Background The 3-hydroxy-3-methylglutaryl-CoA reductase inhibitors (statins) reduce serum cholesterol level and cardiovascular morbidity and mortality. However, the effect of statins on glucose metabolism is unclear. Some studies have suggested that statins may cause hyperglycemia by increasing calcium concentration in the islet cells leading to decrease in insulin release or by decreasing GLUT 4-mediated peripheral glucose uptake.
Methods We analyzed the data in 345,417 patients (mean age 61 ± 15 years, 94% males, 6% diabetic, 20% statin users) from the Veterans Affairs VISN 16 database. We studied change in fasting plasma glucose (FPG) in this population over a mean time of 2 years between the first available measurement and the last measurement form the most recent recorded visit. Data were limited to patients who had 2 FPG measurements. Diagnosis of diabetes had to be present before the first FPG measurement.
Results Among patients without diabetes, FPG increased with statin use from 98 mg/dL to 105 mg/dL, and among nonstatin users, FPG increased from 97 mg/dL to 101 mg/dL (increase in FPG with statin use P < 0.0001). Among patients with diabetes, FPG increased with statin use from 102 mg/dL to 141 mg/dL, and among nonstatin users, FPG increased from 100 mg/dL to 129 mg/dL (increase in FPG with statin use; P < 0.0001). After adjustment for age and use of aspirin, β-blockers, and angiotensin-converting enzyme inhibitors, the change in FPG in nondiabetic statin users was 7 mg/dL (vs 5 mg/dL in nonstatin users, P < 0.0001) and for diabetic statin users it was 39 mg/dL (vs 32 in nonstatin users, P < 0.0001).
Conclusions Statin use is associated with a rise of FPG in patients with and without diabetes. This relationship between statin use and rise in FPG is independent of age and use of aspirin, β-blockers, and angiotensin-converting enzyme inhibitors.
The use of HMG-CoA reductase inhibitors (statins) has been proven beneficial in reducing mortality and morbidity in varied patient populations. A number of studies1-4have provided evidence that the administration of statins inhibits the development of coronary atherosclerosis and reduces mortality from heart disease as well as overall mortality in patients with type 2 diabetes mellitus. The Collaborative Atorvastatin Diabetes Study trial demonstrated that 10 mg of atorvastatin daily decreased the risk of first cardiovascular event, including stroke, in patients with type 2 diabetes mellitus.5
Although statins are very effective and safe cholesterol-lowering agents, serious adverse events like liver enzymes elevation and rhabdomyolysis are associated with their use. More recently, some statins have been linked to increased risk of developing diabetes mellitus or worsening of blood sugar control in diabetic patients.6This aspect of statin pharmacology has not been thoroughly emphasized because of lack of conformational data. The mechanism/s by which statins may affect glycemic control has been speculated upon,7,8but not clear.
The aim of the current study was to evaluate the effect of statin use on fasting plasma glucose (FPG) in a large patient population with and without diabetes. Toward this goal, we studied the effect of statins, as a group, on FPG levels in a cross-sectional data mined from the Department of Veterans Affairs VSIN 16 comprising approximately 15 million veterans followed in 10 hospitals in the southern United States.
The United States Veterans Affair Healthcare System (VISN16) provides care to over 400,000 veterans per year. These patients receive their medical care and drugs free of charge or at minimal cost from the Department of Veterans Affairs, assuring high degree of compliance with physician recommendations and follow-up visits. The Veterans Affairs Computerized Patient-care Review System provides detailed information on patient demographics, comorbid conditions, laboratory findings, and therapy.
VISN16 patient care database was examined with regard to the change in FPG among statin users versus nonstatin users. The data were obtained by mining the VISN16 Data Warehouse. The VISN16 Data Warehouse contains clinical and demographic information on all patients cared for at the 10 medical centers that comprise the Central Veterans Affairs Healthcare Network in the mid-south region of the United States. Access to the Data Warehouse is monitored and controlled by dedicated data administrators.
The present study was approved by the Institutional Review Boards of the University of Arkansas for Medical Sciences, Central Arkansas Veterans Healthcare System, the related Research and Development Committee, and by the VISN-wide committee that regulates access to the Data Warehouse.
The primary outcome of the study was change in the FPG between the first available measurement and the last measurement from the most recent recorded visit in statin versus nonstatin users. The patients were subgrouped as having diabetes or not having diabetes. Data were limited to patients who had at least 2 FPG measurements. In patients with diabetes, diagnosis of diabetes mellitus had to be present before the first measurement of FPG. Fasting plasma glucose was measured using Glucose oxidase assay in all the centers where the data were collected. Laboratory data entry was the basis to ascertain that the given value is from a fasting sample.
The terms used in this study are the following:
Statin users-patients in cohort with a prescription for a statin (simvastatin, atorvastatin, fluvastatin, lovastatin, pravastatin, cerivastatin, rosuvastatin);
Nonstatin users-patients in cohort with no prescription for a statin.
Other pertinent terms in this study are the following:
Hypertension-patient documented to have hypertension in the chart (generally systolic blood pressure reading ≥140 mm Hg or a diastolic blood pressure ≥90 mm Hg) or on antihypertensive regimen; diagnosis with a 401, 405, or 458 prefix;
Diabetes mellitus-diagnosis with an ICD9 250 prefix;
Dyslipidemia-diagnosis with a prefix of 272 including pure hypercholesterolemia, pure hypertriglyceridemia, mixed hyperlipidemia and hyperchylomicronemia.
All statistical analyses were performed using SASv9 software package (SAS Institute Inc., Cary, NC). Descriptive statistics included means and SDs for continuous measurements and proportions for categorical ones. We applied 2-sample t tests and χ2 tests, respectively, to test the equality of means/proportions between statin user and nonuser groups. Association between statin use and the last FPG as well as change in FPG was tested using simple regression models (presented as crude models), as well as multivariate regression models in order to control for baseline FPG differences and other covariates at each step (presented as adjusted models). Purposeful selection methods were used during that process to evaluate significance and confounding effects of the covariates. Covariates considered for the final model for the outcome of interest included age, sex, smoking, hypertension, total cholesterol, low-density lipoprotein (LDL) cholesterol and other medications (angiotensin converting enzyme [ACE] inhibitors, β-blockers, calcium blockers, and aspirin). Adjusted means for the last FPG and change in FPG are presented separately for subjects with and without diabetes. Similarly, changes in lipid profiles and final measurements of those by statin use were estimated using analysis of covariance with adjustments for baseline values. All associations below the α level of 0.05 were considered significant.
We identified 345,417 subjects (mean age 61 ± 15 years, 94% males) who met the criteria for inclusion in the study. Of these, 69,083 (20.0%) were statin users. Diabetes mellitus was prevalent in 6% of patients.
Table 1 shows the demographics and baseline characteristics of statin users and nonstatin users. Patients receiving statins were older and had a significantly higher prevalence of diabetes, hypertension, and smoking. Baseline serum total cholesterol and LDL-cholesterol levels were higher and high-density lipoprotein-cholesterol levels lower in statin users, as compared with nonstatin users (all P < 0.0001). The use of β-blockers, aspirin, calcium channel blockers, and ACE inhibitors was higher in statin users, indicating higher prevalence of cardiovascular disease and risk factors as compared with nonstatin users (P < 0.0001).
The most common statins prescribed were simvastatin (69% of patients), followed by lovastatin (23% of patients), atorvastatin (7% of patients), and others (1%). Statin use resulted in a significant reduction in total and LDL-cholesterol and a small increase in high-density lipoprotein-cholesterol (Table 2). Final values of serum total cholesterol and LDL-cholesterol were lower in statin users than among nonusers.
The effect of statin use on FPG was studied in crude and adjusted models comparing statin to nonstatin users in diabetic and nondiabetic groups. The model was adjusted for baseline differences in FPG and significant covariates remaining in the model which include age, body mass index (BMI), and medication use. Table 3 shows crude and adjusted changes in FPG with and without statins in patients with clinically evident diabetes. The change in FPG was 39 mg/dL in statin users compared with 29 mg/dL in nonstatin users, the difference that was highly significant (P < 0.0001). When adjusted for baseline differences and significant covariates (age, BMI, and other medications), the difference in FPG change was still highly significant, 39 mg/dL in statin users compared with 32 mg/dL in nonstatin users (P < 0.0001; Fig. 1).
Table 4 shows the changes in FPG levels with and without statins for patients without diabetes. The change in FPG was 7 mg/dL in statin users compared with 4 mg/dL in nonstatin users (P < 0.0001). When adjusted for baseline differences and significant covariates (age, BMI, and other medications), the difference in FPG change for 2 groups was still highly significant (P < 0.0001; Fig. 1).
The present analysis from a very large patient database shows that routine use of statins significantly increases FPG in patients with and without diabetes. It is of note that almost 19,000 patients receiving statins were having diabetes. The mean increase in FPG in patients with diabetes was 10 mg/dL greater in statin users versus nonstatin controls. The increase in FPG with statin use was not restricted to patients with diabetes because we observed a mean 3 mg/dL increase in FPG in patients without diabetes (approximately 325,000) over the 2-year period of follow up.
The effect of statins on the risk of developing clinical or subclinical diabetes mellitus (abnormal glycemic control) has previously been addressed only in a few randomized controlled trials because the impact on overall cardiovascular events was the primary end point. The West of Scotland Coronary Prevention Study trial included 6595 men randomized to either pravastatin 40 mg/d or placebo. Among the 5974 nondiabetic men, 139 patients became diabetic during the 3.5 to 6.1 years of follow up. After adjustment for BMI, blood glucose, triglycerides, and other characteristics, patients assigned to pravastatin had a hazards ratio of 0.70 (when compared with placebo) in terms of transition to diabetes.9The Pravastatin or Atorvastatin Evaluation and Infection Therapy study randomized patients with history of acute coronary syndrome to either pravastatin 40 mg/d versus atorvastatin 80 mg/d. It is of note that the atorvastatin-treated patients, who had a greater decline in total and LDL-cholesterol than pravastatin-treated patients, had a higher risk of developing HbA1C greater than 6% compared with the pravastatin group.6In the MRC/BHF Heart Protection Study,1020,536 patients with and without diabetes were randomized to receive simvastatin 40 mg/d or placebo and followed up for almost 5 years. Among the 14,573 subjects without diabetes, 4.6% developed new diabetes in the simvastatin group developed versus 4.0% in the placebo group (P = 0.1). Furthermore, among the random sample of nearly 1087 diabetic patients at the baseline in this study, HbA1c levels increased in both simvastatin (0.15%) and placebo (0.12%) treated patients, although the difference did not reach statistical significance.3In the Anglo-Scandinavian Cardiac Outcomes Trial-Lipid Lowering Arm trial, there was no difference in the development of diabetes mellitus between the atorvastatin and placebo groups of patients with hypertension followed for a mean of 3.3 years.11In the Collaborative Atorvastatin Diabetes Study trial, no difference was noted between the atorvastatin and placebo groups with respect to changes in HbA1C levels or the need for insulin as therapy for diabetes mellitus.5However, the strongest evidence in favor of an adverse effect of statin therapy on glucose metabolism comes from the recently reported JUPITER study,12a randomized placebo-controlled primary prevention trial that addressed cardiovascular events and mortality in nondiabetic patients with "normal range" LDL(<130 mg/dL) and elevated hsCRP (level >2 mg/L). Over a follow-up period of 1.9 years, patients in the rosuvastatin arm (n = 8900) experienced a small, but significantly, greater increase in incident diabetes (3% vs 2.4% among 8900 placebo-treated patients), and higher HbA1C values (mean 5.9 vs 5.8 mg/dL).
The present study is the largest to date to evaluate the impact of statin therapy on FPG levels in patients with and without diabetes. In this "real world" patient population, FPG levels increased significantly in both groups of patients over the follow up period. The increase in FPG levels was observed to be greater in patients with diabetes than in nondiabetic patients (7 and 3 mg/dL, respectively). Importantly, statin-treated patients had a significantly greater increase in FPG. Our findings of an increase in FPG are concordant with the atorvastatin arm of the PROVE-IT substudy,6but at variance from the results of some of the other studies.5,10,11There are a number of possible explanations for this discordance. Most of the aforementioned statin trials were designed primarily to evaluate effect of statin therapy on cardiovascular morbidity and mortality. These trials were not specifically designed to evaluate the effect of statins on FPG, and the number of patients in whom this issue was specifically examined by measuring FPG repeatedly was very small. Stringent inclusion and exclusion criteria that are essential to derive clear conclusions also deny the possibility of studying an important subset of real world patients that does not meet them. That is why retrospective observational studies like ours have potential to offer a new insight that could not be gained otherwise.
Because of the nature of our data, we were unable to examine any impact on the frequency of clinically evident diabetes or alterations in drug therapy in our subjects. Notably, the patients were followed up for over 2 years in this observational study; this is a rather short period considering the natural history of diabetes. We do not know if this association would continue, worsen, or get better over a longer period of follow up. The biological significance of elevated FPG identified in this study remains unknown, but data from the JUPITER trial12would suggest that the abnormality in FPG may translate into the clinical syndrome of diabetes with a rise in HbA1C. In addition, it has been shown that there is a graded relationship between FPG and the extent of coronary artery disease severity.13Current estimates indicate that most individuals, up to 70%, with prediabetic status may eventually develop clinical diabetes mellitus.14Numerous longitudinal studies indicate that impaired FPG is associated with a modest increase in hazards ratio (1.1-1.4) for cardiovascular disease.15It is noteworthy that in the adjusted model in our current study, the mean FPG value in the nondiabetic statin users increased from 97 to 104 mg/dL which is in the impaired fasting glucose range (100-125 mg/dL). The worsening of FPG in patients with diabetes was even more pronounced.
The precise mechanism/s by which statins may exert any influence on glucose metabolism are unclear. Statins have the potential to alter glycemic control by decreasing various metabolites such as isoprenoid, farnesyl pyrophosphate, geranylgeranyl pyrophosphate, and ubiquinone (CoQ10), all of which are dependent on mevalonic acid production. Isoprenoid in particular enhances glucose uptake via GLUT-4 in adipocytes.16Reduction in CoQ10 may result in delayed ATP production in pancreatic beta cells, and thereby impair insulin release.17Simvastatin has been shown to inhibit glucose-induced increase in intracellular calcium in pancreatic beta cells leading to inhibition of insulin secretion in a dose-dependent manner.18
Limitations and Advantages of Data Mining
There are a number of drugs like hydrochlorothiazide, niacin, and steroids besides statins that could have affected FPG levels in this study population. Important among them are diuretics. Unfortunately, data on the number of patients taking diuretics (thiazides in particular, and their dose) were not available. Furthermore, the number of female patients was small because of the nature of Veterans Afairs population demographics. The reproducibility or variability of FPG measurements could have been a confounder. However, the fact that FPG rise was greater in diabetic patients as well as nondiabetic patients with statin use suggests that the alterations in FPG were not a chance observation. One could also argue that providers have waited too long to start statin therapy in diabetic patients and the rise in FPG is a natural disease progression, however, the significant rise in FPG that also happened in patients without diabetes suggests that this cannot be the case. Our study was based on data mining which has inherent limitations. The main disadvantage of data mining is that the information is only as good as the data that goes into the database. The mining procedure itself is fraught with pitfalls. Because of the enormous sample size, even the smallest effect, though not clinically meaningful, may be highly statistically significant. On the other hand, a major advantage of data mining-based studies is that unbiased information can be obtained on data sets that are so large that they would be inconceivable in a traditional patient-based setting. Thus, data mining offers an opportunity to examine clinically important questions with unparalleled statistical power. In the present study, the aggregate data were validated by querying the database in different ways to obtain the same information.
Our observational study in a very large number of Veteran patients demonstrated that statin therapy is associated with rise of FPG levels over a mean duration of 2 years in patients with and without diabetes. Considering the prevalence of statin usage in current day practice, this important observation merits further investigation. Clinicians should be cognizant of this potential adverse association of statin use on FPG, and careful monitoring is advised.