Recent reports indicate that statins are associated with an increased risk for new-onset diabetes mellitus (DM) compared with placebo and that this relation is dose dependent. The aim of this study was to perform a comprehensive network meta-analysis of randomized controlled trials (RCTs) investigating the impact of different types and doses of statins on new-onset DM. RCTs comparing different types and doses of statins with placebo were searched for using the MEDLINE, Embase, and Cochrane databases. A search of RCTs pertinent to this meta-analysis covering the period from November 1994 to October 2012 was conducted by 2 independent investigators using the MEDLINE, Cochrane, Google Scholar, and Embase databases as well as abstracts and presentations from major cardiovascular meetings. Seventeen RCTs reporting the incidence of new-onset DM during statin treatment and including a total of 113,394 patients were identified. The RCTs compared either a statin versus placebo or high-dose versus moderate-dose statin therapy. Among different statins, pravastatin 40 mg/day was associated with the lowest risk for new-onset DM compared with placebo (odds ratio 1.07, 95% credible interval 0.86 to 1.30). Conversely, rosuvastatin 20 mg/day was numerically associated with 25% increased risk for DM compared with placebo (odds ratio 1.25, 95% credible interval 0.82 to 1.90). The impact on DM appeared to be intermediate with atorvastatin 80 mg/day compared with placebo (odds ratio 1.15, 95% credible interval 0.90 to 1.50). These findings were replicated at moderate doses. In conclusion, different types and doses of statins show different potential to increase the incidence of DM.
Although generally well tolerated, statins, compared with placebo, have been associated with a higher incidence of new-onset diabetes mellitus (DM) in several experimental studies and a meta-analysis of 13 randomized controlled trials (RCTs). Another meta-analysis of 5 studies showed a dose-dependent effect of statins on the incidence of DM. On the basis of these findings, the US Food and Drug Administration has recently added information to statin labels regarding the impact of these agents on DM. The constellation of statins is wide, with differences concerning active compounds, associated effects, and therapeutic doses. The recent concerns about the safety of statins pose therapeutic dilemmas regarding which type and dose of statin may minimize the risk for developing DM. To date, appropriately powered head-to-head comparisons among statins with regard to the DM end point are lacking. Given the perceived need to understand the specific risk for developing DM associated with one statin compared with another and to relate that to the administered dose of the drug, a network meta-analysis is timely and warranted. Accordingly, we performed a comprehensive network meta-analysis of RCTs investigating the impact of different types and doses of statins on new-onset DM.
Methods
Established methods were used in compliance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement for reporting systematic reviews and meta-analyses in health care interventions. A search of pertinent RCTs conducted from November 1994 to October 2012 was performed by 2 independent investigators covering the MEDLINE, Cochrane, Google Scholar, and Embase databases as well as abstracts and presentations from major cardiovascular meetings, using the search string “statins AND/OR diabetes.” The internal validity of the RCTs was assessed by 2 independent reviewers.
Citations were screened at the title and abstract level and retrieved as full reports. Inclusion criteria were (1) studies in humans, (2) RCTs, and (3) studies comparing patients treated with high-dose statins versus placebo or with high- versus moderate-dose statins connected in a network with a third comparison (placebo or statin) and reporting the incidence rates of new-onset DM in both arms. To be consistent with other large meta-analyses and to provide robust estimates, we excluded trials with follow-up ≤1 year and including <1,000 patients.
A network meta-analysis was planned with respect to new-onset DM as an end point to compare (1) high-dose statins versus placebo and different high doses of statins, 2) moderate-dose statins versus placebo and different moderate doses of statins, and (3) high-dose versus moderate-dose statins. This research tool remains a well-established method capable of comparing different treatments using a common reference treatment while maintaining the randomization design and integrating data from direct and indirect comparisons.
New-onset DM was defined as any adverse event report of DM, or starting glucose-lowering medication, or a fasting plasma glucose level ≥7 mmol/L (either 1 or 2 values, depending on the frequency of measurement in the trial). Dichotomous outcome variables were compared using odds ratios (ORs) and 95% credible intervals (CIs) by means of network meta-analysis using a Bayesian hierarchical random-effects model. Analysis was based on noninformative prior findings for effect sizes and precision. Convergence and lack of autocorrelation were checked and confirmed after 10,000 iterations. The final summary statistics were based on a further 100,000 iterations, after discarding the initial 10,000-iteration burn-in. Sensitivity analysis was conducted by repeating the main computations using a fixed-effect method. Pairwise contrasts were examined, and heterogeneity was assessed using the I 2 statistic, with values <25%, ≥25% and ≤50%, and >50%, respectively, representing mild, moderate, and severe heterogeneity. Inconsistency between direct and indirect evidence sources was assessed by inspection of the model fit and by comparing the results of the pairwise meta-analyses with the estimates from the network meta-analysis. We computed the probability that each statin agent was the best treatment in terms of inducing less DM. The ranking of the competing drugs was assessed with the median of the posterior distribution for the rank of each drug; cumulative ranking probabilities curves for competing statin treatments were built. The surface under the cumulative ranking curve (SUCRA) was derived by using the posterior probabilities for each treatment to be among the n best options; SUCRA would take a value of 1 when the treatment is certain to be the best and of 0 when the treatment is certain to be the worst. A Bayesian random-effects meta-regression was performed to formally explore whether the effect on DM is related to the power of statins to reduce cholesterol or rather to a molecule-dependent mechanism. Additionally, the potential effect on the results of different body mass indexes (BMIs) across the studies as a proxy for different cardiovascular risk profiles was investigated; specifically, in the meta-regression analysis, the Bayesian OR for DM of each statin versus placebo was regressed against the percentage of low-density lipoprotein (LDL) cholesterol reduction using BMI as a covariate. The goodness of fit of the model to the data was assessed using the residual deviance.
Results
The flow diagram of the study is shown in Figure 1 . Seventeen RCTs fulfilling the eligibility criteria and comprising a total of 113,394 patients were eventually included for data abstraction. Table 1 lists the main characteristics of the included studies. Fourteen RCTs compared a statin with placebo, and 3 studies compared high- with moderate-dose statin therapy. The high daily doses of statins used in the RCTs were atorvastatin 80 mg, lovastatin 20 to 40 mg, pravastatin 40 mg, rosuvastatin 20 mg, and simvastatin 40 mg. The moderate doses were atorvastatin 10 mg, pravastatin 10 to 20 mg, and rosuvastatin 10 mg, all administered once daily. The included statins were connected in a network, as shown in Figure 2 . The final model reached a posterior mean residual deviance of 34.46 (the number of data points was 34), indicating a very good fit of the model. The estimated effects from the pairwise meta-analyses were comparable with the results of the network meta-analysis, suggesting no evidence of inconsistency.
| Study | Year | Trial Population | Trial Design | Compared Regimens | DM at Baseline | Mean Duration of Follow-Up (yrs) | New DM Cases in Compared Regimens | Mean BMI (kg/m 2 ) | Mean Age (yrs) | Relative LDL Reduction | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Yes | No | ||||||||||
| ASCOT-LLA | 2003 | Hypertension, CV risk factors, no history of CAD | Double-blind | Atorvastatin 10 mg vs placebo | 2,532 (24.6%) | 7,773 (75.4%) | 3.3 ∗ † | 154 (3.9%) vs 134 (3.5%) | 28.6 † | 63.0 † | 34.8% at 12 mos ‡ |
| HPS | 2003 | History of CVD | Double-blind | Simvastatin 40 mg vs placebo | 5,963 (29.0%) | 14,573 (71.0%) | 5.0 | 335 (4.6%) vs 293 (4.0%) | 27.2 | 65.0 | 29.4% average in trial |
| JUPITER | 2008 | No CVD | Double-blind | Rosuvastatin 20 mg vs placebo | 0 (0%) | 17,802 (100.0%) | 1.9 ∗ | 270 (3.0%) vs 216 (2.4%) | 28.4 ∗ | 66.0 ∗ | 50.0% at 12 mos |
| WOSCOPS | 2001 | No previous MI, elevated cholesterol | Double-blind | Pravastatin 40 mg vs placebo | 621 (9.4%) | 5,974 (90.6%) | 4.8 | 75 (2.5%) vs 93 (3.1%) | 25.9 | 55.0 | 23.7% at 12 mos |
| LIPID § | 2003 | MI or UA in previous 3 yrs | Double-blind | Pravastatin 40 mg vs placebo | 2,017 (22.4%) | 6,997 (77.6%) | 6.0 | 126 (3.6%) vs 138 (3.9%) | Not reported | 62.0 ∗ | 25% (during 5 yrs) |
| CORONA | 2007 | Systolic CHF | Double-blind | Rosuvastatin 20 mg vs placebo | 1,477 (29.5%) | 3,534 (70.5%) | 2.7 ∗ † | 100 (5.6%) vs 88 (5.0%) | 27.0 ‡ | 73.0 † | 45.1% at 3 mos) ‡ |
| PROSPER | 2002 | Elderly patients with CVD or at high risk | Double-blind | Pravastatin 40 mg vs placebo | 781 (13.5%) | 5,023 (86.5%) | 3.2 | 165 (6.6%) vs 127 (5.1%) | 26.5 | 76.0 | 30.7% at 12 mos |
| MEGA | 2006 | No CVD, elevated cholesterol | Open trial | Pravastatin 10–20 mg vs no treatment | 1,746 (22.3%) | 6,086 (77.7%) | 5.3 | 172 (5.7%) vs 164 (5.3%) | 23.8 | 58.3 | 17.1% at 12 mos |
| AFCAPS/TexCAPS | 1998 | No CVD | Double-blind | Lovastatin 20–40 mg vs placebo | 394 (6.0%) | 6,605 (94.0%) | 5.2 † | 72 (2.3%) vs 74 (2.4%) | 27.0 † | 58.0 † | 26.7% at 12 mos |
| 4S | 1994 | Previous MI or angina | Double-blind | Simvastatin 20–40 mg vs placebo | 202 (4.5%) | 4,242 (95.5%) | 5.4 ∗ | 198 (9.4%) vs 193 (9.1%) | 25.9 | 58.6 | 36.7% at 12 mos |
| ALLHAT-LLT | 2002 | CAD or CAD risk factors | Open trial | Pravastatin 40 mg vs no treatment | 4,268 (41.2%) | 6,087 (58.8%) | 4.8 † | 238 (7.9%) vs 212 (6.9%) | 29.0 | 66.4 | 18.1% at 24 mos |
| GISSI-HF | 2008 | CHF | Double-blind | Rosuvastatin 10 mg vs placebo | 1,196 (26.1%) | 3,378 (73.9%) | 3.9 ∗ | 225 (13.6%) vs 215 (12.5%) | 26.7 | 67.0 | 34.9% at 12 mos |
| GISSI Prevenzione | 2000 | MI within past 6 mos | Open trial | Pravastatin 20 mg vs no treatment | 811 (19.0%) | 3,460 (81.0%) | 2.0 ∗ | 96 (5.5%) vs 105 (6.1%) | 26.3 | 59.3 | 11.5% at 12 mos |
| PROVE-IT–TIMI 22 | 2004 | Recent ACS | Double-blind | Atorvastatin 80 mg vs pravastatin 40 mg | 767 (18.4%) | 3,395 (81.6%) | 2.0 | 101 (5.9%) vs 99 (5.9%) | 29 | 58 | 22% |
| TNT | 2005 | Stable CAD | Double-blind | Atorvastatin 80 mg vs atorvastatin 10 mg | 2,406 (24.1%) | 7,595 (75.9%) | 5.0 | 418 (11.0%) vs 358 (9.4%) | 28 | 61 | 22% |
| IDEAL | 2005 | Previous MI | Double-blind | Atorvastatin 80 mg vs simvastatin 20–40 mg ‡ | 1,427 (16.0%) | 7,461 (84.0%) | 4.8 ∗ | 240 (6.4%) vs 209 (5.6%) | 27 | 62 | 16% |
| SPARCL | 2006 | Previous stroke or TIA | Double-blind | Atorvastatin 80 mg vs placebo | 794 (16.8%) | 3,937 (83.2%) | 4.9 | 166 (8.7%) vs 115 (6.0%) | 27.15 | 62.5 | NA |
† Data from total cohort including patients with DM at baseline.
‡ If, at 24 weeks, plasma total cholesterol level was >190 mg/dl (5.0 mmol/L), the dose of simvastatin could be increased to 40 mg/day. The dose of atorvastatin could be decreased to 40 mg/day for adverse events.
§ Includes only patients with normal fasting glycemia at baseline.
A total of 9 studies including 64,137 patients contributed to the analysis of new-onset DM in patients treated with high-dose statins, compared with a placebo control group. In the overall cohort, there was a total of 4,610 cases of new-onset DM: 7.28% (2,335 of 32,070) in the high-dose statin group and 7.09% (2,275 of 32,067) in the control group. As shown in Figure 3 , treatment with rosuvastatin 20 mg/day, compared with placebo, was associated with a numeric 25% relative increase in the risk for developing new-onset DM. The impact on the risk for DM with atorvastatin 80 mg/day was less evident. Conversely, therapy with pravastatin 40 mg/day was associated with the lowest risk for DM, almost comparable with placebo treatment. The results for simvastatin 40 mg/day were substantially comparable with those for rosuvastatin 20 mg/day. Pravastatin 40 mg/day was associated with a consistent relative risk reduction of new-onset DM (16%) compared with rosuvastatin 20 mg/day, and atorvastatin 80 mg/day resulted in an approximately 8% relative risk reduction for new-onset DM compared with high-dose rosuvastatin.
Eleven studies including 63,558 patients reported rates of DM in patients treated with either moderate doses of statins or placebo. In patients treated with moderate-dose statins, there were 2,601 cases of DM among 31,764 patients (8.18%) compared with 2,527 cases among 31,794 patients (7.95%) in the control group. Figure 3 shows that even moderate-dose rosuvastatin therapy still created the highest risk for DM. Treatment with atorvastatin 10 mg/day was almost comparable with placebo, whereas pravastatin 10 mg/day was associated with a numerically lower risk for DM compared with placebo. The results also show a numerically lower risk associated with pravastatin treatment compared with rosuvastatin. At moderate doses, the risks for DM with atorvastatin and rosuvastatin were comparable.
Figure 3 presents the effect on new-onset DM of high versus moderate doses of statins. The risk for DM was generally increased with higher dose statin regimens. With rosuvastatin 20 mg/day, a 12% increase in the relative risk for DM was observed, compared with rosuvastatin 10 mg/day. High-dose pravastatin was associated with a slightly increased relative risk (7%) compared with low-dose pravastatin. The risk for DM did not appear to increase comparing lower with higher dose atorvastatin.
Figure 4 summarizes the likelihood of being the best high-dose statin treatment in terms of inducing less DM compared with placebo. Pravastatin at high doses was found to have the highest probability to be in the top ranks for the best treatment, presenting a relative SUCRA of 0.53. Atorvastatin at high doses ranked second, with a SUCRA of 0.37. The lowest 2 positions belonged to high-dose rosuvastatin and simvastatin, associated with SUCRAs of 0.30 and 0.29, respectively.
Figure 5 summarizes the probabilities of noninferiority to placebo for various threshold values, expressed as ORs, for statin therapy compared with placebo.
