In our crude model, we found that statin use at randomization was associated with a 17% lower rate of new lesions of keratinocyte carcinoma among high-risk veterans. Adjustment for individual material confounders by Cox regression did not alter this estimate. After approximately 1 year, the difference in risk remained approximately constant and was consistent for BCC and SCC separately. Results of our propensity score–matched analysis, however, suggests that statin use is not associated with the rate of keratinocyte carcinoma. Because of the width of the CIs, these data are inconclusive, but they do not refute the hypothesis that statins may prevent cancer.
Our Cox regression results are consistent with findings from a pooled analysis of clinical trials that found that use of fluvastatin was associated with a 24% reduction in the occurrence of nonmelanoma skin cancer (40
). To our knowledge, no other studies have specifically shown an effect of statins on risk for keratinocyte carcinoma. Most research evaluating the effect of statins on skin cancer has focused on melanoma. A recent systematic review of randomized clinical trials showed no statistically significant reduction of melanoma incidence among users of statins and fibrates (41
). Late-acting etiologic actions are more important for SCC than for BCC (30
), and we hypothesized that statin exposure in this study occurred late in the keratinocyte carcinoma etiologic process. The point estimate for SCC was more protective than for BCC (although not statistically separable), which is consistent with this previous knowledge.
Because our study was inconclusive, it does not refute the hypothesis that statin use is associated with a lower risk for some cancers. Indeed, recent data suggest that statins lower risk for prostate cancer (42
), especially with long-term use (43
) or in advanced disease (22
), but seem to have no benefit in colorectal cancer (46
). However, an epidemiology study by Farwell and colleagues (48
) found an approximate 25% reduction in total cancer among statin users in the Veterans Affairs New England Healthcare System data (48
). Others found associations between statins and lower risk for renal cell carcinoma (49
) but no association with lung and breast cancer (50
Data on cancer incidence from large randomized trials of statins have been similarly inconclusive. A recent meta- analysis of randomized trials showed no effect of statins on cancer incidence and death (51
). It is likely, however, that the studies in the meta-analysis were affected by exposure (52
) and outcome (53
) misclassification and insufficient follow-up time (54
). The relative lipophilicity of statins may influence their cancer prevention efficacy by affecting their target-cell penetration (52
). Hydrophilic agents, such as pravastatin, may not penetrate certain tumor cells, which might render them ineffective or even harmful (52
). The importance of statin lipophilicity may also vary by tumor site. This argues against the notion that statin chemoprevention is a class effect. Moreover, differences in tumor site and characteristics may affect a cancer’s sensitivity to statins (53
). Each of these issues may have biased the results of the meta-analysis toward the null and may similarly affect individual randomized trials and epidemiology studies.
Many previous studies did not have power to assess cancer outcomes, especially at specific sites; did not rigorously measure cancer outcomes; and were not conducted in participants at high risk for cancer (7
). The VATTC investigators ascertained cases of keratinocyte carcinoma consistently and validly, especially for BCC and the full spectrum of SCC (27
). This rigorous measurement has been unavailable in previous studies (40
). Our approach decreases the possibility of outcome misclassification. However, the reliability estimates from an internal VATTC validation study suggest that, despite these efforts, there is still some error in the outcome measurement, with κ
values for interrater agreement of SCC as low as 0.62 for our outcome of invasive SCC (27
). Any residual misclassification is unlikely to be related to receipt of a statin or error in other variables and would therefore attenuate our estimates. In addition, BCC, which was measured with less error, accounted for most outcomes.
Statin users may be more likely than nonusers to seek the services of health professionals. This could lead to increased probability of diagnosing keratinocyte carcinoma in this group that could bias the estimate of the statin–keratinocyte carcinoma relation toward no effect. However, the VATTC investigators examined participants every 6 months and ascertained other keratinocyte carcinoma events through usual care pathology reports. This method makes this bias unlikely in our study. Assuming a true protective effect of statins, long-term users of these drugs may have been less likely to enroll in the VATTC because 1 inclusion criterion was a history of 2 or more keratinocyte carcinomas. However, because these data suggested that statin use was of long duration in the VATTC participants, selection bias by this mechanism is unlikely.
Because we used claims data to ascertain exposure, we cannot be sure that participants who received statins actually took their medication. However, recent work has shown that VA pharmacy claims accurately assess medication use (4
). Our study is an improvement over previous work assessing the effect of statin use on various cancers, which may have been affected by exposure misclassification (24
). However, in our study, veterans may have filled prescriptions outside of the VA pharmacy system, which could have led to underascertainment of medication use. To address the possibility that veterans could have filled prescriptions elsewhere, we included only participants who used the VA pharmacy system. This approach ensured similar completeness of drug data among the statin users and non-users. Moreover, veterans have a financial incentive to fill their prescriptions within the VA system, and evidence suggests that VA pharmacy claims are valid and reliable for ascertaining chronic disease burden (55
). Any exposure misclassification by this mechanism would probably dilute our results.
Our drug use data started in September 1998. Because of this left-censoring, we could not ascertain the date of statin treatment initiation or estimate the actual duration of statin use. However, the available data suggested that statin therapy was of long duration. Results of the time-varying Cox regression model suggest that keratinocyte carcinoma is not immediately sensitive to discontinuation of a statin, which is consistent with the notion that keratinocyte carcinoma has at least a moderate-length empirical induction period and that short-term therapy may be insufficient. In the case of a true biological discontinuation effect or exposure misclassification, an attenuation of the primary estimate in the fixed-time model compared with the time-varying model would be expected. We found no evidence of this, but our study was underpowered to test the effect of statin discontinuation. However, the consistent small decrease in keratinocyte carcinoma risk associated with statin use in these models may reflect residual confounding, which is better controlled in the propensity score–matched analysis.
The VATTC study participants were at high risk for keratinocyte carcinoma. The incidence of keratinocyte carcinoma varies by sex and geographic region (56
). The rates among the unexposed participants in this study were 50 to 100 times greater than those observed in the general population (56
). Whether our findings are generalizable to populations at low or moderate risk is unknown. We found the expected association between well-known risk factors and the occurrence of keratinocyte carcinoma, including male sex, older age, and sun sensitivity (data not shown) (30
Confounding by indication is an unlikely noncausal explanation of our estimates because risk factors for kera-tinocyte carcinoma are unlikely to influence the decision to prescribe statins, but we cannot rule out the possibility of unmeasured confounding. However, the VATTC investigators collected measurements on many potential con-founders. This ascertainment of potential confounding factors is a strength of the study. Conducting the study in this setting was preferable to using data from cardiovascular disease prevention studies. There was little material confounding in our study, including among variables that we considered plausibly associated with statin use (for example, health behaviors, such as sunscreen use). This provides further evidence that unmeasured confounding is not at play. The estimated association between the use of other cholesterol-lowering drugs and keratinocyte carcinoma was compatible with chance despite a point estimate that suggests a protective effect. The implication of this finding is unclear, but it does not refute the hypothesis that statins may prevent cancer independently of cholesterol synthesis (20
). Previous work has shown that statin users are generally healthier than similarly aged nonusers (57
). The rate ratios for other preventive therapies (for example, antihypertensives and other cholesterol-lowering agents) and keratinocyte carcinoma were close to 1.0, which suggests that the “healthy-user” effect does not explain our Cox regression findings.
Statins are prevalent, are well tolerated, and have been shown to improve outcomes for persons with cardiovascular disease. Preclinical research has elucidated potential mechanisms of a chemopreventive effect of statins, including beneficial effects on apoptosis and angiogenesis (20
). This study provides inconclusive evidence on the potential protective effect of statin use on the occurrence of cancer. Future studies must test the effect of individual statins on other types of cancer and assess whether this association varies by type of statin, dose, and duration of use.