Statin use and the risk of developing diabetes: a network meta-analysis

Purpose: Randomized controlled trials have shown mixed findings regarding the association of statins and diabetes. This systematic literature review and network meta-analysis (NMA) was performed to update evidence on this association to possibly assist clinicians in making more informed treatment choices.

Methods: We identified studies relevant to our NMA by performing study searches in databases like Embase, Cochrane, and PubMed, published between August 2010 and June 2014. Pre-2010 studies were identified from bibliography of previously published meta-analyses. Unpublished study data were found from clinicaltrial.gov. Data synthesis was performed by pairwise meta-analysis and NMA within a Frequentist framework.

Results: Twenty nine trials in which 1 63 039 participants had been randomized were included in this review; among these 1 41 863 were non-diabetic patients. The direct meta-analysis showed that statins, as a class, significantly increased the likelihood of developing diabetes by 12% (pooled OR 1.12; 95%CI 1.05-1.21; I² 36%; p = 0.002; 18 RCTs). In the NMA, atorvastatin 80 mg was associated with a highest risk of diabetes, with OR of 1.34 (95%CI 1.14-1.57) followed by rosuvastatin (OR: 1.17; 95%CI: 1.02-1.35). The ORs (95%CIs) for simvastatin 80 mg, simvastatin, atorvastatin, pravastatin, lovastatin and pitavastatin were 1.21 (0.99-1.49), 1.13 (0.99-1.29), 1.13 (0.94-1.34), 1.04 (0.93-1.16), 0.98 (0.69-1.38) and 0.74 (0.31-1.77), respectively. High-dose atorvastatin increased the odds of developing diabetes even when compared with pravastatin, simvastatin and low-dose atorvastatin in the NMA.

Conclusions: Based on the results, statins, as a class, increased the risk of diabetes significantly in the pairwise meta-analysis. Overall, there appears to be a small increased risk of incident diabetes, particularly with more intensive statin therapy, although more data would be valuable to increase the robustness of this interpretation, given that the lower confidence intervals of our study analyses are close to, or just crossing one. Copyright © 2016 John Wiley & Sons, Ltd.