SEARCH

SEARCH BY CITATION

Keywords:

  • prostate cancer neoplasms;
  • statins;
  • mortality;
  • cholesterol

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

BACKGROUND:

Statins are some of the most commonly prescribed medications in medical practice, and prostate cancer is the most common malignancy among men. Although there has been no consistent evidence that statins affect cancer incidence, including prostate cancer, several reports suggest they may decrease the rate of advanced prostate cancer. However, no study to date has specifically examined statin use and prostate cancer mortality. The authors conducted this population-based case-control investigation to examine this association.

METHODS:

This was a matched case-control study. Cases were residents of New Jersey ages 55 to 79 years who died from prostate cancer between 1997 and 2000. The cases were matched individually to population-based controls by 5-year age group and race. Medication data were obtained identically for cases and controls from blinded medical chart review. Conditional logistic regression was used to adjust for confounders.

RESULTS:

In total, 718 cases were identified, and cooperation was obtained from 77% of their spouses (N = 553). After a review of medical records, 387 men were eligible, and 380 were matched to a control. The unadjusted odds ratio was 0.49 (95% confidence interval, 0.34-0.70) and decreased to 0.37 (P < .0001) after adjusting for education, waist size, body mass index, comorbidities, and antihypertensive medication. There was little difference between lipophilic and hydrophilic statins, but more risk reduction was noted for high-potency statins (73%; P < .0001) compared with low-potency statins (31%; P = .32).

CONCLUSIONS:

Statin use was associated with substantial protection against prostate cancer death, adding to the epidemiologic evidence for an inhibitory effect on prostate cancer. Cancer 2012. © 2011 American Cancer Society.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

Statins have been sold in the United States since 1987 and have become the dominant drugs used to treat hyperlipidemia. By 1999 to 2002, nearly 25% of adults aged >60 years were using statins.1 Because of early interest in a possible association between low serum cholesterol levels and cancer2 and the pleiotropic effects of these drugs, several meta-analyses have sought evidence that statin use may affect cancer incidence or overall cancer mortality.3-6 However, those meta-analyses did not identify an association between statin use, either positively or negatively, and the incidence of cancer.

The data on prostate cancer are similar to those for overall cancer.7 However, several studies that have used advanced or aggressive prostate cancer as an endpoint have demonstrated a protective association of statin use.7-11 A recent report indicated that statin medication is associated with a dose-dependent reduction in the risk of biochemical recurrence.12 Those studies were summarized in 2 reviews.13, 14 Because those investigations included relatively few prostate cancer deaths, we examined this issue using data from a previously reported case-control study that we conducted to evaluate prostate cancer screening in New Jersey.15 The value of this database is supported by an earlier analysis15 that correctly predicted the finding of the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial that prostate-specific antigen (PSA) screening provides no benefit from mortality in US patients.16

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

Study Participants

We identified potential cases from New Jersey vital records. Only white and black men who died from prostate cancer, as defined by the underlying cause of death on the death certificate, between 1997 and 2000 at ages 55 to 79 years were eligible. To prevent possible misclassification of nonprostate cancer deaths, we excluded 10% of potential cases in whom medical records did not document metastatic prostate cancer. We required cases to have been married at the time of death to increase the probability that there would be a knowledgeable, surviving informant to assist in identifying medical care providers. To prevent a selection bias and possible confounding, controls also had to have been married. The overall response rate for cases was 77% (N = 553), and 387 (70%) were eligible after the interview and medical records review based on confirmation of age, diagnosis, and residence and on successful acquisition of usable exposure information. The control response rate was 57% (N = 610), and 442 were eligible after review. The chief reasons for ineligibility for both cases and controls were age outside our criteria, not being married, and inability to obtain medical records. There were no systematic differences in reasons for ineligibility between cases and controls. We were able to match 380 cases with controls.

For each case, we matched 1 control participant by age (in the same 5-year age group), race as designated by the interviewee, and for the amount of available time of exposure, as indexed by the date of death of the case going back to 1989. Potential controls ages 55 to 64 years who were living in New Jersey were identified by Northeast Research, Inc. or The Watsroom, Inc. using random digit dialing methods.17 Potential controls ages 65 to 79 years were identified by Westat, Inc. from New Jersey Medicare tapes. Controls with prostate cancer (10.5%) were eligible as long as there was no evidence of advanced disease at the time the matched case died and that the diagnosis came after the date of suspicion of cancer in the case. The study was approved by Institutional Review Boards from the University of Medicine and Dentistry of New Jersey-Robert Wood Johnson Medical School and the New Jersey Department of Health and Senior Services. Informed consent was obtained from control subjects and from spouses of deceased cases.

Data Measurements

Data on demographics, education, occupation, and personal measurements were obtained from spousal report for cases and directly from controls. Medical records were reviewed from all health care providers known to have cared for the individual from 1989 onward. This review was performed to obtain information on comorbidity, clinical stage, Gleason score, pathology results, and all PSA levels determined before diagnosis. All chronic medications were abstracted and recorded identically for both cases and controls from blinded medical chart review. Because of the difficulty in obtaining accurate dates for chronic medication use, we recorded the use of a medication as “ever used.” All listings of medications by study data abstracters were further reviewed by a research nurse knowledgeable about the major classes of prescription medications. Information on the cancer was supplemented when necessary from the New Jersey Cancer Registry. This registry is part of the National Cancer Institute Surveillance, Epidemiology, and End Results (SEER) Program and was established in 1979. New Jersey State Law mandates that hospitals, physicians, dentists, and clinical laboratories must report all new cancer cases within 6 months to the New Jersey Cancer Registry.18 Residents of New Jersey diagnosed with cancer outside the state are identified through information provided by agreements with neighboring states. Existing data in the registry are verified periodically using information from hospitals and physicians, death certificates, motor vehicle and income tax records, as well as federal databases, such as the National Death Index.

Analysis

Statin use was the principal exposure analyzed. We also examined other medications, such as hypertensive medications, nonsteroidal anti-inflammatory drugs (NSAIDS), medications used for coronary heart disease (other than those classified as antihypertensive), lipid-lowering agents other than statins, diabetic medications, and medications used for erectile dysfunction. We did not capture specific dates or cumulative doses of medications. We further classified statins as high or low potency and as hydrophilic or lipophilic.19 We explored possible effect modification of hypertensive medication on statins with an interaction term in the logistic model. Education was categorized into less than high school graduate, high school graduate with or without some college, college graduate, and graduate or professional degree. The number of comorbidities was tabulated as none, 1 to 2, 3 to 5, 6 to 10, and >10. A comorbidity was counted if present before the year of prostate cancer diagnosis. We calculated body mass index (BMI) as weight in kg divided by the square of height in meters and ordered into <25 kg/m2, 25 to 29.9 kg/m2, 30 to 34.9 kg/m2, and ≥35 kg/m2 categories, representing normal weight, overweight, mild obesity, and morbid obesity, respectively. Waist size in inches was divided into categories of ≤32 inches, 32.1 to 36 inches, 36.1 to 40 inches, and >40 inches. These measurements were obtained for a period before the decedent was clinically ill and at a similar time period for the matched control.

Descriptive analysis was performed with contingency tables and chi-square tests for categorical variables. Means were obtained and compared for continuous variables using Student t tests. All P values ≤ .05 were considered statistically significant, and 95% confidence intervals (CIs) were used.

In the matched analysis, we used conditional logistic regression with the SAS statistical software package (SAS Institute, Inc., Cary, NC) with 1:1 matching for race, age, and amount of observation time (the amount of time before the diagnosis or suspicion of diagnosis in the case and an equivalent amount of time for the matched control). We first obtained univariate odds ratios (ORs) for statin use and then adjusted for education level, BMI, waist size, and the number of comorbidities. Our final model also was adjusted for antihypertensive use (ever/never). It is believed that obesity is associated with more aggressive prostate cancer20, 21 and may confound the association between statin use and prostate cancer mortality. Similarly, use of an antihypertensive is an indirect marker of the metabolic syndrome, which has been linked to prostate cancer mortality.21 In another report on the same study subjects,15 we observed no correlation between PSA screening and prostate cancer mortality; therefore, screening status was not considered as a confounder.

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

Table 1 indicates that the men who died from prostate cancer and the controls were well matched for age and race. The slight discrepancy in race between cases and control was because of 1 mismatched pair, which was excluded from further analysis. Controls were more educated than cases: more controls had achieved a college or graduate degree, and more cases had not graduated from high school. The median number of comorbidities was 3 for cases and 2 for controls. Controls who agreed to participate were 13.8% black and had a median age of approximately 72 years compared with those who refused to participate, who were 16.4% black and had a median age of 73 years. Table 2 lists the Gleason scores and stages at diagnosis for the cases. Forty percent of the cases who died presented with distant disease at diagnosis. The median length of survival from suspicion of cancer to the date of death was 2.9 years for cases with distant disease compared with 6.2 years for cases with nondistant disease. Forty controls (10.5%) had a history of prostate cancer that was identified subsequent to the date of suspicion of prostate cancer in the matched case.

Table 1. Characteristics of Cases and Controls
 No. (%)
CharacteristicCasesControls
  • Abbreviations: SD, standard deviation.

  • a

    One case-control pair was mismatched for race

Race  
 Black39 (10)40 (11)a
 White341 (90)340 (89)
Education  
 <High school82 (22)50 (13)
 High school graduate195 (51)193 (51)
 College graduate57 (15)65 (17)
 Graduate or professional school41 (11)70 (18)
 Not available5 (1)2 (1)
No. of comorbidities  
 052 (14)105 (28)
 1-295 (25)101 (27)
 3-5114 (30)89 (23)
 6-1081 (21)60 (16)
 >1038 (10)25 (7)
Age at diagnosis: Mean±SD, y67.2 ± 5.766.5 ± 5.9
Table 2. Gleason Scores and Clinical Stage of Cases
VariableNo. (%)
  1. Abbreviations: LN, lymph node.

Gleason score 
 ≤546 (12.1)
 648 (12.6)
 791 (23.9)
 883 (21.8)
 975 (19.7)
 1012 (3.2)
 Not specified25 (6.6)
Clinical stage 
 Localized: T1/T2, negative LN status144 (37.9)
 Regional: T3/T4 or positive LN status61 (16.1)
 Distant153 (40.3)
 Not assessed22 (5.8)

In unadjusted analysis (Table 3), the odds that a man dying from prostate cancer was exposed to a statin was half the odds of a control participant (OR, 0.49; 95% CI, 0.34-0.70). Although a lower education level, more comorbidities, and larger waist size all predicted an increased risk of prostate cancer death, their inclusion in the model did not significantly change the risk associated with statin use. After adjusting for exposure to any antihypertensive medication, the odds of dying from prostate cancer associated with statin use decreased to 0.37 (95% CI, 0.23-0.60). Antihypertensive agents, which were used by approximately 66% of cases, were associated with an increased OR of 2.9; (95% CI, 1.87-4.49) for prostate cancer death. We explored a possible interaction of antihypertensive medication with that of statins. The interaction term had borderline statistical significance and suggested that most of the protection offered by statins was for men who also were exposed to an antihypertensive medication (data not shown).

Table 3. Effect of Statin and Antihypertensive Use on Prostate Cancer Mortality
ModelOR95% CIP
  • Abbreviations: CI, confidence interval; OR, odds ratio.

  • a

    Adjusted for education level, body mass index, waist size, and number of comorbidities and matched for race and age.

  • b

    Adjusted for the same variables plus antihypertensive medication use.

Unadjusted model   
 Statin use0.490.34-0.70<.0001
Adjusted modela   
 Statin use0.450.29-0.71.0006
Full modelb   
 Statin use0.370.23-0.60<.0001
 Antihypertensive use2.901.87-4.49<.0001

An analysis of statin type suggested that there was little difference between lipophilic and hydrophilic statin medications (Table 4). However, there was a trend based on potency in which high-potency statins were associated with a 73% risk reduction (P < .0001), whereas the risk reduction was only 31% for low-potency statins (P = .32).

Table 4. Effect of Statin Type on Prostate Cancer Mortality
Statin TypeORa95% CIP
  • Abbreviations: CI, confidence interval; OR, odds ratio.

  • a

    Adjusted for education, body mass index, waist size, and number of comorbidities and matched for race and age. The reference category was “no statin” use for both comparisons.

  • b

    Hydrophilic statins include pravastatin, atorvastatin, and fluvastatin. Lipophilic statins include lovastatin, simvastatin, and cerivastatin.

  • c

    May also have been exposed to a hydrophilic statin.

  • d

    May also have been exposed to a weak statin. High-potency statins include cerivastatin, atorvastatin, and simvastatin. Low-potency statins include pravastatin, lovastatin, and fluvastatin.

Hydrophilic statinb0.410.19-0.86.02
Lipophilic statinc0.350.20-0.61.0002
High-potency statind0.270.15-0.48<.0001
Low-potency statin0.690.33-1.45.32

We also investigated whether calendar year had any effect on statins, because the more potent statins were prescribed more often later during the period under study. We categorized calendar years into 3 categories: 1989 to 1992, 1993 to 1995, and 1996 to 1999, with roughly similar number of men in each category. After adding interactions of these periods with any statin use, the inverse association of exposure to statins with prostate cancer mortality was weakest in the earliest years (OR, 0.887; P = .75), strongest in the latest years (OR, 0.11; Pinteraction = .0009), and intermediate in the middle years (OR, 0.32; Pinteraction = .07). When we analyzed exposure only to high-potency statins, this smoothed out the trend considerably, and these interactions were no longer significant. Because PSA testing may be a measurement of compliance and possibly of adherence, we added an interaction term for statin use with PSA screening (ever vs never) in the model, but this was not significant (Pinteraction = .68).

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

Statin use in this population of New Jersey men was associated inversely with prostate cancer death. This association was equivalent to a 50% reduction in prostate cancer deaths. To our knowledge, this is the first population-based study to demonstrate statin protection specifically from prostate cancer death, and it adds to a growing body of literature suggesting that statins protect against advanced prostate cancer. The protective association in this study persisted regardless of adjustment for possible confounders, such as markers of obesity, educational level, and the number of chronic comorbidities. Adjustment for the use of any antihypertensive medication further increased the apparent protective effect of statins.

We identified 7 previous observational studies that examined statin use and advanced prostate cancer (including some admixture of fatal cases in 2 studies). Of these, 4 were cohorts reporting hazard ratios ranging from 0.26 to 0.93, 2 of which were statistically significant, for the use of a statin for ≥5 years.7, 8, 10, 22 The other 3 studies were case-control designs and reported ORs for “ever use” of statins from 0.75 to 0.9: one of those ORs was statistically significant but all 3 studies had lower confidence limits that were consistent with a substantial clinical effect of ≤0.6.9, 23, 24

Statins are given most commonly for hyperlipidemic conditions, especially in patients who have other cardiovascular risk factors, such as hypertension, obesity, and diabetes, which are components of the metabolic syndrome.25 There is evidence that obesity is associated with advanced and fatal prostate cancer, although probably not with prostate cancer incidence.26 For these reasons, if statins have no biologic activity against the development of advanced or fatal prostate cancer, then a positive association with prostate cancer mortality may be expected based on statins being prescribed for conditions associated with this outcome. Thus, it is not surprising that the estimate for statin protection becomes stronger after adjusting for antihypertensive use, an indirect marker for the metabolic syndrome. These results also are consistent with the recent study by Farwell et al indicating that statin users were 60% less likely to be diagnosed with high-grade prostate cancer than antihypertensive users.27

Our analysis suggests a possible mechanism that may be responsible for this risk reduction. We did not observe a significant difference between lipophilic and hydrophilic statins, but there was a significant difference based on potency. High-potency statins were associated with a 73% risk reduction compared with no statins and had nearly 2.5 times the protection observed with low-potency statins. This points to cholesterol-lowering as a possible mechanism. Decreasing cholesterol changes the lipid composition in cell membranes that has been linked to changes in intracellular signaling, including that of the Akt pathway.28, 29 The v-Akt murine thymoma viral oncogene homolog-1/mammalian target of rapamycin (Akt-mTOR) signal-transduction pathway has been implicated in increasing the response in prostate cancer cells to hypoxia-inducible factor-1 (HIF-1) that is central to cancer cell survival in a hypoxic environment.28

Another possible mechanism reflective of cholesterol-lowering potency is that of decreasing mevalonate, which is an important precursor for isoprenoids. Just like for cholesterol, high-potency statins cause large decreases in mevalonate and, in turn, isoprenoids. Isoprenoids facilitate intracellular signaling by Ras and Rho, which promote both cell survival and cell proliferation.29, 30 Statins also may reduce the risk of cancer through their anti-inflammatory,31 proapoptotic,32 and antiangiogenic effects.31

There are some limitations to this investigation. Although medical records usually capture medications that are used regularly and require a prescription, they do not systematically capture all comorbidities. For example, it was difficult to identify documentation of “hyperlipidemia” or “dyslipidemia” even when an individual was on a statin. For this reason, our estimate of the number of comorbidities is probably a lower boundary of the actual total. However, the number of comorbidities did not change the association of statin use with prostate cancer mortality.

Because we could not obtain accurate measurements of the cumulative dose or starting dates of medications, we used “any exposure” as a positive history for medication use. This prevented us from examining a dose effect other than what we could characterize by the potency of the statin. Similarly, we could not pinpoint the timing of the initial prescription, so we were limited in determining whether the potential biologic effect was on the development of the cancer or on its progression. We do note that most previous studies have not demonstrated a correlation with prostate cancer incidence. Also, the greatest opportunity for exposure to statins in our patients was before the diagnosis in the case (and an equivalent amount of time for the matched control).

Case-control studies are more prone to selection bias, so we carefully compared our controls with the population of New Jersey men with respect to PSA screening. Controls had a PSA screening rate that was nearly identical to the 2001 Behavioral Risk Factor Surveillance Survey for New Jersey, which supports the representativeness of the population.33 Controls were chosen to be representative of the source population that gave rise to the cases: resident New Jersey men who were capable of developing lethal prostate cancer. It is expected that controls would be a mix of individuals without prostate cancer and those with nonlethal prostate cancer (at the time of death of the matched case). Thus, controls were allowed to have nonadvanced prostate cancer, but we limited eligibility so that a control with prostate cancer had to have the diagnosis after his matching case was diagnosed. This was necessary to prevent a bias toward protection from PSA testing in the PSA screening study. However, we do not believe this possible source of selection bias was significant for statin exposure, because our controls were very similar to the population of New Jersey men with respect to prostate cancer prevalence and PSA screening for this calendar period. Exposure to medications for hypertension, a comorbidity closely related to that of hyperlipidemia for which statins are indicated, had a strong and opposite correlation to the protective association observed for statins. If our findings are a result of selection bias, then it is unlikely that these 2 groups of medicines would have opposite associations with prostate cancer mortality. This design, however, does limit our ability to determine whether the observed protective association is from the prevention of more aggressive prostate cancer or the prevention of progression of existing prostate cancer.

There is also the possibility of a selection bias from competing risk because of cardiovascular death. If men on statins are more likely to die from cardiovascular death than those not on statins, then it is possible that this could result in an apparent lower rate of death from prostate cancer for those men on statins apart from any biologic action. We believe that, if present, this bias is not significant, because adjusting for comorbidity did not change the estimates; and; as noted above, individuals with cardiovascular risk factors are more likely to have high-grade and fatal prostate cancers. Moreover, the same trend would be expected with antihypertensive medications, which is not the case.

Information bias is another concern for observational studies. We matched on the opportunity for exposure for each case and control dyad—1989 through the time of death of the case—but we cannot state with certainty how comparable the documentation of statin use may be between cases and controls. Cases were more likely to have contact with health care providers in their last few years, so this may result in more thorough documentation of medication use (conservative bias). Alternatively, cases may not have had statins prescribed, even if indicated, in their last few years because of competing concerns with their cancer (bias toward protective association). However, the largest time interval for potential exposure for both the case and the matched control was before the diagnosis of prostate cancer in the case, which served to decrease this source of bias. There is certainly some nondifferential misclassification based on incomplete information from medical records and the level of adherence or true exposure, but this would bias the results to the null.

Our response rate for controls was only 57%, but they were generally comparable to the source population. Cases had a good response rate, but we limited their eligibility to those who had a surviving spouse who could provide enough information for us to identify the decedent's source of health care and whose medical records provided proof of metastatic, symptomatic prostate cancer. This may have limited eligibility, but it ensured that cases had prostate cancer as the principal cause of death. By using only married cases and, for comparability, married controls, we may have decreased our generalizability; married men may be more adherent to medications than nonmarried men; thus, our result may be less in this population. Finally, although we adjusted for several potential confounders and matched for race, age, and potential time of exposure, we are aware that, similar to any observational study, this does not ensure the absence of unquantified biases and unknown confounding.

The strengths of this study are that it is population-based and that chart review verified that all cases died from prostate cancer. An advantage of using disease-specific mortality as an endpoint is that it is not confounded by screening, which may be a problem in studies of cancer incidence. The key predictor variable, statin use, although not quantitative, nevertheless was documented similarly for both cases and controls from a review of medical records. The greater protection afforded by the more potent statins suggests a specific drug effect.

In summary, we describe an inverse association of prostate cancer death and the use of statin medications. This association is consistent with the protective effect cited in previous reports that have focused on advanced prostate cancer. This correlation is more striking with high-potency statins. In view of the good safety record of this class of drugs and the shared risk factors for cardiovascular disease and aggressive prostate cancer, we believe that it is now time to directly test the value of statins for inhibiting progression of prostate cancer in a randomized clinical trial.

FUNDING SOURCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES

Stephen Marcella, George Rhoads, and Jeffrey Carson received financial support from the National Cancer Instituted Grant: NCI-R01 CA71734-01A1. Stephen Marcella, George Rhoads, and Pamela Ohman-Strickland are members of the Cancer Institute of NJ supported by Comprehensive Cancer Center Support Grant: P30CA072720.

CONFLICT OF INTEREST DISCLOSURES

The authors made no disclosures.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SOURCES
  8. REFERENCES