information on nonstatin lipid-lowering medications (cholestyramine, colesevelam, colestipol, ezetimibe, niacin), triglyceride-lowering medications (clofibrate, fenofibrate, gemfibrozil), and diabetes medications (insulin, sulfonylureas, thiazolidinedione).
We identified any liver disease defined by the following diagnoses recorded between January 1, 1997, and HCC index date: alcoholism (291.xx, 303.0x, 303.9x, 305.0x), cirrhosis (571.2, 571.5, 571.6), hepatitis C infection (070.41, 070.44, 070.51, 070.54, V02.62), hepatitis B infection (070.22, 070.23, 070.32, 070.33, V02.61), obesity (278), and alcoholic liver disease (571.0, 571.1, 571.2, 571.3). We also identified filled prescriptions for HCV treatment, HBV treatment, aspirin, non steroidal anti-inflammatory drugs (NSAID), and angiotensin converting enzyme (ACE) inhibitors. A subset with severe liver disease was identified with cirrhosis codes (571.2, 571.5, 571.6, 572.3). We also adjusted for filled diabetes medications.
Given the observational nature of the study, confounding by indication is a major concern and therefore we attempted to adjust for the propensity to give statins. We constructed a logistic regression model to generate a propensity score that expresses the likelihood of receiving a statin among members of the diabetes cohort. The outcome variable in the propensity model is any filled statin prescriptions, and the covariates were chosen, based on their biological plausibility and their significance in unadjusted analyses (p<.1), and entered in stepwise logistic regression model. The covariates that constituted the final model were coronary artery disease, myocardial ischemia, liver disease, obesity, and diabetic nephropathy (data not shown). We subsequently adjusted for the propensity score in the conditional logistic regression models examining filled statin prescriptions with the risk of HCC (see below)22.
Differences in covariate distribution between cases and controls were determined using χ2 tests for categorical variables and t tests for continuous variables. We estimated odds ratios (ORs) as estimates of the relative risk for HCC associated with statin use and their accompanying 95% confidence intervals (CIs) using conditional logistic regression. In addition to matching variables (age, sex, date of entry in the diabetes cohort), adjustment was carried out for possible confounders including race, hepatitis C virus (HCV) infection, alcoholic liver disease (ALD), cirrhosis, alcoholism, as well as combination of factors such as HCV plus ALD. We adjusted for the propensity score by adding propensity score as a covariate in the final model examining the association between statin and HCC, and by conducting a stratified analysis where the association between statin and HCC was examined separately in two groups of patients with high and low propensity to receive statin.
To examine potential effect modifiers, we conducted analyses stratified by groups with and without any liver disease, and by groups with and without cirrhosis.
The primary exposure was a filled prescription for any statin prior to the index HCC date. In order to reduce the possibility of confounding by indication, we also assessed any statin prescription, excluding those given in the one year directly preceding the index HCC date. To examine duration-effect relationship, we evaluated the effect of number of prescriptions fills/refills (0, 1–3, >3) and the cumulative duration of any statin prescriptions examined in the following categories: 0–180, 181–360, 361–540, 541–720, and > 720 days compared with the reference group (no statin prescription). We also examined these duration categories, excluding the one year preceding the index HCC date in each case.
We evaluated OR by the dose-duration and the total number of prescriptions of simvastatin, which was the most commonly used statin in the study population. We computed the average daily dose of each filled simvastatin prescription (strength multiplied by the average daily quantity), the dose duration for the prescription (average daily dose multiplied by days of use prior to HCC index date), and subsequently the cumulative dose duration prior to HCC index date (sum of dose durations over all simvastatin fills). For any two overlapping prescriptions, we added the dose duration of the first prescription to that of the second prescription if the overlap period was less than 30 days, assuming that patients may have used these prescriptions sequentially. For an overlap of greater than 30 days, we deducted the dose duration of the first prescription for the period following the start of the second prescription, assuming that the second prescription replaced the first one. We compared the distribution of dose-duration for simvastatin measured as quartiles between cases and controls.
We conducted a nested chart validation of the cases or control status as well as statin exposure at the Michael E. DeBakey Veterans Affairs Medical Center (MEDVAMC) in Houston, TX and the St. Louis VA in St. Louis, MO. We identified 104 patients (34 cases, 95 controls) who had at least one health care encounter at either VA between 1999 and 2002. A trained clinician blinded to the case/control and statin exposure status reviewed electronic medical records. HCC diagnosis was identified from progress notes, radiology, laboratory, or pathology reports. Statin use was defined as filled prescriptions within the study period. Underlying etiology (HCV, HBV, alcoholic liver disease, fatty liver) for HCC cases was also identified.
The study population consisted of 1303 cases with HCC diagnosed between January 1, 2001, and December 31, 2002, and 5212 controls without HCC matched on age, sex, and incidence density (ratio, 1:4). Mean age was 72 years. Most subjects (99%) were men, and 13% were African Americans. As expected from incidence density sampling and matching, cases and controls had no significant differences in age and duration between entry and index dates (Table 1). The case group had significantly higher proportions of liver disease, HCV treatment and HBV treatment; however both were quiet low. The case group had slightly lower proportions of patients with filled prescriptions for aspirin and ACE inhibitor.
During a mean exposure period of 2.4 years that preceded index dates for cases and controls, approximately 49% had a recorded filled statin prescription, 4% had nonstatin cholesterol-lowering drugs, and 9% had triglyceride-lowering drugs. In cases and controls, 41.6% of whom had had a prescription filled for at least 6 months, 32.3% had it filled for at least 12 months, and 22.7% had it filled for at least 18 months. The mean overall duration of any given statin prescription was 578.2 days (SD=334.4). Liver disease recorded as early as January 1, 1997, or within 4–6 years prior to the index date, was recorded in a larger population of cases (67.4%) than controls (8.7%).
The propensity score (to use statins) was generated from the model that contained coronary artery disease, myocardial ischemia, liver disease, obesity, and diabetic nephropathy as covariates, and was categorized into three equal tertiles: low (found in 12.6% of the study population), medium (39.1%), and high propensity (48.3%). A propensity score categorical variable (high propensity vs. other) was used a covariate in all the final models examining the statin-HCC associations. Further, the statin-HCC models were conducted as a sensitivity analysis in a group limited to patients with high propensity score.
We found high accuracy for our definitions of case-control as well as statin prescriptions. After reviewing a sample of medical records at the MEDVAMC, we found all 95 controls were correctly classified as not having HCC. All 57 patients who had filled prescription statins recorded in administrative records were correctly identified, and 65 patients classified as having had no statin prescriptions during CYs 1999–2002 in administrative data were also correctly identified (100% positive predictive value and 100% negative predictive value for administrative data). Of cases identified as HCC in the VA Medicare dataset, 89% were correctly identified as HCC by chart reviews. The underlying etiology in HCC cases was HCV (23%), alcoholic liver disease only (23%), fatty liver disease (38%), while there was no mention of risk factors or cirrhosis in 15%.
The proportions of patients with filled statin (or simvastatin) prescriptions were significantly lower among cases than controls (Table 1). The difference was mainly in the proportions of patients receiving more than 3 fills/refills, which was significantly lower in cases than that in controls (18.2% vs. 43.5%). Among those with at least one statin prescription, controls had an average of 9.9 (SD 7.1) fills/refills as compared to 9.4 (6.9) in cases. On the hand, there were no significant differences between cases and controls in the proportions of patients with filled prescriptions for triglyceride lowering and non-statin cholesterol reducing medications.
Having filled prescriptions for any statin was associated with a reduced risk of HCC in the primary analyses as well as in analyses stratified by the presence of known liver disease of any kind prior to HCC (Table 2). These significant associations were observed in unadjusted conditional logistic regression models and were also present but attenuated in the models adjusted for HCV, ALD, cirrhosis, alcoholism, race, HCV treatment, apirin/NSAID, ACE inhibitor, and propensity score. The relative odds of HCC were reduced in all groups (values ranged between .46 and .79). The lowest relative risk for all patients (adjusted and unadjusted) as well as for patients without liver disease (unadjusted) occurred when the duration of filled prescriptions was at least 18 months. When the analysis excludes statin prescriptions recorded within the one year preceding HCC diagnosis, the associations between statins and HCC were less strong.
Running the same analysis using simvastatin alone produced similar reductions in relative risk of HCC (relative risk values varied between 0.45 and 0.72) (Table 3). As with the analyses that used any statin, the reduced relative risk values in the adjusted population were less than those in the unadjusted population. In both analyses (Tables 2 and and3)3) but especially those for simvastatin, there was a trend toward lower OR (i.e., more risk reduction) with having >3 filled prescription than with 1–3 prescriptions as compared with no statins.
In an analysis restricted to patients with cirrhosis (data not shown), risk estimates for statins were of similar magnitude to those observed in the primary analysis. However, due to the small number of subjects (367 cases and only 86 controls), there was no statistical significance. In another analysis restricted to patients in the highest tertile for propensity score to use statins, risk estimates for statins also persisted in direction and significance; for example, the OR associated with having 1 to 3 filled statin prescriptions was 0.63 (0.33–1.18) and that of using more 3 filled statin prescriptions was 0.67 (0.49–0.93).
When the duration of prescription filling was divided into mutually exclusive periods, the reduced relative risk values for any statin or for simvastatin persisted (Table 4). There was a weak trend toward duration response relationship between statin and HCC risk. For any statin or for simvastatin alone, the association between the statin and HCC was weakest with the shortest duration of filled prescriptions. For example, for any statin the 0.59 relative risk at <6 months was the highest value of the periods measured; however, there was no significant or consistent trend of reduction with longer duration (0.40, 0.57, 0.42, 0.39) for 6–12, 12–18, 18–24, or >24 months, respectively.
When exposure to simvastatin was measured in quartiles of dose duration, the lowest dose duration produced weaker associations with HCC than higher dose duration. Some of the associations with the lowest dose duration were not significant in the analyses stratified by liver disease status. Likewise, the magnitude of inverse association with HCC was least for the quartile with the lowest dose and duration (OR, 0.52; CI = 0.41–0.65) than the higher quartiles; however, there was no discernible trend in the OR values of the second, third, and fourth quartiles that followed (0.49, 0.41, and 0.44, respectively).
Finally, we determined if the observed associations were unique to statins as compared to other lipid lowering medications. In the analyses shown in Table 5, there were no significant associations between HCC and either triglyceride or non statins cholesterol lowering drugs.
In this nested case control study of patients with diabetes, we found a significant inverse association between having statin prescriptions filled and the risk of developing HCC. There was a trend toward stronger risk reduction with longer and more frequent statin prescriptions. Reduced HCC risk was similar, whether the prescriptions were for simvastatin or any other statin dispensed. The risk reduction observed with statin ranged between 25% and 40%. These findings largely persisted in analyses that adjusted for propensity to use statins, excluded use within one year prior to HCC diagnosis, and analyses in patients with as well as without known liver disease or cirrhosis. There were no significant associations between HCC risk and filling prescriptions for non-statin cholesterol reducing or triglyceride lowering medications.
A number of potential mechanisms responsible for the effect of statins in reducing risk of HCC have been investigated, including inhibiting downstream products of the mevalonate pathway, primarily geranylgeranyl pyrophospate (GGPP) and farnesylpyrophosphate (FPP). Derivatives of the mevalonate pathway GGPP and FPP are important in the activation of a number of cellular proteins, including small guanosine-5′-triphosphate binding proteins, such as K-ras, and the Rho family 1,23,24. Statins interfere with the production of GGPP and FPP and disrupt the growth of malignant cells, eventually leading to apoptosis 4. It is also believed that p21 and p27 are cyclin-dependent kinase inhibitors that inhibit the growth of cancerous cells. Statins are also thought to inhibit the activation of the proteosome pathway, limiting the breakdown of both p21 and p27, allowing these molecules to exert their growth-inhibitory effects and in turn retard cancer cell mitosis 3,5.
Statins also have been shown to inhibit proliferation of stellate cells and their production of collagens in a dose dependent manner 25,26. Finally, statins may have an effect specific to HCV. Recently, it has been shown that HCV replication depends in part on geranylgeranylation of a host protein but high concentrations of statins can disrupt HCV RNA replication. The effect was presumably due to severe depletion of mevalonic acid, which in turn led to low cellular levels of GGPP, the substrate for the geranylgeranyl-transferase 13.
If physicians are less likely to prescribe statins because of their hepatotoxicity, patients with liver disease, especially those with severe liver disease, are less likely to be prescribed statins, which in a study like ours could lead to a spurious inverse association between statin use and HCC. We attempted to mitigate the effect of this potential bias even though we recognize that residual confounding could exist. We took several steps to avoid and evaluate for possible effect of confounding by indication because the population of patients who received statins might be different from the population that do not in ways that are not related to HCC. First, we examined the effect in stratified analyses in patients with and without liver disease. The findings indicate significant risk reduction in both groups but the magnitude of statin-related risk reduction is larger among patients with known liver disease than those without liver disease. Second, we estimated and adjusted the analyses for the propensity to use statins using propensity score method22, and found that the adjustment attenuated but did change the direction or remove the significance the observed inverse association between statin and HCC. Third, we conducted an analysis in which statin prescriptions recorded within one year preceding HCC were excluded assuming that in this time period liver disease is likely to be severe and overt; in these analyses, we found the associations between statins and HCC to be less strong in general. These lost significance in the adjusted analysis and in some of the stratified analyses of patients with and without liver disease. However, the number of cases and controls in some of these categories was relatively small. For example, only 55 cases and 606 controls were examined for at least 18 months in patients without liver disease. Lastly, we examined the association between HCC and filled prescriptions of nonstatin cholesterol-lowering drugs as well as triglyceride-lowering drugs. None of the exposures related to either of these two classes was significantly associated with HCC.
To account for possible changes in prescription as well as disease management practices, we employed sampling on incidence density where cases and controls are on contemporaneous as well as equal time durations. Although a randomized controlled trial would be the preferable study design to obtain unbiased estimates, conducting such a trial for HCC while possible is infeasible because of the small fraction of cases that occur annually. Such a trial also will not be justified without preliminary evidence from observational studies.
Apart from the novelty of findings, our study has several strengths including the large sample size, the relatively long period of potential exposure, accurate definitions of the case/control and statin prescription validated by chart reviews, and the close matching between cases and controls. Our choice of conducting a nested case-control study among patients with diabetes was related to the known higher likelihood of developing HCC, and to the higher likelihood of using statins to treat commonly found lipid abnormalities.
On the other hand, the study has limitations related to the observational retrospective nature of its design and the reliance on a convenience sample of patients identified in VA Medicare datasets. The relatively short duration of exposure (average 2.4 years) has limited the duration response analyses; however, the limited exposure period imposed by the convenience sample may not be interpreted as the same as short duration of statin use. It is possible that the short term patterns of statin use observed within the study frame are reflective of long term use that was not captured. The accuracy and completeness of potential confounders such as hepatitis C, alcoholic liver disease, and cirrhosis is unclear. These conditions are more likely to be identified and recorded in patients with HCC, and therefore the effect of misclassification is to reduce any observed association between statin and HCC. Approximately one third of HCC cases had none of these risk factors in the VA administrative date or Medicare claims data. A likely explanation is the absence of diagnostic codes for non-alcoholic fatty liver disease (NAFLD). Our chart review supports the overall the proportions of cases with HCV and alcoholic liver disease reported from administrative datasets but also indicates that NAFLD was found in approximately 38% of cases. Given that all patients in this cohort had established diabetes, one expects NAFLD to be disproportionately high in the study. Statin filled prescriptions were used to approximate actual intake of medications; the adherence to these medications in unknown. However, filled prescriptions correlate highly with self reported intake of medication in previous studies 27,28. Lastly, the findings were obtained from a predominantly men veteran population and therefore they may not be generalizable to women or non veterans.
In conclusion, this large nested matched case-control study in patients with diabetes provides the first indication of a cancer preventive effect for statins specific to HCC. These finding needs to be confirmed in future studies.
This research was supported in part by the Department of Veterans Affairs, Health Services Research and Development Service IIR 02-081, and by a grant from the American College of Gastroenterology. Dr. El-Serag is supported by NIH K24DK078154-03.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.