Article Text

Download PDFPDF

Extended report
Risk of diabetes mellitus associated with disease-modifying antirheumatic drugs and statins in rheumatoid arthritis
  1. Gulsen Ozen1,2,
  2. Sofia Pedro3,
  3. Marie E Holmqvist4,
  4. Michael Avery5,
  5. Frederick Wolfe3,
  6. Kaleb Michaud1,3
  1. 1University of Nebraska Medical Center, Omaha, Nebraska, USA
  2. 2Faculty of Medicine, Marmara University, Istanbul, Turkey
  3. 3National Data Bank for Rheumatic Diseases, Wichita, Kansas, USA
  4. 4Clinical Epidemiology Unit, Department of Medicine, Karolinska Institutet, Stockholm, Sweden
  5. 5Regional Health, Rapid City, South Dakota, USA
  1. Correspondence to Dr Kaleb Michaud, 986270 Nebraska Medical Center, Omaha, NE 68198, USA; kmichaud{at}


Objective To investigate the rate of incident diabetes mellitus (DM) in patients with rheumatoid arthritis (RA) and the impact of disease-modifying antirheumatic drug (DMARD) and statin treatments.

Methods We studied patients with RA and ≥1 year participation in the National Data Bank for Rheumatic Diseases without baseline DM from 2000 through 2014. DM was determined by self-report or initiating DM medication. DMARDs were categorised into four mutually exclusive groups: (1) methotrexate monotherapy (reference); (2) any abatacept with or without synthetic DMARDs (3) any other DMARDs with methotrexate; (4) all other DMARDs without methotrexate; along with separate statin, glucocorticoid and hydroxychloroquine (yes/no) variables. Time-varying Cox proportional hazard models were used to adjust for age, sex, socioeconomic status, comorbidities, body mass index and RA severity measures.

Results During a median (IQR) 4.6 (2.5–8.8) years of follow-up in 13 669 patients with RA, 1139 incident DM cases were observed. The standardised incidence ratio (95% CI) of DM in patients with RA (1.37, (1.29 to 1.45)) was increased compared with US adult population. Adjusted HR (95% CI) for DM were 0.67 (0.57 to 0.80) for hydroxychloroquine, 0.52 (0.31 to 0.89) for abatacept (compared with methotrexate monotherapy), 1.31 (1.15 to 1.49) for glucocorticoids and 1.56 (1.36 to 1.78) for statins. Other synthetic/biological DMARDs were not associated with any risk change. Concomitant use of glucocorticoids did not alter DM risk reduction with hydroxychloroquine (HR 0.69 (0.51 to 0.93)).

Conclusions In RA, incidence of DM is increased. Hydroxychloroquine and abatacept were associated with decreased risk of DM, and glucocorticoids and statins with increased risk.

  • Rheumatoid Arthritis
  • DMARDs (biologic)
  • DMARDs (synthetic)
  • Outcomes research

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Rheumatoid arthritis (RA) is associated with increased cardiovascular (CV) morbidity and mortality1–3 likely due to complex interactions between RA-related inflammatory activity, medications and traditional CV disease (CVD) risk factors.4–6 Among these risk factors, type 2 diabetes mellitus (DM) is one of the most important.7 ,8 Although studies investigating DM prevalence in RA have had inconsistent results,9–12 it has been shown that both inflammatory activity and some RA medications impact glucose metabolism, insulin resistance and consequently DM development.13–16 Key inflammatory cytokines in RA, particularly tumour necrosis factor-α (TNF-α) and interleukin-6 (IL-6), have been associated with increased adiposity and insulin resistance by triggering key steps in the insulin signalling pathways.13 ,15 ,16

Besides affecting inflammation, disease-modifying antirheumatic drugs (DMARDs) may also affect DM risk by directly altering glucose metabolism. However, DM risk modification by DMARDs varies in effect.14 ,17–20 For example, glucocorticoids (GC), despite strong anti-inflammatory actions, may lead to hyperglycaemia and insulin resistance in a dose-dependent and duration-dependent manner.14 Alternatively, hydroxychloroquine (HCQ),21 methotrexate (MTX)18 and TNF inhibitors (TNFi)22 have been shown to favourably alter glucose metabolism. Large epidemiological studies also showed decreased risk of new-onset DM with HCQ in RA.19 ,20 ,23 However, these studies had heterogeneous comparison groups and did not address HCQ duration, dose and cessation for DM risk. Similar issues affected studies showing DM risk reduction in TNFi, and none have examined newer biologics.17 ,19

In addition to well-known risk factors such as obesity and physical inactivity, recent meta-analyses and observational studies found that statin use is also associated with increased DM risk in the general population.24–27 It is unknown whether statins exert the same effect in RA considering its association with decreased RA activity.28 ,29

Given the changing use of treatments and CVD concerns, we sought to investigate the associations of DMARDs, GC and statins with incident DM in patients with RA in a large US-wide observational cohort. We also sought to determine the timing, dosing and sustainability of the observed effects of HCQ in DM risk in patients with RA.

Patients and methods

In a US-wide longitudinal observational cohort, the National Data Bank for Rheumatic Diseases (NDB), participants were mainly recruited from rheumatologists who confirmed the diagnoses. Participants completed semiannual, comprehensive questionnaires as previously described.30 We included patients with RA who completed at least two questionnaires between January 2000 and December 2014, while those with prevalent DM at study entry were excluded. Follow-up continued until the participant reported DM or was censored at death, loss to follow-up or end of study period.

The primary outcome was incident DM defined as patient report of new DM diagnosis or initiating use of an antidiabetic medication. Diagnosis date of DM was assigned a random month of onset during the 6-month period, as done previously.14

Treatment exposure was measured at enrolment and every 6 months with questionnaires.14 Initially, we determined the impact of each DMARD individually on DM risk compared with its non-use (see online supplementary text). From this, we defined four mutually exclusive, hierarchical DMARD groups: (1) MTX monotherapy (reference), (2) any abatacept (ABA) independent of other DMARDs, (3) any other biological (non-ABA) or non-biological DMARDs in combination with MTX, (4) all others. HCQ, GC, statins and non-steroidal anti-inflammatory drugs (NSAIDs) were evaluated separately.

Statistical analysis

Baseline characteristics of the patients with RA with and without incident DM were compared. Crude incidence rates were calculated by dividing the number of events during follow-up by the corresponding person-time at risk. Standardised incidence ratios (SIRs) were estimated using the incidence rates of DM for the United States, reported by the Centers for Disease Control and Prevention, stratified by calendar year, sex and age.31 Since only incidence rates of DM are reported for individuals under 80 years of age, we excluded those older than 80 at NDB entry when estimating SIRs.

We examined the association between drug exposure and incident DM by using multivariable Cox proportional hazards. A goodness of model fit was compared for each DMARD group. The lowest Akaike information criterion32 by far was with the above-hierarchical DMARD groups with additional dichotomous HCQ, GC, NSAIDs and statin variables (see online supplementary text).

The final model also included the following: age, sex, ethnicity, income and employment status, Rheumatic Disease Comorbidity Index,33 hypertension, RA duration (log-transformed), health assessment questionnaire (HAQ), smoking status and body mass index (BMI) categorised according to WHO classification.34 Three-year calendar periods were also evaluated to account for US treatment trends.

In sensitivity analyses, the impacts of HCQ-ever versus HCQ-never use, daily dose and treatment duration, and prednisone daily dose were examined. The association between treatment duration and risk of incident DM was assessed through Poisson regression. DM risk after discontinuation of HCQ was also evaluated with Cox regression models by selecting patients who were on HCQ (new initiators) but discontinued HCQ (at 3rd, 6th and ≥12th months off HCQ) in comparison with HCQ-never-used patients. The HR was estimated for increasing values of off-treatment duration. For comparison, we took the same approach for new users of MTX. The final model including DMARD groups and other covariates was also analysed in patients without history of CVD (ischaemic heart or peripheral arterial disease, cerebrovascular accident or heart failure), that is, statin use for primary prevention. Potential interactions of concomitant use of GC, either with DMARDs or statins, were also tested.

Finally, a marginal structural model was applied to new initiators of HCQ versus MTX to minimise the bias of confounding by indication. Due to the significant reduction in sample size, results were not presented (see online supplementary text).

In all analyses, treatment exposures were time-dependent covariates. In order to prevent bias from removing observations due to missing data, unanswered covariates of completed questionnaires were replaced by using multiple imputation by chained equations to create imputed datasets for analyses35 (annual income had 4% missing, all other variables had <1% missing). For non-consecutive observations (8%), the last observation was carried forward, and incident DM diagnosis was assigned at the beginning of the first non-missing 6-month period. All p values were two-sided, conducted at a significance level of 0.05. All statistical analyses were performed using Stata V.14.0 (StataCorp, College Station, Texas, USA).


After excluding 1456 (9.6%) prevalent baseline DM cases, our study had 13 669 patients with RA, with a baseline mean age of 58.6±13.4 years and disease duration of 14.4±12.4 years. The baseline characteristics of the patients by future incident DM and DMARD exposure groups are presented in table 1 and online supplementary table S1, respectively. During a median (IQR) follow-up of 4.6 (2.5–8.8) years, 1139 (8.3%) incident DM cases were observed.

Table 1

Baseline characteristics, overall and by future incident diabetes*

The overall incidence rate (95% CI) for DM was 1.59 (1.50 to 1.68) per 100 person-years. The incidence rate of DM in RA found to be increased (age-adjusted and sex-adjusted SIR, 1.37 (1.29 to 1.45)) compared with the incidence rates in US adult population.31 The incidence rates in females and males were 1.57 (1.47 to 1.67) and 1.68 (1.48 to 1.91) corresponding to age-adjusted SIRs of 1.39 (1.29 to 1.48) and 1.30 (1.13 to 1.49), respectively. The incidence rates and SIRs by disease activity and DMARD groups, HCQ, GC and statins are shown in table 2.

Table 2

Crude incidence rates (95% CI) and SIRs (95% CI) of diabetes in rheumatoid arthritis by disease activity* and treatment compared with US population

The fully adjusted time-dependent Cox regression models showed a significant DM risk reduction with HCQ (HR 0.67 (0.57 to 0.80)). In comparison with MTX monotherapy, ABA (HR 0.52 (0.31 to 0.89)) was also associated with DM risk reduction in patients with RA (table 3). Alternatively, the risk of incident DM significantly increased with current use of GC (HR 1.31 (1.15 to 1.49)) or statins (HR 1.56 (1.36 to 1.78)). Besides adjustment for differences in statin-exposed and non-exposed patients (see online supplementary table S2), adjusting for other comorbidities separately (thyroid or CVD) did not change the association with statins (data not shown). Other significant factors associated with increased DM risk included lower annual income, non-Caucasian ethnicity, higher BMI and more comorbidity (see online supplementary table S3).

Table 3

Association of different treatments with incident diabetes in patients with RA

Sensitivity analysis indicated both HCQ doses of <400 mg/day (median (IQR): 200 mg (200–300)) and ≥400 mg/day (median (IQR): 400 mg (400–600)) were associated with DM risk reduction, though higher doses were more prominent (table 4). Poisson regression models revealed that DM risk reduction with HCQ started after 2 years of treatment (Relative risk (RR) 0.76 (0.58 to 1.00)), and continued to decrease with longer duration of >4 years (RR 0.69 (0.59 to 0.81)). In patients initiating HCQ (N=686), DM risk reduction was consistent with those with prior HCQ exposure, ≥3 years of HCQ treatment (RR 0.44 (0.23 to 0.86)). Patients who initiated and then discontinued HCQ (N=342) had a non-significant risk reduction up to 6 months compared with HCQ never-used patients: HR 0.65 (0.21 to 2.0) for ≥3 months, 0.88 (0.28 to 2.75) for ≥6 months and 1.27 (0.31 to 5.10) for ≥1 year off-HCQ. In contrast, DM risk after MTX initiation and discontinuation was consistently high compared with MTX never-used patients (HR range 1.5–2.8 (0.63 to 8.35)).

Table 4

Risk of incident diabetes by prednisone dose, HCQ dose, and duration and concomitant glucocorticoid use

When examining concomitant use of GC, either with HCQ, ABA, or statins, DM risk reduction with HCQ remained significant, while that of ABA vanished. Furthermore, the increased risk with statins was potentiated with GC (table 5). Notably, when HCQ and statins were used together, the increased risk with statins disappeared (HR 0.92 (0.68 to 1.25)). After exclusion of patients with prevalent CVD (N=2535), statin use was still associated with an increased risk of incident DM (HR 2.31 (1.86 to 2.87)) (table 5).

Table 5

Association between DMARDs and statins and incident diabetes in patients without cardiovascular disease (N=11,134; incident diabetes N=837 (7.8%))


In this large US-wide observational cohort study, we found HCQ and ABA associated with decreased risk of incident DM, whereas GC and statins were associated with increased risk.

The risks of incident DM in RA with HCQ,14 ,19 ,20 ,23 TNFi17 ,19 and GC14 have been evaluated in a few different settings. Previous observational studies reported that HCQ-ever use was associated with 38%–71% DM risk reduction compared with never use,20 ,23 and current use was associated with 46% risk reduction compared with the use of any non-biological non-MTX DMARD.19 A recent UK administrative database study indicated ∼20% risk reduction with HCQ use versus non-use.14 We also observed a 33% risk reduction with HCQ and a 34% reduction with >4 years of HCQ compared with non-users. This relatively lower risk reduction with HCQ in the UK and our cohort compared with previous studies may be due to differing demographics, comparison groups and adjustment for different disease, treatment and diabetes-related covariates (NSAIDs, statins, newer biologics, HAQ and BMI). Finally, more prescription of HCQ to patients with greater DM risk after first publication of this effect may also be a reason for the difference, although we did not observe any trends of HCQ use or effect by calendar year.

From the clinical perspective, determining the minimum dose and duration of HCQ for DM risk reduction and the effects after cessation is also important. We found lower daily doses (<400 mg) of HCQ also decreased DM risk after at least 2 years of treatment (63% of HCQ-exposed were on >2 years of treatment, median (IQR) duration=49 (12–134) months). Previously, it was shown that higher cumulative doses, reflecting longer treatment duration, were associated with DM risk reduction;20 ,36 however daily protective doses were not addressed. Improvement in insulin sensitivity has been reported to start within 8 weeks of HCQ treatment in healthy individuals,37 ,38 although this short-term improvement has not been shown in stable patients with RA.39 The only study evaluating the association of DM risk with HCQ treatment duration reported that >4 years of treatment was required for risk reduction.20 The reason for the difference may be due to lower number of patients, and consequently lower incident DM cases in the group with >4 years HCQ treatment in the previous study (384 of 1808 HCQ-ever used, 21%).20 In this study, we also showed for the first time that DM risk reduction associated with HCQ tends to continue up to 6 months after cessation of HCQ. This finding is important because it indicates that HCQ offers long-lasting beneficial metabolic effects beyond its anti-inflammatory actions. Future studies are warranted to confirm this finding, as the number of patients who initiated and discontinued HCQ was substantially lower (N=342).

The mechanisms underlying the DM risk reduction with HCQ are usually explained by improvement in insulin sensitivity and pancreatic β-cell functions37 ,38 which may be independent of anti-inflammatory actions. The improvement of adiponectin levels without significant change in serum inflammatory cytokines (TNF-α, IL-6) with HCQ in non-diabetic individuals38 and less significant DM risk reduction with more potent DMARDs support this hypothesis.19

Concerning other DMARDs, MTX was associated with a slight decrease in glycosylated haemoglobin in diabetic patients with RA.18 However, a large cohort study revealed no DM risk reduction with MTX compared with other non-biological DMARDs.19 We observed a decreased DM risk with MTX when only compared with DMARD non-users in multivariate analysis. As the gold standard therapy for RA, ‘MTX monotherapy’ was chosen as the reference in our primary analysis. In comparison with ‘MTX monotherapy’, TNFi monotherapy or concomitant with MTX/other DMARDs also did not modify DM risk. This was inconsistent with prior studies in which TNFi either improved insulin sensitivity22 or decreased incident DM risk.17 ,19 However, these latter studies had relatively younger cohorts, shorter follow-up or no information about BMI or RA activity/severity measures in addition to using ‘other DMARDs’ as a reference group.17 ,19 Finally, the TNFi-associated body fat increase, regardless of disease activity change,40 may impact DM risk, but this hypothesis necessitates further investigation.

Notably, we also found significant DM risk reduction with ABA compared with MTX monotherapy, which has not been previously reported. Immunologically, ABA may slow the decline of pancreatic β-cell functions in type 1 DM,41 though its effects on type 2 DM are unknown. Our results, along with evidence from a recent study reporting improvement in insulin sensitivity with 6 months ABA treatment in 15 patients with RA,42 suggest ABA has favourable effects on insulin resistance. Nevertheless, considering the relatively lower number of patients taking ABA (N=839), investigation in a larger sample is needed.

Another noteworthy finding was the DM risk increase with statins. This association has been previously reported in meta-analyses of randomised controlled trials (RCT) and observational cohort studies of the general population, ranging from 9% to 87% increase.24–26 ,43 Although statins were associated with decreased overall mortality,44 improvement in endothelial functions and atherosclerotic plaques45 ,46 and even amelioration in disease activity in RA,28 ,29 the impact of statins on the risk of incident DM has not been evaluated before. The effects of statins in primary prevention of CVD in patients with RA have been recently investigated in an RCT of 2986 patients with RA.47 Despite the early termination of study due to low event rates, a non-significant decrease in CV events was reported in the preliminary results without any DM data.47 We found a 55% increase in DM risk with statins that was comparable with GC. Adjustment for BMI, smoking, physical activity and other comorbidities including CVD, considering the common risk factors for DM and hypercholesteremia, did not change this association. Moreover in the analysis of statin-using patients for primary prevention of CVD only, the increased risk with statins persisted. Currently, the mechanisms for the higher incidence of DM with statins are not fully understood. The suggested explanations include statin-induced insulin resistance in muscles and liver48 and genetic variations in 3-hydroxy-3-methylglutaryl-CoA reductase gene.49 Statin-induced insulin resistance may also be potentiated by chronic inflammation and concurrent GC treatment in RA. To better understand the net effects of statins in RA, including whether CV morbidity and mortality benefits outweigh the risk of DM, further research is warranted.

The well-recognised increased DM risk with GC is also confirmed in this study.14 ,50 Additionally, the effects of concomitant GC with other DMARDs were examined. Persistent DM risk reduction with GC and HCQ may indicate a therapeutic approach to decrease DM risk in patients who require long-term GC treatment. This effect was also seen for concomitant use of statins and HCQ. Given that both drugs are commonly used in RA, further investigations are needed to clarify whether GC affects lipid-lowering actions and CV mortality impact of statins.

Our study has important limitations to address. First, although we found a decreased DM risk with HCQ and increased DM risk with statins, the preferential prescription of HCQ to patients with less severe disease and statins to patients who may already be at high risk for DM with frequent presence of obesity, physical inactivity and family history of lipid/glucose metabolism disorders may lead to confounding by treatment indication. To minimise this, we adjusted for several confounding factors related to disease activity/severity and DM risk. We also applied marginal structural models to new initiators of HCQ or MTX (data not shown) and found a non-significant risk reduction with HCQ compared with MTX. With this methodology, less-biased inferences were obtained for the effect of time-varying treatment in the presence of time-varying confounders, which were simultaneously affected by earlier treatment and will affect later treatment. However, restricting the study to new users, with the goal of approximating a RCT, drastically reduced the sample size and study power (only 9% of patients taking HCQ were new users). Second, serological status, tender/swollen joint counts and acute phase reactants were unavailable; therefore, patient activity scale and HAQ measures were used to adjust for RA activity/severity, and the best performing model included HAQ. Finally, we did not have access to plasma glucose or HbA1c values, and could not determine with great accuracy as to when patients developed DM, only when they were diagnosed, which could impact treatment associations.

In conclusion, our findings suggest that DM incidence is increased in patients with RA and that HCQ and ABA are associated with reduced risk of incident DM in RA while GC and statins are associated with increased risk. HCQ confers a sustainable and treatment duration-dependent favourable effect and eliminates the increased risk associated with GC or statins. Considering the increased CV mortality in RA and the importance of DM to this outcome, our findings can inform clinicians about determining the appropriate treatment decisions in high DM-risk patients with RA. Although further research is required to better understand the effects of statins on RA, given the more frequent presence of other CV risk factors in statin-using patients, careful monitoring for DM should be considered in these patients.



  • Handling editor Tore K Kvien

  • The results have been presented before in abstract form in American College of Rheumatology Annual Meeting 2016.

  • Contributors All authors participated in the conception and design of the study. GO, SP and KM analysed and drafted the manuscript. All authors contributed to the interpretation of the results and reviewed the manuscript.

  • Funding KM was supported by the Rheumatology Research Foundation Investigator Award.

  • Competing interests None declared.

  • Ethics approval This study was conducted with the approval of the Via Christi Regional Medical Center Institutional Review Board, and patients provided informed written consent prior to study enrolment.

  • Provenance and peer review Not commissioned; externally peer reviewed.