Statins and Prostate Cancer Risk: A Case-Control Study

Abstract

One in six men over age 60 years will be diagnosed with prostate cancer (1). Prostate cancer is the second leading cause of cancer deaths among US men (2). Therefore, the need to identify and to develop means to prevent this common malignancy is a high priority. Recent research on the chemopreventive potential of a group of cholesterol-lowering drugs, 3-hydroxy-3-methylglutaryl coenzyme A (HMG CoA) reductase inhibitors (statins), has shown intriguing results (3, 4). Statins lower serum cholesterol levels by inhibiting 3-hydroxy-3-methylglutaryl coenzyme A reductase, the rate-determining enzyme in the mevalonate pathway (4). In addition to their cholesterol-lowering ability, statins have been shown to inhibit prostate and breast cancer cell proliferation and to induce tumor-specific apoptosis (4–6) in in vitro studies.

While the mechanism remains unclear, clinical trials with cancer as a secondary endpoint have shown a nonsignificant inverse association between statin use and total cancer incidence (7, 8). Additionally, Bjerre and LeLorier's meta-analysis (9) of five trials of statins and cardiovascular disease, powered to identify a cancer-inducing effect, reported no association between statin use and risk of developing incident cancers.

Studies investigating statin use and incident cancer risk as the primary outcome have been largely observational and have yielded conflicting results. Graaf et al. (3) demonstrated an inverse association between long-term statin use and risk of incident cancer. Blais et al. (10) reported in their nested case-control study of 6,721 subjects that statin users were considerably less likely to be diagnosed with any cancer. More recently, in a case-control study using computerized medical records, Kaye and Jick (11) found no association between total cancer risk and statin use. One clinical trial of death due to hepatocellular carcinoma noted a suppression of tumor cell growth and extended survival time with use of pravastatin (12). None of these studies reported a significant association specifically for prostate cancer. However, Kaye and Jick (11) did note a significant increase in risk of prostate cancer with untreated hyperlipidemia. To our knowledge, the only study to specifically evaluate risk of prostate cancer reported a nonsignificant increase in prostate cancer risk with statin use (odds ratio (OR) = 1.2, 95 percent confidence interval (CI): 0.8, 1.7) (13).

Previous studies of statins and cancer risk are limited because of one or more of the following issues: 1) most trials addressed cancer as a secondary outcome (7, 8); 2) many studies addressing cancer as a primary outcome inadequately or crudely characterized statin use, such that it was not possible to account for variations of statin type (3, 10), dose (13), and duration of use (11); or 3) studies inadequately provided information regarding potential confounding factors, such as dietary intake (3, 7–13) and use of other lipid-lowering agents (9, 12, 13). In the current analyses, we address a number of these concerns by examining detailed prescription information in men enrolled in a case-control study of diet and prostate cancer risk. The study population consists of cases with biopsy-confirmed prostate cancer and clinic controls with a normal (<4 ng/ml) prostate-specific antigen test within the past 12 months.

MATERIALS AND METHODS

Subjects were recruited as part of a case-control study of diet and other risk factors for prostate cancer among veterans receiving care at the Portland Veterans Affairs Medical Center (PVAMC). All veterans referred to the PVAMC urology clinic for a prostate biopsy between December 2001 and August 2004 were eligible. Subjects who had a previous diagnosis of prostate cancer, ongoing treatment for another cancer, or documented cognitive impairment that would hinder their ability to accurately respond to a questionnaire or those participating in another research study were excluded. Over this time period, 782 men were referred for biopsy, 697 met our eligibility criteria, and 552 (79.2 percent of those eligible) were successfully contacted. Of those contacted, 349 (63.4 percent) consented to participate, 111 (20 percent) declined, 35 (6 percent) rescheduled, 40 (7 percent) missed either their interview or their biopsy appointment, and 15 (3 percent) were contacted but did not consent for other reasons. Of the 349 subjects who agreed to participate in our study, 298 (85 percent) underwent biopsy, and 51 (15 percent) were referred for biopsy but either chose not to have the biopsy or were determined, on the basis of ultrasound findings, declining prostate-specific antigen levels, or reduced life expectancy, to be an inappropriate candidate for biopsy. Of the 298 subjects who had a biopsy, 100 (34 percent) were diagnosed with prostate cancer, 178 (60 percent) had a negative biopsy, and 19 (6 percent) were diagnosed as having prostatic intraepithelial neoplasia, a potential precursor of prostate cancer.

Potential clinic controls were identified from the PVAMC primary care clinic. Controls were frequency age matched with men undergoing biopsy by 5-year age category. Eligible controls were over 50 years of age; had a normal serum prostate-specific antigen test result in the past 12 months, no history of prostate cancer or any other cancer, no history of prostatic intraepithelial neoplasia, and no history of dementia; and were not receiving treatment for prostate conditions. A total of 540 men with a normal prostate-specific antigen test were identified as possible clinic controls, 404 of whom met our eligibility criteria and were considered acceptable for participation by their primary care physician. Of these, 349 (86 percent of those eligible) subjects were successfully contacted, 202 (58 percent) were successfully interviewed, 61 (18 percent) declined to participate, 12 (3 percent) rescheduled, 61 (18 percent) missed their appointment, and 13 (4 percent) did not consent for other reasons.

Statin data were recorded from an electronic pharmacy database at the PVAMC. All prescriptions for statins from May 1997 to the date of enrollment were recorded. The date of initial prescription, the date discontinued or date of new prescription, the daily dose, and the number of pills per prescription were entered into our database. Prescription days were calculated as the date discontinued or the date a new prescription was initiated minus the date the previous prescription was initiated. In situations where no discontinued date was available, it was calculated conservatively as the date of initiation plus the documented number of pills per prescription. Refills were not assumed. Subjects were considered to have used statins if they had been prescribed a statin drug for at least 3 months prior to enrollment. Statin prescriptions were limited solely to those available on the PVAMC formulary from May 1997 through April 2004. These included lovastatin, simvastatin, atorvastatin, and fluvastatin. Although statins were initially included in the PVAMC formulary in 1990, electronic pharmacy records were not reliably accurate until 1997.

All subjects completed an interviewer-administered food frequency questionnaire (14) to assess usual dietary intake and a risk factor questionnaire to assess other potential risk factors for prostate cancer. The risk factor questionnaire included self-report of family history of cancer, nonsteroidal antiinflammatory drug (NSAID) use, alcohol consumption, smoking, and other comorbid conditions. This questionnaire does not capture subjects' physical activity. All subjects signed informed consent forms and Health Insurance Portability and Accountability Act (HIPAA) authorization forms approved by the PVAMC and Oregon Health & Sciences University institutional review boards prior to the interview. Prostate cancer cases completed the questionnaire prior to their biopsy and diagnosis with cancer.

All statistical analyses were performed using the SAS/PC program, version 8.2 (SAS Institute, Inc., Cary, North Carolina). Differences in covariate distribution among cases and controls were determined using a chi-square test for categorical variables and a t test for continuous variables. Tests were considered statistically significant at a p value (two sided) of less than 0.05. Odds ratios as estimates of the relative risk for prostate cancer associated with each level of statin use and 95 percent confidence intervals were obtained using unconditional logistic regression. All models were adjusted for age. Additional adjustment was carried out for factors considered to be associated with the likelihood of statin prescription and the risk of prostate cancer, including body mass index, calorie intake, fat and cholesterol intake, use of any other lipid-lowering drugs (including niacin and bile-acid binding resins), use of NSAIDs, diabetes, and race. Dietary intake variables were entered into the models as quartiles of intake based on the distribution of intake among the control subjects. Only mutually exclusive variables were maintained in the final expanded model. Models with and without adjustment variables are presented.

Stratified analyses were conducted to determine if the association between statins and prostate cancer varied by disease severity. Disease severity was determined according to the clinical staging of the American Joint Committee on Cancer (15) and the Gleason score of pathology. Cases were characterized by stage as having localized (stage T1–T2) or distant (stage T3–T4) disease and by grade as having less aggressive (Gleason score of ≤6) or more aggressive (Gleason score of ≥7) disease (16).

Statin duration and intensity variables were calculated for each subject by the total days used for each statin at each prescribed dose. The duration of statin use was calculated as the number of days each type of statin was used. Intensity of statin use was calculated in two ways, cumulative dose and average daily dose. The cumulative dose of individual statins was calculated as the total days each statin was used at each different dosage, multiplied by that dosage, and summed across all dosages. For example, the cumulative dose of lovastatin (mg) equals the sum (days used lovastatin at 40 mg/day × 40) + (days used lovastatin at 20 mg/day × 20). The average daily dose was calculated as the cumulative dose divided by the total days used. If a subject used more than one statin, use of each type of statin was calculated as described. Thus, one individual may contribute to the total dose of both lovastatin and simvastatin. Total statin use was calculated as any use of statins compared with no use. Statin variables for duration and intensity were categorized. For totals, categories represent nonusers and quartiles of users based on the distribution of use among controls. For individual statin types (lovastatin and simvastatin), variables were categorized as nonusers and users of doses equal to or below the median and above the median. The duration and intensity of statin use were modeled separately. Because the intensity and duration variables are interrelated, they are highly correlated, and thus it is not possible to model them together.

RESULTS

Records from 100 incident prostate cancer cases and 202 clinic controls are included in the analyses of prostate cancer risk. The median age was 65.5 years for prostate cancer cases and 65 years for clinic controls. Among prostate cancer cases, the majority had localized disease (92 percent with stage T1–T2c). However, Gleason scores of pathology ranged from 6 to 10, with 43 percent having a Gleason score of 6 and 57 percent having a Gleason score of greater than or equal to 7.

As shown in table 1, clinic controls had significantly higher total fat (p = 0.05) and calorie (p = 0.02) intake, as well as greater body mass index (p = 0.04) compared with prostate cancer cases. Prostate cancer cases were also significantly more likely to be African American, consistent with findings from previous studies (17, 18).

The prevalent use of any statin and specifically lovastatin and simvastatin was significantly greater (p = 0.01, 0.01, and 0.04, respectively) among clinic controls. Because atorvastatin and fluvastatin were new to the PVAMC formulary, use of these statins was dramatically lower than that of lovastatin and simvastatin. In further analysis, atorvastatin and fluvastatin are calculated as part of total statin use, but individual data are not provided.

Use of NSAIDs was not significantly different across the subject groups (table 2). There was not a significant difference in the number of subjects with diabetes, a family history of cancer, or a family history of prostate cancer. Consistent with higher use of statins, a higher percentage of the control subjects was diagnosed with hyperlipidemia compared with prostate cancer cases. Finally, because an elevated level of prostate-specific antigen is the primary reason that men are referred for prostate biopsy, it is not unexpected that the median prostate-specific antigen level for clinic controls (median = 1.0 (range = 0–4.0) ng/ml) was significantly lower than for prostate cancer cases (median = 6.0 (range = 0.5–66.0) ng/ml).

After adjustment for age, men with any recorded statin use had a 50 percent lower risk of prostate cancer. After adjustment for age, race, body mass index, NSAID use, diabetes, total calories, and use of other lipid-lowering drugs, men with any recorded statin use had a 65 percent reduction in risk of prostate cancer compared with nonusers. When the data were stratified by Gleason score, the percentage of reduction in risk of less aggressive (Gleason score 6) disease was no longer significant, but the percentage of reduction in risk of more aggressive (Gleason score ≥7) disease was 76 percent (table 3). When statin use was categorized by duration, cumulative dose, and average daily dose, there was a significant trend toward decreasing prostate cancer risk with increasing duration of use and cumulative dose of statin, and there was a significant, although less pronounced, decrease in risk with increasing average daily dose of statin (table 3). These trends remained largely consistent after the analyses were stratified by type of statin. There was a reduction in prostate cancer risk with an increase in the duration and cumulative dose of both lovastatin and simvastatin (table 3). However, while the effect appeared to be duration related for simvastatin, lovastatin appeared to have a threshold effect. This was supported by the average daily dose data. The trend for reduction in risk with increasing average daily dose of simvastatin remained consistent with the overall trend for average daily dose. However, while the odds ratios for lovastatin remained below one, risk reduction was not consistently enhanced with increasing dose. This lack of a trend in risk reduction with increasing duration and dose of lovastatin use may be due to the relatively small amount of variation in dosage.

Based on the magnitude (odds ratio) and the statistical significance (p value) of the associations, the duration of use is a stronger predictor of risk reduction than is intensity as determined by average daily dose. However, the magnitude and significance of the association are greatest when intensity is modeled as cumulative dose (a combination of both duration and dose), suggesting that both duration and dose are important contributors to the association between statins and prostate cancer. When both duration and intensity are modeled together, neither is significant; however, this is not unexpected as these variables are not independent of each other.

Because of the small number of subjects with recorded use of nonstatin lipid-lowering agents (i.e., niacin, gemfibrozil, cholestyramine, fenofibrate, and so on), we were unable to determine with certainty if these agents were also independent predictors of decreased risk. However, adjustment for the reported use of any other lipid-lowering drugs in the full model did not significantly alter the odds ratio for statins (data not shown).

DISCUSSION

The mechanism whereby statins may inhibit cell proliferation and induce apoptosis in prostate cancer is not fully understood. Yet, a number of potential mechanisms are currently under investigation, including but not limited to the following: 1) inhibiting downstream products of the mevalonate pathway, primarily geranylgeranyl pyrophospate (GGPP) and farnesylpyrophosphate (FPP); 2) inhibiting breakdown of M_r 21,000 protein (p21) and/or M_r 27,000 protein (p27) cyclin-dependent kinase inhibitors; or 3) reducing cholesterol availability to the lipid membrane raft and thus decreasing raft-related signaling.

Briefly, 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, N-ras, and the Rho family (5, 19, 20). Statins interfere with the production of GGPP and FPP and disrupt the proper function and growth of malignant cells, eventually leading to apoptosis (4). p21 and p27 are cyclin-dependent kinase inhibitors that are known to inhibit the growth of cancerous cells. Investigators have reported that statins 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 to retard cancer cell mitosis (6, 21). Finally, lipid rafts are cholesterol membrane domains rich in proteins that mediate signal transduction processes, including epidermal growth factor signaling (22). In prostate cancer, researchers have demonstrated that treatment of LNCaP cells with filipin, a cholesterol-binding agent, resulted in loss of the lipid raft integrity and apoptotic death (23). In addition, this same group reported that treatment of mice bearing LNCaP tumor xenografts with simvastatin results in a decrease in the cholesterol content of lipid rafts and a subsequent dose- and time-dependent apoptosis (24).

We report a strong inverse association between any statin use and prostate cancer risk, and we demonstrate that this risk reduction is greatest for risk of clinically more aggressive (Gleason ≥7) disease. While this finding is consistent with the direction of the association reported in several other observational studies (3, 10), the magnitude of the association is dramatically stronger and is in the opposite direction of the association reported in a study by Coogan et al. (13). There are several differences between our study and those previously published. One rather startling difference is the prevalence of recorded use of statin drugs. Overall, 49 percent of our clinic controls and 34 percent of prostate cancer cases had been prescribed statins at least once for a minimum of 3 months. In most other published observational studies of statins, prevalence of use has varied from 6 percent to 10 percent (3, 10, 11, 13). A number of reasons may account for this large difference in prevalent statin use, including time period of the data collection, data source, and study location and population. In three of the four previous studies, data collection started in the mid-1980s and ended between 1994 and 2001 (3, 10, 13). In the fourth study, data collection ended in 2002, but statin use was counted only if it began no later than 2001 (11).

Statins entered the US market in 1987. However, recorded use did not dramatically increase until the late 1990s and early 2000s (25). Data from the National Ambulatory Medical Care Survey clearly demonstrate the rapid increase in prescriptions for cholesterol-lowering drugs, thought to be primarily statins, since the early 1990s. Since 1995, among men aged 65 or more years, the number of hyperlipidemia drugs ordered per 100 population has increased from 25.1 in 1995–1996 to 79.5 in 2001–2002 (25, 26). Similar to our study population, this survey represents only the portion of the population that seeks out medical care. Hence, the estimate of statin use is likely higher than seen in the overall population. It is possible that our higher prevalence of recorded use reflects changes in prescription practices that took place prior to and during the 1997–2004 time period, changes that may not have been reflected in studies that ended in 2001.

It must also be noted that our population consists solely of veterans with active access to health care. It is possible that these men are more likely to receive regular lipid screening and to be prescribed lipid-lowering medications than is the general population. Finally, the population in this study is largely older men, with only 8 percent of our subjects aged less than 55 years. As demonstrated by data from the National Ambulatory Medical Care Survey, prescriptions for lipid-lowering medications increase with age. Other studies have included subjects ranging in age from 21 to 89 years (7–9, 11–13).

Because statin use data were obtained from an historical pharmacy database, there is little chance of differential reporting bias or exposure misclassification. Our prescription data were obtained from a data warehouse that includes pharmacy records for all Veterans Affairs (VA) clinics associated with the participating PVAMC. As described below, all but five men in this study received primary care from one of the VA clinics. It is possible that some men obtained statin prescriptions from outside the VA system; these prescriptions would not be captured in our analyses. However, most men who receive care within the system prefer to obtain their prescription medications through the VA system to avoid out-of-pocket costs.

While our data collection methods provide a detailed characterization of statin use in this population, there remain a number of limitations inherent to the study design chosen. As with all case-control studies, there is the concern that exposure data are being ascertained at the wrong time in relation to disease development, or that the exposure of interest is simply a marker for some other factor that impacts disease development. For example, as suggested by data from Kaye and Jick (11), uncontrolled hyperlipidemia may increase prostate cancer risk, and statin use may simply be a marker of controlled hyperlipidemia. However, if controlling hyperlipidemia were the primary protective effect, one would expect to see similar findings in studies of other lipid-lowering agents. Blais et al. (10) addressed this possibility by comparing cancer rates among subjects using statins and those using bile acid-binding resins, and they demonstrated a significant reduction in cancer risk among statin users compared with bile acid-binding resin users (rate ratio = 0.72, 95 percent CI: 0.57, 0.92). While there were too few men in our study using other lipid-lowering agents to conduct a similar analysis, the inclusion of other lipid-lowering drugs as an adjustment variable did not alter our findings. Additionally, although cases and controls differed with regard to a number of other health characteristics commonly related to hyperlipidemia (body mass index, diabetes, and total fat intake), adjustment for these factors also did not alter our findings.

Another limitation in our study is the possibility of selection bias. Of eligible controls, 42 percent refused to participate or did not show up for scheduled appointments. If men that agreed to participate were healthier than those who refused, our results may be biased. Again, adjustment for other factors that may reflect health behavior did not significantly modify the associations presented. Alternatively, 36.6 percent of the eligible biopsy subjects that were contacted refused or were unable to participate. If the distribution of prostate cancer diagnosis and severity differed among these men, the association between statin use and disease severity may be distorted. The catchment area for the cases and controls also differed somewhat. More clinic controls reported residing in the area directly surrounding the PVAMC (45 percent of controls vs. 33 percent of cases). Men outside the metropolitan area may receive their primary care from local VA clinics. If statin prescription patterns differ among clinic locations, this may bias our results. Yet, it is unlikely that there was a great deal of difference, as VA providers have access to the same VA practice guidelines for prescribing statins. Finally, while all clinic controls had recorded primary care visits, five cases had no recorded primary care visit, suggesting that they may receive primary care outside the VA system and that we may not have record of their statin use. We repeated our analyses, dropping these five cases. The results did not differ substantially from those presented (adjusted OR for any statin use = 0.39, 95 percent CI: 0.22, 0.70) (data not shown).

A strength of our clinic-based, case-control design is the reduced likelihood of subject misclassification. All cases had biopsy-confirmed cancer, and controls were defined not solely as those without a cancer diagnosis but were further narrowed to men with a negative prostate-specific antigen screening test result. There remains the possibility that some controls were misclassified using the prostate-specific antigen cutoff of 4.0 ng/ml, yet when our data were reanalyzed using a cutoff of 2.5 ng/ml as a normal prostate-specific antigen test result, only 27 control subjects were dropped, and our results were not substantially different from those presented (adjusted OR for any statin use = 0.41, 95 percent CI: 0.22, 0.74).

These analyses represent results from one of only a handful of studies of statin use and prostate cancer risk. We report a significant reduction in risk of prostate cancer and primarily more aggressive disease with statin use. With prostate cancer as our endpoint, we were able to collect well-characterized data on statin type, dose, and duration of use. While intriguing, these results must be interpreted cautiously and with consideration of the weaknesses inherent to a case-control design. However, if our results are confirmed in a larger prospective study, they may provide the evidence necessary to consider the use of statin drugs in prostate cancer prevention.

Supported in part by US Public Health Service grant 5 M01 RR000334, this paper is the result of work also supported with resources and the use of facilities at the Portland Veterans Affairs Medical Center, Portland, Oregon.

Presented at the annual meeting of the American Society of Clinical Oncology, New Orleans, Louisiana, June 5–8, 2004.

Conflict of interest: none declared.

References