Statins and Cancer Risk:  A Meta-analysis FREE

Retrospective analyses suggest that statins reduce the risk of developing cancer.^1- 7 This includes marked reductions in colon, breast, lung, and prostate cancers.^1- 7 Given the inherent biases and weaknesses associated with retrospective analyses, these studies are hypothesis-generating but cannot prove causality. Pharmacologically, statins may reduce cancer via reductions in inflammation, neovascular formation, and cell proliferation but can also inhibit selenoprotein synthesis and decrease natural killer cell function, which might enhance cancer risk.^8- 11

Statins have been studied in numerous large-scale, randomized, active- or placebo-controlled trials for primary and secondary prevention of coronary artery disease. In these trials, statins reduced the risk of a first myocardial infarction and overall mortality.^12- 38 With long-term follow-up and collection of cancer data in a majority of studies, insight into the risk of cancer among statin-naive persons and statin users can be derived. Therefore, we performed meta-analyses of all randomized controlled statin trials evaluating cancer.

We conducted a systematic literature search of MEDLINE from 1966 through July 2005, EMBASE from 1990 through July 2005, CINAHL from 1982 through July 2005, Web of Science from 1994 through July 2005, CANCERLIT from 1975 through July 2005, and the Cochrane Systematic Review Database to identify randomized clinical trials of statin use with a primary or secondary end point of cancer diagnosis or cancer death. A search strategy using the Medical Subject Heading and text key words HMG-CoA reductase inhibitor, HMG-CoA RI, statin, pravastatin, simvastatin, lovastatin, atorvastatin, cerivastatin, rosuvastatin, and fluvastatin was used. All searches were limited to studies of humans published in English. A manual search of abstracts presented between 2002 and July 2005 at the annual meetings of the American Heart Association, the American College of Cardiology, the American Society of Clinical Oncologists, and the American Society of Hematology was conducted. In addition, a manual review of references from primary or review articles was performed to identify any additional relevant studies. All potentially relevant articles were reviewed independently by 3 investigators (K.M.D., C.I.C., and C.M.W.). To be included in this meta-analysis, studies had to be (1) randomized trials of statins, (2) placebo- or routine treatment–controlled, (3) have a mean (or median) duration of patient follow-up of at least 1 year, (4) enroll a minimum of 100 patients, and (5) report data on the incidence of either cancer diagnosis or cancer death.

The following methodological features most relevant to the control of bias were assessed: randomization, random allocation concealment, masking of treatment allocation, blinding, and withdrawals. All studies were evaluated by 3 independent reviewers (K.M.D., C.I.C., and C.M.W.), with disagreement resolved by consensus.

All data were independently abstracted by 3 investigators (K.M.D., C.I.C., and C.M.W.) through use of a standardized data abstraction tool. Disagreements were resolved by consensus among the 3 reviewers. The following information was sought from each article: author identification, year of publication, geographic location of the study, study funding source, type of study design (prospective or retrospective, randomized or observational, presence and type of control, blinded or open-label), study population, sample size, duration of patient follow-up, statin used (specific agent, agent hydrophilicity or lipophilicity, and derivation of agent [natural or synthetic]), type of cancer diagnoses included (breast, prostate, colon, respiratory, gastrointestinal, or melanoma), definitions of cancer diagnosis and cancer death (when reported), and method of data collection within trials for cancer end points. In cases in which there was more than 1 published report on the same population or group of patients, the most recent article was selected for analysis, although previous articles could be reviewed to supplement missing data where applicable.

Incidence of cancer and cancer death were treated as dichotomous variables. Weighted averages were reported as odds ratios (ORs) with 95% confidence intervals (CIs) calculated using StatsDirect statistical software, version 2.4.5 (StatsDirect Ltd, Cheshire, England) and a DerSimonian and Laird random-effects model. Risk difference was also calculated for both cancer incidence and cancer death. Statistical heterogeneity was measured using the Q statistic (P<.10 was considered representative of significant statistical heterogeneity). Heterogeneity was also assessed through visual examination of L’Abbe plots.

To establish the effect of clinical heterogeneity between studies on meta-analysis' conclusions, subgroup analysis was conducted. Since the effect of statins on cancer may vary from one subtype of cancer to another, the effect of these agents on the incidence of different types of cancer diagnosis (breast, prostate, colon, respiratory, gastrointestinal, or melanoma) was evaluated. In addition, because numerous pharmacologic differences exist between available statins, comparisons were conducted of each individual statin studied (atorvastatin, pravastatin, simvastatin, lovastatin, and fluvastatin), as well as of hydrophilic (atorvastatin, pravastatin, and fluvastatin) or lipophilic (simvastatin, cerivastatin, and lovastatin) agents and naturally (pravastatin, simvastatin, and lovastatin) or synthetically (atorvastatin, cerivastatin, and fluvastatin) derived agents. Studies of poorer methodological quality, such as unblinded or open-label trials, may exhibit exaggerated treatment effects. Excluding them may result in increased internal validity but could reduce external validity of the analysis. In addition, the selection of a random- vs fixed-effects model in meta-analyses is controversial. The use of a random-effects model in the calculation of CIs results in wider intervals and, thus, a more conservative estimate of treatment effect compared with a fixed-effects model. To reconcile these issues, sensitivity analysis was conducted whereby the meta-analysis was reanalyzed excluding unblinded or open-label studies and using a Mantel-Haenszel fixed-effects model.

Several methods were used to assess the potential for publication bias. Visual inspection of funnel plots for both the cancer diagnosis and cancer death end points was conducted. The Begg rank correlation method and the Egger weighted regression method were also used to statistically assess publication bias for cancer incidence and cancer death (P<.05 was considered representative of statistically significant publication bias).

Our initial search yielded 8943 potential literature citations (Figure 1). Of these, 7097 were excluded by electronically limiting the search to humans, English language, and clinical trials and through review of citations. Abstracts from 1846 articles were reviewed and an additional 1669 trials were excluded (528 were not clinical trials, 20 were not conducted in humans, 767 had a study duration of <1 year, 254 enrolled <100 patients, 90 did not report cancer incidence or death, and 10 evaluated statins for the treatment of cancer), leaving 177 studies for full publication review. Of these, 150 were excluded (31 were not clinical trials, 19 had a study duration of <1 year, 51 enrolled <100 patients, and 50 did not report cancer incidence or death); thus, 26 studies (n = 86 936 participants) were found to conform to our inclusion criteria.^12- 38 (Table 1). A total of 22 studies^{12- 13,15- 16,18- 25,27- 28,30- 37,39} reported data on cancer death, and 20 studies^{14- 17,19,21- 26,29- 35,37- 39} reported usable data on incidence of cancer diagnosis. Three of the studies used an open-label design.^{20,24- 25,39} Two additional studies included a follow-up period of 2 to 5 years during which patients were treated with standard care after the initial randomized, double-blind, placebo-controlled trial.^15,17 Patient enrollment ranged between 151 and 20 536 patients.^12- 39 The mean patient age ranged between 50 and 76 years and participants were mostly male (73%).^11- 36 The included studies evaluated atorvastatin (n = 3),^12- 13,18 cerivastatin (n = 1),¹⁴ fluvastatin (n = 3),^16,22,29 lovastatin (n = 3),^23,37- 39 pravastatin (n = 11),^{17,20- 21,24- 25,27,30- 35,39} or simvastatin (n = 3)^{15,19,26,28,36} vs placebo or standard care. Fluvastatin, cerivastatin, and atorvastatin were the only synthetically derived statins we identified with published data on cancer end points.^{12- 14,16,18,22,29} Hydrophilic agents were evaluated in 18 studies,^{12- 13,17- 18,20- 21,23- 25,27,30- 35,37- 39} while lipophilic agents were evaluated in 9 studies.^{14- 16,19,22,26,28- 29,36} Duration of patient follow-up for cancer ranged from 1.9 years²⁴ to 10.4 years¹⁵ in the included studies.

The 95% CIs cross 1.00 for overall incidence of cancer diagnosis and cancer death analyses for patients receiving statin therapy vs controls (OR, 1.02; 95% CI, 0.97-1.07) and (OR, 1.01; 95% CI, 0.93-1.09), respectively (Figure 2 and Figure 3). This correlates to a risk difference of 0.0002 (95% CI, −0.003 to 0.0034) for cancer diagnosis and −0.0005 (95% CI, −0.0022 to 0.0012) for cancer death. Calculations of ORs were not possible for 3 cancer death studies due to the absence of any cancer deaths in their control groups and do not appear in Figure 3. In a reanalysis adding a nominal value (0.5 death) in all 2 × 2 cells to enable calculation of ORs, no statistically or clinically meaningful differences occurred vs the initial analysis. No statistically significant heterogeneity was observed between trials for either analysis with the Q statistic (P>.37 for both). In addition, L’Abbe plots did not show evidence of heterogeneity (Figure 4). Review of funnel plots could not rule out the potential for publication bias for either analysis. Publication bias was not evident when the Begg rank correlation method (P = .68 for cancer incidence and P = .35 for cancer death) and the Egger weighted regression method (P = .63 for cancer incidence and P = .23 for cancer death) were used (Figure 5).

No statistically significant differences were observed between patients receiving statin vs control for any of the prespecified cancer subtypes (Table 2). No statistical heterogeneity was observed between trials in any of the above-mentioned analyses (P>.14 for all), with the exception of breast cancer incidence (P = .047).

When we limited the analysis to individual statins, statins with high or low lipophilicity, or natural or synthetic statins, no significant differences were observed vs control (Table 2). The 95% CIs included 1.00 for each of these subgroup analyses. Except for the effect of simvastatin alone on cancer incidence analysis (P = .09), no statistical heterogeneity was observed between trials in the above-mentioned analyses (P>.11 for all).

In sensitivity analysis, the meta-analysis' conclusions remained robust to methodological changes. The results of the incidence of cancer diagnosis or cancer death analyses were not altered when a fixed-effects model was used (cancer diagnosis OR, 1.02; 95% CI, 0.97-1.07 and cancer death OR, 1.00; 95% CI, 0.93-1.09). When only randomized, double-blind, placebo-controlled statin trials were evaluated, there was no significant difference in cancer incidence (OR, 1.05; 95% CI, 0.99-1.12). This was not significantly altered when the first 5.4 years of the Scandinavian Simvastatin Survival Study (4S) and the first 6 years of the Long-term Intervention With Pravastatin in Ischaemic Disease (LIPID) study were added to this analysis (since this was the period for which they were randomized, double-blind, and placebo-controlled) (OR, 1.03; 95% CI, 0.98-1.09). Cancer death was not altered in either analysis (with or without 4S and LIPID).

Several previous retrospective case-control studies have suggested that statins have a marked effect on cancer incidence.^1- 7 Poynter and colleagues¹ performed a population-based case-control study of statin use (almost entirely pravastatin and simvastatin) and colorectal cancer in northern Israel. This retrospective trial of 3968 people found a 47% reduction in the risk of colon cancer with statin use. In another recent series of case-control studies, the incidences of lung, prostate, and breast cancer were all reduced by approximately 50%.^2- 4 A Dutch study reported a 20% risk reduction in cancer in patients receiving statins for 4 or more years in a population-based, nested case-control study.⁵ A similar nested case-control study conducted in Quebec noted that statin-treated patients were 28% less likely to be diagnosed as having cancer than those receiving bile acid–binding resins.⁶

Three previous meta-analyses have evaluated the impact of statins on cancer incidence. They included between 19 592 and 30 817 patients comprising 3 to 13 studies and did not find any differences in the incidence of cancer.^39- 40 Our meta-analysis was markedly larger than these previous analyses, we evaluated cancer death, and we evaluated individual types of cancer. In our current meta-analysis, statins did not reduce the incidence of cancer or cancer death. No reductions were noted for cancers of the breast, colon, gastrointestinal tract, prostate, respiratory tract, or skin (melanoma) when statins were used. Like in the case-control study by Poynter and colleagues,¹ the patients in our meta-analysis were primarily treated with simvastatin and pravastatin. As such, we evaluated pravastatin alone and simvastatin alone on cancer incidence and death and found no impact. Our results are in agreement with 3 previous case-control studies that found that statins did not reduce the incidence of cancer.^41- 43

We thought that hydrophilic statins, with their impaired ability to penetrate biological membranes, might provide different effects than lipophilic statins, which readily enter cells, but this was not evident in our study.⁴⁴ Similarly, naturally derived statins have a markedly different structure than synthetic statins but neither type affected the results.⁴⁵

Two large active-controlled clinical trials of atorvastatin have been conducted, but only the Treatment to New Targets (TNT) trial reported cancer results.⁴⁶ The TNT trial compared low- vs high-dose atorvastatin therapy (10 mg/d vs 80 mg/d), and the incidence of cancer was 1.5% vs 1.7%, respectively (P = .42).⁴⁶ This suggests that the dose of statin and intensity of low-density lipoprotein cholesterol reduction do not substantially alter the risk of cancer, which is heartening given the new impetus for low-density lipoprotein cholesterol levels well below 100 mg/dL. This is consistent with the results of 2 additional smaller studies in which higher-intensity dosing did not affect cancer incidence vs lower-intensity therapy.^47- 48

There are some limitations to this meta-analysis. First, differences in cancer surveillance and reporting may have contributed to the differences in cancer rates among studies. However, the definitions used and surveillance intensity was consistent within each study for the active and control groups so the relative impact should still be accurate. To standardize our results, we evaluated only studies reporting number of patients diagnosed as having cancer and, therefore, excluded the Antihypertensive and Lipid-Lowering Treatment to Prevents Heart Attack Trial (ALLHAT-LLT) from this analysis because it only reported the cumulative number of cancer diagnoses made (cancer death data from ALLHAT-LLT was used).¹³ Our use of cancer death is also important since the end point of death would be consistently applied between studies. Second, many of the studies included patients with a history of cancer, but the percentage of patients with preexisting cancer was not uniformly given, and whether the diagnoses of cancer were new cancers or recurrences was not elucidated. Third, information on other confounding variables, such as smoking, was not available for analysis, but, given the large populations involved, we would have anticipated that any differences in background factors would have been evenly distributed via randomization. Finally, as with any meta-analysis, the potential for publication bias is a concern. Publication bias results from ease of finding studies with significant or positive results, potentially leading to overrepresentation of a drug's benefit in systematic reviews. Our meta-analysis' funnel plots appear slightly asymmetrical. Therefore, in evaluating our funnel plots, we can report only that there may be publication bias, which is difficult to quantify. However, publication bias was not evident when the Begg rank correlation method and the Egger weighted regression method were applied to the data.

Statins have a neutral effect on cancer and cancer death risks in randomized controlled trials. We could find no type of cancer that statins benefited or subtype of statin that reduced the risk of cancer.