Diabetes Care Volume 37, August 2014 2225

Giovanni Corrao,1 Buthaina Ibrahim,1

Statins and the Risk of Diabetes: Federica Nicotra,1 Davide Soranna,1
Luca Merlino,2 Alberico L. Catapano,3,4
Evidence From a Large Elena Tragni,3 Manuela Casula,3
Guido Grassi,4,5 and Giuseppe Mancia5,6
Population-Based Cohort Study
Diabetes Care 2014;37:2225–2232 | DOI: 10.2337/dc13-2215

EPIDEMIOLOGY/HEALTH SERVICES RESEARCH

OBJECTIVE
To investigate the relationship between adherence with statin therapy and the
risk of developing diabetes.

RESEARCH DESIGN AND METHODS

The cohort comprised 115,709 residents of the Italian Lombardy region who were
newly treated with statins during 2003 and 2004. Patients were followed from the
index prescription until 2010. During this period, patients who began therapy with
an antidiabetic agent or were hospitalized for a main diagnosis of type 2 diabetes
were identified (outcome). Adherence was measured by the proportion of days
covered (PDC) with statins (exposure). A proportional hazards model was fitted to
estimate hazard ratios (HRs) and 95% CIs for the exposure-outcome association,
after adjusting for several covariates. A set of sensitivity analyses was performed
to account for sources of systematic uncertainty.

RESULTS
During follow-up, 11,154 cohort members experienced the outcome. Compared
1
with patients with very-low adherence (PDC <25%), those with low (26–50%), Department of Statistics and Quantitative
Methods, Division of Biostatistics, Epidemiology
intermediate (51–75%), and high (‡75%) adherence to statin therapy had HRs
and Public Health, University of Milano-Bicocca,
(95% CIs) of 1.12 (1.06–1.18), 1.22 (1.14–1.27), and 1.32 (1.26–1.39), respectively. Milan, Italy
2
Operative Unit of Territorial Health Services, Re-
CONCLUSIONS gion Lombardia, Milan, Italy
3
In a real-world setting, the risk of new-onset diabetes rises as adherence with Department of Pharmacological and Biomolec-
ular Sciences, Centre of Epidemiology and Pre-
statin therapy increases. Benefits of statins in reducing cardiovascular events ventive Pharmacology (SEFAP), University of
clearly overwhelm the diabetes risk. Milano, Milan, Italy
4
IRCCS Multimedica, Sesto San Giovanni, Milan,
Italy
Hydroxymethylglutaryl-CoA reductase inhibitors, known as statins, are the most 5
Department of Health Science, University of
effective drugs for the treatment of hypercholesterolemia (1), and their important Milano-Bicocca, Milan, Italy
6
role in primary and secondary prevention of cardiovascular (CV) morbidity and IRCCS Istituto Auxologico Italiano, Milan, Italy
mortality is well established (2), even among patients with type 2 diabetes (3). Corresponding author: Giovanni Corrao, giovanni.
However, during the past decade, concern has been raised about the use of statins corrao@unimib.it.
and the development of diabetes (4). This did not appear to be the case in the post Received 19 September 2013 and accepted 17
hoc analysis of one of the earliest trials on the CV protective effects of statins, the April 2014.
WOSCOPS (West of Scotland Coronary Prevention Study), which suggested that This article contains Supplementary Data online
at http://care.diabetesjournals.org/lookup/
pravastatin might reduce the risk of diabetes (5). In the more recent JUPITER (Jus-
suppl/doi:10.2337/dc13-2215/-/DC1.
tification for the Use of Statins in Primary Prevention: an Intervention Trial Evalu-
© 2014 by the American Diabetes Association.
ating Rosuvastatin) trial, rosuvastatin was found to be associated with an increased Readers may use this article as long as the work
risk of physician-reported diabetes as well as with a small, though significant, in- is properly cited, the use is educational and not
crease in HbA1c levels (6). Two meta-analyses of clinical trials found that among for profit, and the work is not altered.
2226 Statins and the Risk of Diabetes Diabetes Care Volume 37, August 2014

patients treated with statins, the risk of individuals. Of these, we identified pa- therapy was assessed as the cumulative
developing diabetes was higher than in tients for whom at least one prescrip- number of days during which the medi-
those treated with placebo (7,8). It tion of statins was dispensed during cation was available divided by the num-
should be noted, however, that none 2003 and 2004, and the first dispensa- ber of days of follow-up, a quantity
of these trials were designed to look tion was defined as the index prescrip- referred to as the proportion of days
for diabetes and that the meta-analyses tion. To make data relevant to the study covered (PDC) (11). Patients were cate-
used a range of methods to detect the aim, four patient categories were ex- gorized as having very-low (PDC ,25%),
condition. cluded: 1) patients who received one low (PDC 25–49%), intermediate (PDC
Because statins are among the most or more prescriptions of a statin within 50–74%), and high (PDC $75%) adher-
commonly prescribed drugs and diabe- 3 years before the index prescription, 2) ence. To avoid the arbitrary nature of
tes affects a substantial and increasing patients who received at least one anti- this categorization, in a secondary analy-
proportion of the population world- diabetic agent or were hospitalized for a sis, we used quartile categories of PDC for
wide, efforts aimed at elucidating the primary or secondary diagnosis of dia- the entire period of follow-up to define
magnitude of the association between betes within 3 years before the index increasing levels of exposure to statins.
use of statins and diabetes risk have ma- prescription, 3) patients who did not PDC categories were also considered
jor implications for public health. To reach at least 1 year of follow-up, and according to the potency of the statins
address this issue, we used the data pro- 4) patients who received only one dis- dispensed during follow-up. Based on a
vided by the health-care utilization da- pensation of statins during the first year systematic review and meta-analysis of
tabases of Lombardy, a region of Italy after the date of index prescription. The randomized controlled trials that quan-
with ;10 million inhabitants. This large remaining patients represented the tified the effect of statins on serum LDL
cohort study investigated whether in- study cohort. Each member of the co- cholesterol concentration (12), treat-
creasing levels of adherence with statin hort accumulated person-years of ment with high-potency statins was de-
therapy increases the risk of developing follow-up from the date of index pre- fined $10 mg rosuvastatin, $20 mg
physician-diagnosed diabetes and esti- scription until the earliest among the atorvastatin, and $40 mg simvastatin,
mated the magnitude of the dose- dates of new-onset diabetes (see next whereas all other statin treatments
response relationship. Controlling for section) or censoring (i.e., death from were defined as low potency.
sources of systematic uncertainty was any cause, emigration, or 31 December
of particular concern in this study. 2010 [end of follow-up]). Covariates
For each cohort member, data also in-
RESEARCH DESIGN AND METHODS Outcome Identification
cluded 1) sex, age, and type of statins
Health-Care Utilization Database of Cohort members who experienced at
(atorvastatin, fluvastatin, pravastatin,
Lombardy least one sign suggestive of the occur-
rosuvastatin, and simvastatin) at the in-
The data used for the current study rence of diabetes during follow-up were
dex prescription; 2) concomitant use of
were retrieved from the health-care uti- identified. The date of the first dispen-
other drugs during follow-up (antihyper-
lization databases of Lombardy, a region sation of an antidiabetic agent (Anatom-
tensive, nitrate, digitalis, antiarrhythmic,
of Italy that accounts for ;16% of Italy’s ical Therapeutic Chemical Classification
corticosteroid, oral contraceptive, cyclo-
population. The Italian population is System code A10) or of hospitalization
sporine, antipsychotic, and antidepres-
covered by the National Health Service with a primary diagnosis of diabetes
sant agents); 3) hospitalizations for CV
(NHS), and in Lombardy, an automated (ICD-9 code 250) was assumed as the
diseases in the 3 years before the index
system of databases collects a variety of date of outcome onset, whichever was
prescription; 4) the Charlson comorbidity
information, including 1) an archive of earlier. However, because of the possi-
index score (13), which was calculated
residents who receive NHS assistance, bility that one isolated dispensation of
from the diagnostic information available
reporting demographic and administra- an antidiabetic agent might be an inap-
from inpatient charts in the 3 years be-
tive data; 2) a database on diagnosis at propriate initial drug treatment, in a sec-
fore and 1 year after the date of the index
discharge from public or private hospi- ondary analysis, we used a more-specific
prescription; and 5) the number of serum
tals of the region; 3) a database on out- diagnostic criterion by requiring that at
cholesterol tests and specialist visits dur-
patient drug prescriptions reimbursable least three antidiabetic prescriptions be
ing the first year after the date of the in-
by the NHS; and 4) a database on out- dispensed to consider a patient as hav-
dex prescription.
patient visits, including visits in special- ing the outcome.
ist ambulatory care and diagnostic Assessment of Exposure to Statins Data Analysis
laboratories accredited by the NHS. For We identified all prescriptions dis- The x2 test was used for the trend and
each patient, we linked these databases pensed to the cohort members during linear regression model when appropri-
through a single identification code. Full follow-up. The period covered by a pre- ate to test differences in demographic
details of the procedures are reported scription was calculated from the num- and clinical characteristics with increas-
elsewhere (9). ber of tablets in the dispensed canisters, ing levels of statin adherence. The inci-
Cohort Selection and Follow-up assuming a treatment schedule of one dence rate of diabetes was calculated
The target population comprised all NHS tablet per day (10). For overlapping pre- by the ratio between the number of
beneficiaries aged 40–80 years living in scriptions, the individual was assumed cases of incident diabetes and the num-
Lombardy. According to the 2011 Italian to have completed the former one be- ber of person-years accumulated during
census, this population was 5,097,075 fore starting the second. Adherence to follow-up.
care.diabetesjournals.org Corrao and Associates 2227

Cox proportional hazards regression Finally, although our estimates were one in five patients and ;6% of patients
was used to estimate the hazard ratio adjusted for a number of factors and did not have a serum cholesterol test or
(HR) and 95% CI for the association be- because relevant clinical features were attend a specialist visit, respectively.
tween exposure to statins and time of di- not available in the databases, we ad- During follow-up, 11,154 cohort
abetes onset. The predictor variables of dressed the potential bias generated members experienced the study out-
interest were the factors constructed ac- by unmeasured confounders. For exam- come, with an incidence rate of 14.9
cording to the categories of PDC, using the ple, we considered obesity as a poten- new cases of diabetes per 1,000 person-
very-low-adherence category as refer- tially important unmeasured confounder years. More than one-half (57.3%)
ence. Because drug exposure may vary because obesity is a major predictor of had very-low or low adherence to statin
over time, assessment of its value requires diabetes (15) and can influence adher- therapy. Very-low or low adherence
proper accounting for the cumulative and ence to statin therapy (16,17), which was more common in women; in patients
varying nature of the measure. This was means that failing to adjust for its effects who began therapy with fluvastatin
done by including adherence categories can move the results toward or away or pravastatin; and among patients who
as time-dependent variables in the model. from the null. We made a quantitative did not use other drugs, were not hos-
The relationship of the risk of diabe- assessment of the obesity-related bias pitalized previously for CV disease, or
tes with increasing exposure to statins by using the Monte Carlo sensitivity anal- did not have other chronic comorbidities.
as a whole as well as to low-potency and ysis (18). We set 1) the prevalence of the Patients with a higher number of serum
high-potency statins separately was as- obesity among users of statins to 37.5% cholesterol tests and specialist visits had
sessed using both unadjusted and ad- (19), 2) the risk of diabetes in obese pa- higher adherence to statin therapy.
justed HRs. Adjustment included these tients to threefold higher than in normal
Adherence to Statin Therapy and Risk
reported covariates as time-fixed (i.e., weight patients (15), and 3) the odds of
of Diabetes
those measured at index prescription) obese patients belonging to the highest
There was a continuous and significant
or time-dependent (concomitant use PDC category from twofold lower to two-
trend toward an increase of diabetes
of other drugs) factors. Trends in HRs fold higher than that of normal weight
risk as adherence to statin therapy in-
were tested, when feasible, according to patients (16,17). The Monte Carlo sensi-
creased both in the unadjusted and in
the statistical significance of the regres- tivity analysis corrected the observed HRs
the adjusted risk models (58% [95%
sion coefficient of the recorded variables for the bias factor calculated from the
CI 51–66%] vs. 32% [26–39%] for high
obtained by scoring the corresponding previously mentioned data and took
adherence vs. very-low adherence, re-
categories. The null hypothesis of equal- into account random uncertainty for ad-
spectively) (Fig. 1A and B). Although
ity in the regression coefficients of low- justing estimates. Full details on the
high-potency statins showed a greater
potency and high-potency statins was Monte Carlo sampling procedure and
action on diabetes risk than low-potency
tested by the z test. computing method are reported else-
statins (Fig. 1C and D), there was no sta-
where (20). For all hypotheses tested, a
tistical evidence that the effect of statins
Sensitivity Analyses two-tailed P , 0.05 was considered
on diabetes risk differed according to
To verify the robustness of the findings, significant.
their potency (Ptrend = 0.372).
the following sensitivity analyses were
RESULTS The relationship did not substantially
performed: 1) As mentioned previously,
change by 1) varying the criteria for cat-
we adopted various ways for categoriz- Patients
egorization of adherence with statin
ing exposure to statins (predefined or The distribution of the exclusion crite-
therapy or 2) considering a more de-
quartile categories of PDC) or for iden- ria is shown in Supplementary Fig. 1.
manding criterion for diabetes diagno-
tifying incident diabetes (one to three The 115,709 patients included in the
sis. In particular, taking the first category
prescriptions of antidiabetic drugs dur- study cohort accumulated 748,049
of exposure as the reference, the HR
ing follow-up), and 2) we checked the person-years of observation, with an
(95% CI) of diabetes increased to 1.11
possibility that our estimates were af- average follow-up of ;6.4 years per
(1.00–1.23), 1.23 (1.16–1.31), and 1.37
fected by detection bias (14) (i.e., that patient.
(1.29–1.45) by increasing quartiles of
long-term exposure to statins and a Table 1 shows the characteristics of
PDC and to 1.19 (1.11–1.27), 1.35
more-regular prescription renewal the entire cohort as well as stratification
(1.26–1.45), and 1.53 (1.44–1.64) by
imply a more regular use of primary by adherence to statin therapy. At the
the alternative criterion for diabetes
care services, triggering the search for index date, the mean age was ;62
diagnosis.
diabetes). Detection bias was investi- years, and 49% were men. Almost one
gated by estimating the statin–incident in three patients began therapy with Detection Bias
diabetes relationship in the cohort atorvastatin or simvastatin. Most pa- The combined action of adherence to
members stratified according to the tients were cotreated with antihyper- statin therapy and frequency of choles-
number of cholesterol tests and outpa- tensive drugs, whereas cotreatment terol tests and outpatient specialist vis-
tient specialist visits during the first year with other drugs was less frequent. its on the risk of diabetes is shown in Fig.
of follow-up. The rationale is that the More than 30% of the cohort was hos- 2. Among patients with very-low adher-
relationship would move toward that pitalized previously for CV diseases. ence to statin therapy, the risk of diabe-
expected under the null among patients Other chronic comorbidities were iden- tes decreased with the number of serum
who did not use laboratory analysis tified in more than one in four patients. cholesterol tests but increased as the
services and/or attend specialist visits. During the first year of follow-up, almost number of specialist visits increased.
2228 Statins and the Risk of Diabetes Diabetes Care Volume 37, August 2014

Table 1—Selected characteristics of cohort members according to categories of adherence to statin therapy
PDC category
,25% 25–50% 50–75% $75% Entire cohort
All patients 42,921 (37.1) 23,357 (20.2) 21,251 (18.4) 28,180 (24.4) 115,709 (100)
Age, mean (SD) 62.0 (9.9) 62.7 (9.3) 62.5 (9.1) 62.7 (9.0) 62.4 (9.4)
Male sex 17,425 (40.6) 10,533 (45.1) 10,873 (51.2) 17,427 (61.8) 56,258 (48.6)
First-line therapy with statins
Atorvastatin 13,601 (31.7) 7,344 (31.4) 6,859 (32.3) 9,938 (35.3) 37,742 (32.6)
Fluvastatin 5,311 (12.4) 2,224 (9.5) 2,105 (9.9) 2,613 (9.3) 12,253 (10.6)
Pravastatin 7,539 (17.6) 4,173 (17.9) 3,627 (17.1) 4,135 (14.7) 19,474 (16.8)
Rosuvastatin 3,507 (8.2) 2,040 (8.7) 1,811 (8.5) 2,238 (7.9) 9,596 (8.3)
Simvastatin 12,963 (30.2) 7,576 (32.4) 6,849 (32.2) 9,256 (32.8) 36,644 (31.7)
Concomitant use of other drugs
Antihypertensive drugs
ACE inhibitors 16,935 (39.5) 10,588 (45.3) 10,701 (50.4) 16,271 (57.7) 54,495 (47.1)
ARBs 11,572 (27.0) 6,935 (29.7) 6,663 (31.4) 8,933 (31.7) 34,103 (29.5)
CCBs 11,756 (27.4) 7,390 (31.6) 7,184 (33.8) 10,738 (38.1) 37,068 (32.0)
b-Blockers 13,211 (30.8) 9,335 (40.0) 9,819 (46.2) 15,881 (56.4) 48,246 (41.7)
Thiazides 171 (0.4) 121 (0.5) 116 (0.5) 187 (0.7) 595 (0.5)
Others 10,394 (24.2) 6,400 (27.4) 6,120 (28.8) 9,081 (32.2) 31,995 (27.7)
Nitrates and/or digitalis 4,265 (9.9) 3,391 (14.5) 3,915 (18.4) 7,751 (27.5) 19,322 (16.7)
Antiarrhythmics 2,595 (6.0) 1,772 (7.6) 1,725 (8.1) 2,770 (9.8) 8,862 (7.7)
High-dose corticosteroids 6,755 (15.7) 3,693 (15.8) 3,237 (15.2) 4,224 (15.0) 17,909 (15.5)
Oral contraceptives 2,268 (5.3) 1,148 (4.9) 917 (4.3) 860 (3.1) 5,193 (4.5)
Cyclosporine 147 (0.3) 81 (0.3) 75 (0.4) 123 (0.4) 426 (0.4)
Antipsychotics 1,082 (2.5) 487 (2.1) 357 (1.7) 470 (1.7) 2,396 (2.1)
Antidepressants 7,855 (18.3) 4,177 (17.9) 3,580 (16.8) 4,399 (15.6) 20,011 (17.3)
History of CV disease 10,848 (25.3) 7,622 (32.6) 8,168 (38.4) 14,967 (53.1) 41,607 (36.0)
Charlson comorbidity index score
0 35,712 (83.2) 18,200 (77.9) 15,515 (73.0) 16,857 (59.8) 86,284 (74.6)
1 3,716 (8.7) 2,844 (12.2) 3,350 (15.8) 6,898 (24.5) 16,808 (14.5)
2 1,985 (4.6) 1,310 (5.6) 1,321 (6.2) 2,487 (8.8) 7,103 (6.1)
$3 1,508 (3.5) 1,003 (4.3) 1,065 (5.0) 1,938 (6.9) 5,514 (4.8)
Number of cholesterolemia tests
0 11,639 (27.1) 4,760 (20.4) 4,094 (19.3) 5,696 (20.2) 26,189 (22.6)
1 15,365 (35.8) 8,252 (35.3) 7,272 (34.2) 9,725 (34.5) 40,614 (35.1)
2 9,966 (23.2) 6,331 (27.1) 5,971 (28.1) 7,594 (26.9) 29,862 (25.8)
$3 5,951 (13.9) 4,014 (17.2) 3,914 (18.4) 5,165 (18.3) 19,044 (16.5)
Number of outpatient specialist visits
0 3,310 (7.7) 1,424 (6.1) 1,151 (5.4) 1,319 (4.7) 7,204 (6.2)
1 4,091 (9.5) 1,923 (8.2) 1,597 (7.5) 2,024 (7.2) 9,635 (8.3)
2 4,390 (10.2) 2,226 (9.5) 1,927 (9.1) 2,432 (8.6) 10,975 (9.5)
$3 31,130 (72.5) 17,784 (76.1) 16,576 (78.0) 22,405 (79.5) 87,895 (76.0)
New cases of diabetes
n 3,128 2,463 2,260 3,303 11,154
Rate (every 1,000 person-years) 11.1 16.5 16.6 18.4 14.9
Data are n (%) unless otherwise indicated. ARB, angiotensin receptor blocker; CCB, calcium channel blocker.

Of note, evidence of a trend toward in- treatment adherence (confounder- adherence category had a prevalence
creasing risk of diabetes as adherence to exposure odds ratio [OR]). As expected, odds 1.84-fold higher than those in the
statin therapy increased was also ob- the trend in exposure-outcome HR was highest adherence category). However,
served in patients who did not receive progressively driven toward the null as the point estimate of exposure-outcome
cholesterol tests (top panel) or outpa- the confounder-exposure OR increased association was never annulled, even for
tient specialist visits (bottom panel). (i.e., as the confounder prevalence pro- the highest considered disparity of treat-
gressively increased among patients ment adherence between obese and
Unmeasured Confounding with high adherence). The significant normal weight patients.
Figure 3 shows the risk of diabetes associ- exposure-outcome excess risk associ-
ated with treatment adherence (exposure- ated with high adherence observed in CONCLUSIONS
outcome HR) after adjustment for an the main analysis became nonsignifi- The results show that compared with
unmeasured confounder expected to cant (1.13 [1.00–1.28]) whether the patients who used statins for a small
be from twofold less to twofold more confounder-exposure OR was $1.84 portion of the follow-up period, those
frequent among patients with high (e.g., obese patients in the lowest who exhibited a more continuous use
care.diabetesjournals.org Corrao and Associates 2229

Figure 1—Effect of adherence to statin therapy on the HRs for diabetes. HR was estimated according to Cox proportional hazards model. A and B:
Crude and adjusted estimates of all statins. C and D: Adjusted estimates of high-potency and low-potency statins. Adjustments were made for age
(continuous), sex, first-line statin therapy, concomitant use of other drugs, history of CV disease, and categories of the Charlson comorbidity index
score (see Table 1 for the complete list of covariates included in the model).

of statins had a 32% excess risk of new- Meta-analyses of clinical trials, however, lines (26) and insulin sensitivity (27), re-
onset diabetes. Furthermore, the results showed a weaker increased risk of new- spectively. Simvastatin and atorvastatin
show that an increased risk of new- onset diabetes by ;10% (7,8,23). Finally, have been shown to decrease insulin se-
onset diabetes is related to all available the SPARCL (Stroke Prevention by Ag- cretion in b-cells (28). The inhibition of
statins at all doses. The results failed to gressive Reduction in Cholesterol Levels) isoprenoid synthesis may explain some
show, however, significant evidence trial showed that high-dose atorvastatin of the observed dysglycemic effects of
that high-potency statins exert a stron- treatment compared with placebo is associ- statins (26). Finally, it has been hypoth-
ger action on diabetes development ated with a 37% excess risk of diabetes (24). esized that statins may influence muscle
than low-potency statins. In summary, The current findings differ from those or liver insulin sensitivity directly, but
this population-based study supports of a large cohort study, bolstering the there is no specific experimental evi-
the notion that continuous use of statins suggestion that statin-induced diabetes dence to support this hypothesis (29).
is associated with a nonmarginal in- is likely a class effect (8), and of a meta-
Strengths and Weaknesses
crease in the risk of developing diabetes analysis of clinical trials showing that in-
The current study is unique in several
in a real-life setting. tensive- versus moderate-dose statin
respects. First, the investigation was
therapy is associated with a greater
Comparison With Available Evidence based on data from a very large unse-
risk of diabetes (25). Two large random-
The current findings confirm and extend lected population, which was made pos-
ized trials, however, consistently showed
the results of the JUPITER trial, which sible because of a cost-free health-care
that atorvastatin 80 mg daily is associated
was the first to find that new-onset di- system for virtually all Italian citizens. Sec-
with a weak and nonsignificant excess of
abetes was 27% more common in pa- ond, the drug prescription database pro-
diabetes compared with atorvastatin
tients treated with statins than in vides highly accurate data because report
10 mg daily (HR 1.10 [0.94–1.29]) or sim-
those receiving placebo (5), as well as of prescriptions by the pharmacies is es-
vastatin 20 mg daily (HR 1.19 [0.98–
the results of the PROSPER (Prospective sential for reimbursement, and filing of an
1.43]) (24).
Study of Pravastatin in the Elderly at Risk) incorrect report about dispensed drugs
trial, which showed that new-onset dia- Plausibility has legal consequences (30). Third, pa-
betes increased by 30% in the pravastatin The mechanisms underlying the diabe- tients were identified from the point of
group compared with the placebo group togenic influence of statins are incom- the initial lipid-lowering therapy, and the
(21). Even more clear-cut risk excesses pletely understood. Atorvastatin and complete sequence of the subsequent
(+20% and +48%) have been reported simvastatin have been shown to de- prescriptions for statins was available.
from observational investigations (8,19,23). crease glucose uptake in adipocyte cell Fourth, we were able to include patients
2230 Statins and the Risk of Diabetes Diabetes Care Volume 37, August 2014

Figure 2—Combined action of adherence to statin therapy and number of cholesterolemia tests (top panel) and outpatient specialist visits (bottom
panel) on the HR for diabetes. HR was estimated according to Cox proportional hazards model. Estimates are adjusted for age (continuous), sex, first-
line statin therapy, cotreatment with other drugs, history of CV disease, and categories of the Charlson comorbidity index score (see Table 1 for the
complete list of covariates included in the model).

without previous clinical evidence (drug adherence remains a source of uncer- Third, our estimates might be af-
treatment and/or hospitalization) of dia- tainty in our estimates. fected by detection bias; that is, patients
betes so that the data relate to the effect Second, because of privacy regula- with long-term adherence to statin ther-
of statin use on new-onset of diabetes. tions, identification codes of prescrip- apy may have been more likely to
Finally, a number of sensitivity analyses tion records were not available for receive a diagnosis of diabetes. How-
were performed, which increased the ro- analysis, so drug-based diagnoses of ever, we found a clear relationship be-
bustness of the findings. diabetes cannot be scrutinized and vali- tween adherence with statins and risk of
The study has a number of potential dated. However, it seems highly unlikely diabetes, even among patients who did
limitations. First, evaluation of statin ad- that diagnostic errors could differen- not undergo laboratory examinations or
herence was based on pharmacy-dispensing tially affect patients according to their outpatient specialist visits. Thus, the ex-
information. This method assumes that adherence to statins. Moreover, the cess risk of diabetes also concerned pa-
the PDC by a prescription corresponds to adherence-diabetes relationship was tients with low use of health services,
the proportion of days of medication confirmed, and indeed amplified, when a making it unlikely that detection bias
use. Small, insignificant differences be- more stringent diagnostic criterion for explains the main findings.
tween the assessed number of dis- new-onset diabetes was used, which raises Finally, whether the observed find-
pensed pills and the actual pill count the possibility that the statin-dependent ings are a result of our inability to fully
were reported by a study investigating risk of developing diabetes might be account for higher adherence to statin
adherence to statin therapy over 12 greater than that resulting from the pri- therapy in patients at higher risk of di-
months (31). Furthermore, data on dis- mary diagnostic criterion we adopted. It abetes is the more relevant question in
pensing history have been shown to be should also be mentioned that the im- interpreting the findings. Baseline fast-
consistent with other adherence mea- proved diagnostic specificity obtained ing serum glucose and other compo-
sures, drug serum levels, and clinical with the stricter criterion minimizes bias nents of the metabolic syndrome, such
drug effects (32). Nevertheless, the use of a risk estimate that includes diagnos- as triglyceride level, BMI, and hyperten-
of medication dispensing as a measure of tic misclassification. sion (24), might be more frequent in
care.diabetesjournals.org Corrao and Associates 2231

Figure 3—Changes in HRs (and 95% CIs) for the risk of diabetes (outcome) associated with high adherence to statin therapy (exposure) after external
adjustment for obesity (confounder). External adjustment was performed by means of Monte Carlo sensitivity analysis. Adjusted estimates were
obtained by means of setting 1) the prevalence of obesity among statin users to 37.5% (12); 2) the risk of diabetes in obese patients to threefold
higher than in normal weight patients (25); and 3) the odds of obese patients in the highest PDC category from twofold lower to twofold higher than
that of normal weight patients (26,27).

individuals with high rather than low ad- beyond obesity, no other unmeasured Consistent with our calculation, a
herence to statin treatment. Because factor is able to annul the investigated meta-analysis reported that treating
our databases have a limited amount relationship, even if it is characterized 255 patients with statins for 4 years
of clinical information, we dealt with by very-high prevalence (i.e., up to would induce one case of diabetes,
confounding in several ways. First, only 37.5%), strongly affects the outcome but in the meantime, it would prevent
statin users were included in the cohort (i.e., up to threefold increased diabetes 5.4 coronary deaths or myocardial in-
to compare the duration of statin ther- risk), and is markedly more diffuse farctions for each millimolar reduction
apy, whereas inclusion of nonusers among adherent patients (i.e., up to in serum LDL cholesterol (6). Thus, the
would be for observational investiga- twofold more than in those with little CV protection offered by statins ap-
tions (8,19). In this way, the potential adherence to statin therapy). In such pears to markedly outweigh the in-
for confounding is reduced by actively conditions, an increased risk of diabetes creased incidence of diabetes, although
comparing patients with the same indi- weaker than that found in our main the adverse effect of diabetes on
cation at baseline (33). Second, our es- analysis is expected. This possibly ex- CV risk (35) suggests that the original
timates were adjusted for a number of plains the stronger association generally statin-dependent protection might de-
available demographic, therapeutic, and reported by observational investiga- cline with time. The decline might be less
clinical characteristics, such as cotreat- tions than reported in clinical trials. than predicted based on epidemiolog-
ment with antihypertensive and other ical data on diabetes and CV morbidi-
agents, history of CV disease, and cate- Implications for Benefits of Statin ty and mortality, however, because
gories of Charlson comorbidity index Treatment whether the prognostic value of drug-
score. Third, the statin-diabetes rela- Assuming that 1) the diabetes incidence induced diabetes is equivalent to that of
tionship was observed within each rate among patients with very-low ad- native diabetes is still uncertain (36).
stratum of frequency of laboratory ex- herence to statin therapy represents the This has been reported to be the case in
aminations and outpatient specialist vis- baseline rate of new-onset diabetes be- some studies (37,38), whereas in other
its. Because medical attention might be cause diabetes is unlikely to develop studies, a few years or even long-term
considered a proxy of clinical profile and during very-short-term statin use and exposure to diabetes induced by antihy-
other unmeasured risk factors for diabe- 2) diabetes develops in patients with pertensive drugs was not found to in-
tes, this further protects our conclu- high adherence to statins at a rate crease CV morbid or fatal events (24,39).
sions, although residual confounding 1.32-fold faster than the baseline rate,
cannot be excluded. For this reason, we expected that ;280 patients would Conclusion
we also performed a sensitivity analysis have to be continuously treated to This large population-based cohort
and showed that the association be- induce one case of diabetes every study extends earlier findings of an in-
tween adherence to statin treatment year. In a meta-analysis of seven ran- creased risk of diabetes with statin ther-
and the risk of diabetes was not an- domized controlled trials of statin use apy by providing evidence of a clear-cut
nulled after correction for an unmea- versus control in patients with diabetes, association between adherence to sta-
sured confounder of great importance statins decreased major CV and cere- tin therapy and risk of new-onset diabe-
for the development of diabetes, obe- brovascular events every 40 patients tes in a real-world setting. It appears
sity. It should be emphasized that treated (34). from event-based investigations that
2232 Statins and the Risk of Diabetes Diabetes Care Volume 37, August 2014

benefits of statins in reducing CV events 8. Zaharan NL, Williams D, Bennett K. Statins 24. Waters DD, Ho JE, DeMicco DA, et al. Pre-
clearly overwhelm the diabetes risk. and risk of treated incident diabetes in a pri- dictors of new-onset diabetes in patients treated
mary care population. Br J Clin Pharmacol with atorvastatin: results from 3 large random-
2013;75:1118–1124 ized clinical trials. J Am Coll Cardiol 2011;57:
9. Corrao G, Cesana G, Merlino L. Pharmaco- 1535–1545
Duality of Interest. G.M. discloses consultancy epidemiological research and the linking of elec- 25. Preiss D, Seshasai SR, Welsh P, et al. Risk of
agreements with Boehringer Ingelheim and tronic healthcare databases available in the incident diabetes with intensive-dose com-
Novartis and participation in speakers bureaus Italian region of Lombardy. Biomed Stat Clin pared with moderate-dose statin therapy:
for Bayer, Boehringer Ingelheim, Merck Sharp & Epidemiol 2008;2:111–125 a meta-analysis. JAMA 2011;305:2556–2564
Dohme, Manar International, Novartis, Recordati, 10. Helin-Salmivaara A, Lavikainen P, Korhonen 26. Takaguri A, Satoh K, Itagaki M, Tokumitsu Y,
Sanofi, Sankyo, and Servier. No other potential MJ, et al. Long-term persistence with statin Ichihara K. Effects of atorvastatin and pravastat-
conflicts of interest relevant to this article were therapy: a nationwide register study in Finland. in on signal transduction related to glucose up-
reported. Clin Ther 2008;30:2228–2240 take in 3T3L1 adipocytes. J Pharmacol Sci 2008;
Author Contributions. G.C. contributed to the 11. Andrade SE, Kahler KH, Frech F, Chan KA. 107:80–89
initial study idea, interpretation of the results, Methods for evaluation of medication adher- 27. Koh KK, Quon MJ, Han SH, et al. Differential
and drafting of the manuscript. B.I. contributed ence and persistence using automated data- metabolic effects of pravastatin and simvastatin
to the protocol, preparation of the data set for bases. Pharmacoepidemiol Drug Saf 2006;15: in hypercholesterolemic patients. Atherosclero-
analysis, data analysis, and interpretation of the 565–574; discussion 575–577 sis 2009;204:483–490
results. F.N. contributed to the sensitivity 12. Law MR, Wald NJ, Rudnicka AR. Quantifying 28. Ishikawa M, Okajima F, Inoue N, et al. Dis-
analysis and interpretation of the results. D.S., effect of statins on low density lipoprotein cho- tinct effects of pravastatin, atorvastatin, and
E.T., and M.C. contributed to the protocol and lesterol, ischaemic heart disease, and stroke: simvastatin on insulin secretion from a beta-
interpretation of the results. L.M. contributed systematic review and meta-analysis. BMJ cell line, MIN6 cells. J Atheroscler Thromb
to abstracting the data and authorizing their use 2003;326:1423 2006;13:329–335
and the interpretation of the results. A.L.C., 13. Charlson ME, Charlson RE, Peterson JC, 29. Ma T, Tien L, Fang C-L, Liou YS, Jong GP. Statins
G.G., and G.M. contributed to the interpreta- Marinopoulos SS, Briggs WM, Hollenberg JP. and new-onset diabetes: a retrospective longitudi-
tion of the pharmacological and clinical pro- The Charlson comorbidity index is adapted to nal cohort study. Clin Ther 2012;34:1977–1983
spective results and review of the manuscript. predict costs of chronic disease in primary 30. Strom BL. Overview of automated data-
G.C. is the guarantor of this work and, as such, care patients. J Clin Epidemiol 2008;61:1234– bases in pharmacoepidemiology. In Pharmaco-
had full access to all the data in the study and 1240 epidemiology. 3rd ed. Strom BL, Ed. Chichester,
takes responsibility for the integrity of the data 14. Delgado-Rodrı́guez M, Llorca J. Bias. J Epi- U.K., Wiley, 2000, p. 219–222
and the accuracy of the data analysis. demiol Community Health 2004;58:635–641 31. Lee JK, Grace KA, Foster TG, et al. How
15. Lindström J, Tuomilehto J. The diabetes risk should we measure medication adherence in
score: a practical tool to predict type 2 diabetes clinical trials and practice? Ther Clin Risk Manag
References risk. Diabetes Care 2003;26:725–731 2007;3:685–690
1. Catapano AL, Reiner Z, De Backer G, et al.; 16. Hüther J, von Wolff A, Stange D, et al. In- 32. Steiner JF, Prochazka AV. The assessment of
European Society of Cardiology (ESC); European complete medication adherence of chronically refill compliance using pharmacy records: meth-
Atherosclerosis Society (EAS). ESC/EAS guide- ill patients in German primary care. Patient Pre- ods, validity, and applications. J Clin Epidemiol
lines for the management of dyslipidaemias: fer Adherence 2013;7:237–244 1997;50:105–116
The Task Force for the Management of 17. Mazzaglia G, Ambrosioni E, Alacqua M, 33. Schneeweiss S, Avorn J. A review of uses of
Dyslipidaemias of the European Society of Car- et al. Adherence to antihypertensive medica- health care utilization databases for epidemio-
diology (ESC) and the European Atherosclerosis tions and cardiovascular morbidity among logic research on therapeutics. J Clin Epidemiol
Society (EAS). Atherosclerosis 2011;217:3–46 newly diagnosed hypertensive patients. Circula- 2005;58:323–337
2. Baigent C, Keech A, Kearney PM, et al.; Cho- tion 2009;120:1598–1605 34. Chen YH, Feng B, Chen ZW. Statins for pri-
lesterol Treatment Trialists’ (CTT) Collaborators. 18. Greenland S. Sensitivity analysis, Monte mary prevention of cardiovascular and cerebro-
Efficacy and safety of cholesterol-lowering treat- Carlo risk analysis, and Bayesian uncertainty as- vascular events in diabetic patients without
ment: prospective meta-analysis of data from sessment. Risk Anal 2001;21:579–583 established cardiovascular diseases: a meta-
90,056 participants in 14 randomised trials of 19. Culver AL, Ockene IS, Balasubramanian R, analysis. Exp Clin Endocrinol Diabetes 2012;
statins. Lancet 2005;366:1267–1278 et al. Statin use and risk of diabetes mellitus in 120:116–120
3. Kearney PM, Blackwell L, Collins R, et al.; postmenopausal women in the Women’s 35. LaRosa JC, He J, Vupputuri S. Effect of stat-
Cholesterol Treatment Trialists’ (CTT) Collabo- Health Initiative. Arch Intern Med 2012;172: ins on risk of coronary disease: a meta-analysis
rators. Efficacy of cholesterol-lowering therapy 144–152 of randomized controlled trials. JAMA 1999;
in 18,686 people with diabetes in 14 randomised 20. Corrao G, Nicotra F, Parodi A, et al. External 282:2340–2346
trials of statins: a meta-analysis. Lancet 2008;371: adjustment for unmeasured confounders im- 36. Mancia G, Grassi G, Zanchetti A. New-onset
117–125 proved drug-outcome association estimates diabetes and antihypertensive drugs. J Hypertens
4. Colbert JD, Stone JA. Statin use and the risk based on health care utilization data. J Clin Epi- 2006;24:3–10
of incident diabetes mellitus: a review of the demiol 2012;65:1190–1199 37. Aksnes TA, Kjeldsen SE, Rostrup M, Omvik P,
literature. Can J Cardiol 2012;28:581–589 21. Shepherd J, Blauw GJ, Murphy MB, et al.; Hua TA, Julius S. Impact of new-onset diabetes
5. Ridker PM, Danielson E, Fonseca FAH, et al.; PROSPER Study Group. PROspective Study of mellitus on cardiac outcomes in the Valsartan Anti-
JUPITER Study Group. Rosuvastatin to prevent Pravastatin in the Elderly at Risk. Pravastatin hypertensive Long-term Use Evaluation (VALUE)
vascular events in men and women with ele- in elderly individuals at risk of vascular disease trial population. Hypertension 2007;50:467–473
vated C-reactive protein. N Engl J Med 2008; (PROSPER): a randomised controlled trial. Lan- 38. Verdecchia P, Reboldi G, Angeli F, et al. Ad-
359:2195–2207 cet 2002;360:1623–1630 verse prognostic significance of new diabetes in
6. Sattar N, Preiss D, Murray HM, et al. Statins 22. Carter AA, Gomes TG, Camacho X, Juurlink treated hypertensive subjects. Hypertension
and risk of incident diabetes: a collaborative DN, Shah BR, Mamdani MM. Risk of incident 2004;43:963–969
meta-analysis of randomised statin trials. Lan- diabetes among patients treated with statins: 39. Kostis JB, Wilson AC, Freudenberger RS,
cet 2010;375:735–742 population based study. BMJ 2013;346:f2610 Cosgrove NM, Pressel SL, Davis BR; SHEP Collab-
7. Mills EJ, Wu P, Chong G, et al. Efficacy and 23. Rajpathak SN, Kumbhani DJ, Crandall J, orative Research Group. Long-term effect of
safety of statin treatment for cardiovascular dis- Barzilai N, Alderman M, Ridker PM. Statin ther- diuretic-based therapy on fatal outcomes in
ease: a network meta-analysis of 170,255 pa- apy and risk of developing type 2 diabetes: subjects with isolated systolic hypertension
tients from 76 randomized trials. QJM 2011; a meta-analysis. Diabetes Care 2009;32:1924– with and without diabetes. Am J Cardiol 2005;
104:109–124 1929 95:29–35