If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
The association between statins and incident diabetes mellitus (DM) in observational studies is much larger than that reported from randomized controlled trials. We sought to assess this association using a novel design controlling for selection bias.
Using data from MarketScan, we identified a cohort of non-diabetic patients who initiated a statin and matched them to patients not taking statins. From the statin-user cohort, we identified two subgroups: patients who received statin refills for >6 months (continuers) and patients who received statin refills <6 months (discontinuers). Patients were followed for a minimum of two years to determine incident DM.
We included 442,526 patients, divided equally between statin users and non-users. Statin use was associated with increased DM (9.9% vs. 4.4%, HR 2.2, p < 0.001). Among the 221,263 statin users, there were 194,357 continuers and 26,906 discontinuers. There was no significant difference in the incidence rate of DM between both groups (10.0% vs. 9.3%, HR 1.03, p = 0.22).
Statin use was strongly associated with incident diabetes when users were compared to non-users but not when continuers were compared to discontinuers. Selection bias confounds the association between statin use and incident diabetes in observational studies.
“Do statins cause diabetes?” “If so, to what extent?”
These questions are the focus of intense debate and the answers have important implications for public health. Statins are one of the most prescribed medications in the world and high level evidence supports their effectiveness for reducing the risk of atherosclerotic cardiovascular disease events in both primary and secondary prevention.
Cholesterol Treatment Trialists' (CTT) Collaborators The effects of lowering LDL cholesterol with statin therapy in people at low risk of vascular disease: meta-analysis of individual data from 27 randomised trials.
Numerous randomized controlled trials (RCTs) demonstrate that the benefit of statin therapy is related to baseline LDL values, the amount of LDL lowering achieved, and the difference in LDL levels achieved between groups.
The level of evidence related to statin-induced side effects and adverse events is less certain because adverse events tend not to be systematically ascertained and/or adjudicated in RCTs whose focus is primarily on demonstrating efficacy.
On the matter of statin-induced diabetes, data from RCTs and observational studies diverge. RCTs find a statistically significant but small absolute risk (Odds Ratio [OR] 1.09) for statin-induced diabetes.
Using a robust and novel observational study design, one specifically intended to control for selection bias, we set out to assess the association between statin exposure and incident diabetes.
Data are from the Truven Health Analytics MarketScan® Database, which captures clinical utilization and prescription drug claims data from more than 250 large employers, health plans, and government and public organizations. The database includes over 50 million individuals covered annually under commercial (private) insurance plans and captures administrative claims with data from inpatient visits, outpatient visits, and pharmacy claims de-identified at the patient level. This study was approved by the Penn State College of Medicine Institutional Review Board. Since the study used deidentified data, a consent form was not required.
During the first year of the study period (January 1st, 2008 to December 31st, 2008), referred to as the observation period, a cohort was selected by using International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM]. All variables were defined as having two or more diagnosis codes. Patients who had a prescription claim for a statin drug or a diagnosis of diabetes mellitus were excluded along with individuals having conditions making them more susceptible to develop diabetes such as cancers, COPD/asthma, end stage kidney disease, organ transplantations, and liver disease. A list of ICD-9 codes are provided in supplementary table 1.
Starting January 1st, 2009, the cohort was followed for the statin exposure which was defined as the first statin prescription claim. The statin exposure day was defined as the index date and this group is referred to as statin users. Patients were excluded from this group if they did not have at least 2 clinic visits following the index date. A group of statin non-users was created by matching patients in the statin user group in a 1:1 manner based on age and sex. Non-users were followed from the index date of their matched, statin-user pair and were only included if they had at least 2 clinic visits following the index date. Data from the baseline period, which started 364 days before the index date and ended on the index date were used to obtain baseline characteristics and co-morbid conditions of both groups.
We then identified two subgroups among the statin user group. The first group was patients who had a prescription claim for a statin drug more than 6 months following the index date. This group was labeled as “statin continuers.”. The second group was patients who did not have a prescription claim for a statin more than 6 months following the index date. This group was labeled as “statin discontinuers”.
Follow up and outcomes
All patients including those in the statin user and non-user cohorts as well as those in the statin continuer and discontinuer cohorts were followed for a minimum of two years. The main outcome variable was incidence of new diabetes defined as two or more ICD-9 codes for diabetes mellitus type II (ICD-9 250.xx, and 249.xx).
Pearson chi-square test was used for unadjusted comparison for categorical variables and student's t test was used for continuous variables. Cox proportional hazard models were used to examine the relationship between different groups and DM adjusting for hypertension, coronary artery disease, peripheral vascular disease, cerebrovascular accident, congestive heart failure, and hyperlipidemia.
A two-tailed p-value of <0.05 was considered to be statistically significant. All statistical analyses were performed with SAS software version 9.4 (SAS Institute Inc., Cary, NC, USA).
A total of 442,526 patients were included in the final analysis, divided equally between statin users and non-users. The mean age was 51.3 ± 7.5 years, and 124,826 (56.4%) were women. Patients were followed for 1140 ± 212.8 days. As shown in Table 1, statin users were significantly more likely to have cardiovascular comorbidities including hypertension (22.5% vs. 4.6%), coronary artery disease (3.1% vs. 1.0%), history of stroke (1.5% vs. < 1%) and hyperlipidemia (33.8% vs. 13.0%).
Table 1. Baseline characteristics of statin-exposed and statin-unexposed individuals.
A total of 31,591 (7.1%) patients developed a new diagnosis of diabetes during the follow-up period. Statin use was associated with an increased incidence of DM (9.9% vs. 4.4%, Hazards ratio “HR” 2.2, 95% Confidence Interval “CI” [2.2-2.3], P < 0.001) (Fig. 1a, Table 2). This risk was found to be statistically significant for all types of statins [HR of 3.2 for fluvastatin, 2.6 for atorvastatin, lovastatin, and rosuvastatin, 2.4 for pravastatin, and 2.3 for simvastatin and pitavastatin] and across all subgroups (Table 2). Statin users of younger age were at higher risk of developing diabetes, and the likelihood ratio of developing diabetes among statin and non-statin users decreased with age. However, statin users remained at higher risk of developing diabetes at any age.
Table 2Incidence of diabetes mellitus in statin-exposed and statin-unexposed individuals.
Among the 221,263 statin users, there were 194,357 continuers and 26,906 discontinuers. The average age for statin continuers was 51.5 years compared to 49.0 years for discontinuers. Statin continuers were more likely to have hypertension, peripheral vascular disease, stroke, hyperlipidemia, and coronary artery disease. (Table 3) The average follow up time was 1145.2 ± 212.4 days for statin continuers and 1101.9 ± 210.9 for discontinuers.
Table 3Baseline characteristics of statin continuers and discontinuers.
The incidence rate of new DM was similar between statin continuers and statin discontinuers (10.0% vs. 9.3%, HR 1.03, 95% CI [0.98-1.10], p = 0.22) (Fig. 1b, Table 4). The incidence of new DM was higher in the statin continuers versus discontinuers in individuals aged 20-29 years (8.9% vs. 6.7%, HR 1.63, 95% CI [1.1-2.4], p = 0.02), and 30-39 years (9.2% vs. 7.7%, HR 1.29, 95% CI [1.1-1.5], p < 0.001) (Table 4). In these subgroups the hazard ratios were significantly attenuated compared to the statin user versus nonuser subgroup comparisons (Table 2). For all other subgroups and female sex, the incident rate of new DM was similar between statin continuers and discontinuers.
Table 4Incidence of diabetes mellitus in statin continuers and discontinuers.
Our main finding was a strong relationship between statin use and new diabetes when statin users were compared to nonusers (9.9% vs 4.4%; HR 2.2; 95% CI 2.2-2.3) but this association was significantly attenuated when we ran the analysis between statin continuers versus discontinuers (10.0% vs. 9.3%; HR 1.03; 95% CI 0.98-1.10). We believe these findings suggest that the relationship between statin use and new diabetes in observational studies is significantly confounded by selection bias and that the application of statistical techniques like multivariable logistic regression and propensity-matching, which are intended to handle imbalances between patient groups, are not sufficient to overcome this limitation.
Observational studies have reported significantly higher odds of developing DM in patients taking statins than would be suggested by RCTs. Culver et al performed a secondary analysis of the Women's Health Initiative data and found significant risk of developing DM in post-menopausal women taking statins (HR 1.71; 95% CI 1.6-1.8).
Culver and Mansi both applied propensity matching to their analyses; however, propensity matching can only account for measured variables and even sophisticated datasets may not contain important factors that influence the prescribers’ choices to recommend a specific therapy or intervention.
Our method, on the other hand, of comparing statin continuers versus discontinuers accounts for all measured and unmeasured variables that may have influenced patient selection for statin therapy. For this reason, we believe these results are more reflective of the true association between statin use and new diabetes and, in this case, are consistent with results from RCTs, which show a statistically significant association with a small absolute effect size. While the association in this subgroup analysis did not quite reach statistical significance the overall HR of 1.03 is close to the OR of 1.09 reported in the meta-analysis by Sattar N et al.
The interpretation of our results, particularly in our sub-analysis of statin continuers versus discontinuers, is limited by several factors. First, we interpret the loss of association between statin continuers versus discontinues and diabetes to mean that the significant association between statin users versus nonusers and diabetes is confounded by selection bias. Though unlikely, we cannot completely rule out that any statin exposure, even if for a limited time, leads to an irreversible increase in new diabetes risk. Our study is also limited by lack of accounting for individual adherence to statin therapy, statin dose and exact time to discontinuation of statin use within the 6-month timeframe. These factors could influence the association between statin use and new diabetes. Finally, we cannot account for why statins were discontinued and these reasons might influence new diabetes risk.
In conclusion, we performed a novel two-step analysis using observational data to assess the association between statin use and diabetes. In the first analysis, statin users compared to nonusers had a significant increase in new diabetes even after accounting for important variables known to increase diabetes risk. In the second analysis, which we consider a novel addition to the literature on this topic, we compared statin continuers to discontinuers from the first analysis and found no statistically significant association between statin continuation and new diabetes. We feel that selection bias in the first analysis, which was corrected for in the second analysis, significantly contributes to the association between statin use and new diabetes and this far outweighs the causal contribution of statin therapy.
These findings suggest the statistically significant but small absolute increase in new diabetes, reported from statin RCTs is likely a true representation of the causal effect of statin therapy to new diabetes; not the much larger association reported in observational studies.
M.R, T.A and A.F. Study design and manuscript writing. G.L: Data analysis