The account from WOSCOP introduced the notion that extended statin use – 40 mg of pravastatin – may be associated with decreased incidence of diabetes [6
]. Of note, this was a post-hoc analysis of 5974 patients of whom 139 developed new-onset diabetes as defined by at least 2 mmol/l (36 mg/dl) rise in blood glucose above baseline values – a criterion for diabetes inconsistent with conventional clinical practice [6
]. Even so, the reported risk ratio of 0.7 for incident diabetes was only marginally statistically significant, as reflected by the upper limit of the confidence interval (95% CI 0.5–0.99), the width of which reflects the small number of cases of incident diabetes and limited precision in the risk estimate [6
]. The WOSCOP investigators acknowledged that their criterion for diabetes restricted the number of patients classified as newly diabetic. In the aftermath of the WOSCOP report, numerous placebo-controlled and standard care-controlled trials of statin therapy have provided conflicting reports regarding statin use and diabetes. Such conflicting observations could stem from the fact that some individual studies lacked the statistical power and event rate (of incident diabetes) required to demonstrate the association. Given such a potential scenario, a systematic review of multiple studies with quantitative synthesis of data – meta-analysis – can be particularly helpful for determining the statistical robustness of any association. In this context, a meta-analysis of 13 [6
] randomized placebo-controlled and standard care-controlled trials was recently conducted using published and unpublished data. Sattar et al.
] demonstrated a 9% increased risk of incident diabetes (odds ratio 1.09; 95% CI 1.02–1.17) in their analysis of 91 140 participants, among whom 4278 developed diabetes. Furthermore, metaregression analysis demonstrated that age was associated with the risk of incident diabetes, with stronger statin-attributable risk found in studies with elderly patients [19••
]. Neither change in low-density lipoprotein (LDL)-cholesterol during treatment nor baseline BMI was found associated with the risk of new-onset diabetes. There was little heterogeneity among the component studies of the meta-analysis.
In another study, Rajpathak et al.
] reported similar findings in their synthesis of five trials [8
] involving 51 619 participants, among whom 1943 developed diabetes. There was no heterogeneity among the component studies. The risk of incident diabetes was 13% (risk ratio 1.13; 95% CI 1.03–1.23); however, upon adding the WOSCOP study results, heterogeneity among the component studies became significant, and attenuation of the effect size resulted in loss of statistical significance [20
]. Notably, Sattar et al.
] included the WOSCOP study in their meta-analysis but eliminated the nonstandard criteria for diabetes employed by Freeman et al.
] in their original post-hoc analyses of WOSCOP. However, Sattar et al.
found that even the inclusion of WOSCOP data, as in the 2001 report, did not change the outcome of their meta-analysis, perhaps due to greater statistical power in their study [6
]. Collectively, the consistency in the findings from two independent meta-analyses suggests that statin use is indeed associated with increased risk of incident diabetes, although small and of no clear practical relevance. Furthermore, the fact that there was no significant heterogeneity between the studies synthesized suggests that the increased incidence of diabetes is secondary to a class effect as opposed to any apparent differences between drugs, such as their compartmentalization (lipophilic versus hydrophilic), metabolism [Cytochrome P450 (CYP)3A4, CYP2C9, UDP glucuronosyltransferase (UGT), etc.], potency, and half-life. Contrary to the findings of the above-discussed meta-analyses, another meta-analysis from Coleman et al.
] reported that statins as a class are not associated with increased risk (risk ratio 1.03; 95% CI 0.89–1.19) of new-onset diabetes. However, in their meta-analysis of five prospective randomized controlled trials involving 39 791 patients, among whom 1407 developed diabetes, they found significant statistical heterogeneity. Notably, as in the study by Rajpathak et al.
, upon removal of the WOSCOP study from their analysis there was little or no heterogeneity, and a significant increase in the relative risk of developing diabetes was noted (risk ratio 1.14; 95% CI 1.02–1.28) [21
]. It is also important to note that this meta-analysis was performed before the publication of JUPITER.
The above three meta-analyses, published sequentially, provide incremental advances in our understanding of the clinical trial evidence for the association between statins and incident diabetes. To put these papers in a chronologic perspective, the reports by Coleman et al.
] (pre-JUPITER) and that by Rajpathak et al.
] (post-JUPITER) both demonstrate increased risk of incident diabetes with statin use, an effect that in each report was attenuated upon inclusion of WOSCOP due to heterogeneity despite the increase in sample size. The most recent report by Sattar et al.
] appears to have conclusively addressed the controversy of statin use and risk of incident diabetes, as it represents the most comprehensive synthesis of data to date – devoid of heterogeneity – with adequate statistical power. Thus, we believe that extended statin use is associated with a modest increase in the risk of incident diabetes. Despite this conclusion, a cautionary note is important. Although meta-analyses are based on rigorous statistical methodology, they are not controlled experiments, but rather observational studies that rely on previously published data. Consequently, meta-analyses are, at best, hypothesis-generating studies. However, there are mechanistic and experimental studies that support the notion that extended statin use may alter glucose metabolism, thus lending a biologic platform to the conclusion that must be derived from the meta-analyses.