To examine the association of tumor necrosis factor α (TNFα) inhibitor use and the risk of developing diabetes mellitus in a rheumatoid arthritis (RA) inception cohort.
To examine the association of tumor necrosis factor α (TNFα) inhibitor use and the risk of developing diabetes mellitus in a rheumatoid arthritis (RA) inception cohort.
Adults diagnosed with RA between January 1, 2001, and December 31, 2009, were identified (n = 1,881). Prevalent cases of diabetes mellitus (n = 294) were excluded. Information on sociodemographic data, medical history, body mass index (BMI), laboratory measures, and medications was collected from the electronic health record. Incident diabetes mellitus was defined using the 2010 American Diabetes Association criteria or physician-established diagnosis. Time-varying Cox proportional hazards regression models were used to adjust for age, sex, race, BMI, rheumatoid factor (RF) and anti–cyclic citrullinated peptide antibodies (anti-CCP), erythrocyte sedimentation rate (ESR), and use of nonsteroidal antiinflammatory drugs (NSAIDs), glucocorticoids, hydroxychloroquine, and methotrexate.
A total of 1,587 incident RA patients without diabetes mellitus were included. The anti-TNFα users (n = 522) had a lower median age but greater baseline BMI; maximum ESR, RF, and anti-CCP positivity; and NSAID, glucocorticoid, or methotrexate use. The median followup time for the ever and never TNFα inhibitor users was 44.9 months (interquartile range [IQR] 23.7–73.0 months) and 37.1 months (IQR 16.3–65.1 months), respectively (P < 0.001). Of the 91 patients developing diabetes mellitus, 16 were ever and 75 were never TNFα inhibitor users, yielding incidence rates of 8.6 and 17.2 per 1,000 person-years (P = 0.048), respectively. Adjusting for covariates, the hazard ratio for incident diabetes mellitus in TNFα inhibitor users was 0.49 (95% confidence interval 0.24–0.99, P = 0.049) compared to the never users.
In this inception RA cohort, anti-TNFα use was associated with a 51% reduction in risk of developing diabetes mellitus.
Rheumatoid arthritis (RA) is a chronic autoimmune disease characterized by high levels of circulating tumor necrosis factor α (TNFα). Although the primary site of inflammation in RA is the synovium, changes in body composition associated with TNFα resulting in increased adiposity and decreased lean muscle mass have also been described (1, 2). Circulating TNFα is also an important mediator of insulin resistance (IR), endothelial dysfunction, and other atherogenic changes (3–5).
While the prevalence of diabetes mellitus (DM) is not dramatically increased in RA, patients with RA have been found to have altered glucose metabolism and IR (5, 6). IR often precedes the onset of type 2 DM and has been associated with an increased risk of atherosclerotic cardiovascular disease in patients with RA (7). Medications that both control RA activity and optimize glucose metabolism are therefore desired. Several investigators have reported a decrease in IR with the use of TNFα inhibitors to treat patients with RA; therefore, these drugs potentially have an impact on the development of DM (8–15).
Given the pivotal role of TNFα in the pathogenesis of IR and type 2 DM, we hypothesized that the use of TNFα inhibitors may reduce the risk of developing DM in RA patients. We used a large observational database of patients with incident RA to test this hypothesis and examined the incidence of DM in the patients who did and did not receive anti-TNFα agents.
The Geisinger Health System is a large not-for-profit primary care and multispecialty medical practice located in Central Pennsylvania. The Geisinger Health System uses health information technology infrastructure for managing and using patient data and information, with electronic health records (Epic) in all outpatient clinics, and other digital means of delivering care. Data for this study were extracted by the Geisinger Health System Information Technology Department from the electronic health records based on the definitions and inclusion/exclusion criteria described below.
The Rheumatology Department at Geisinger Health System adopted the electronic health records system in its entirety by January 1, 2001.
All adult (age ≥18 years) individuals with a new diagnosis of RA within the Geisinger Health System between January 1, 2001, and December 31, 2009, were included. A patient with RA was defined as a patient with a primary or secondary diagnosis with an International Classification of Diseases, Ninth Revision (ICD-9) code of 714.0 at ≥2 outpatient encounters with a Geisinger Health System rheumatologist. The first date was considered the “index” date. An inception cohort of RA patients was thus created. The RA diagnosis was validated against the 1987 American College of Rheumatology criteria (16) by one of the authors (SD) who manually reviewed 100 random charts. Of these 100, 97 patients with the study definition of RA met the American College of Rheumatology criteria for a diagnosis of RA. Of the 3 patients who did not meet the criteria, 1 had a diagnosis of psoriatic arthritis, 1 had mixed connective tissue disease, and the third had juvenile idiopathic arthritis. This high concordance rate of 97% between the American College of Rheumatology criteria and the study definition of RA was deemed sufficient to proceed with study analyses (17).
Patients with prevalent DM at the time of their RA diagnosis were excluded based on the following definition of DM: either an outpatient visit with an ICD-9 code of 250, a nonfasting blood glucose level of ≥200 mg/dl, a glycosylated hemoglobin (HbA1c) level of ≥6.5%, or use of any hypoglycemic medication.
Variables extracted from the electronic health records included sociodemographic data on age at the time of the RA diagnosis, sex, and race. Body mass index (BMI) was calculated as weight in kilograms (kg) divided by height in meters squared (m2) for each patient. Medications captured were: nonsteroidal antiinflammatory drugs (NSAIDs), glucocorticoids, hydroxychloroquine, methotrexate, and TNFα inhibitors (adalimumab, etanercept, golimumab, infliximab). Exposure to TNFα inhibitors, methotrexate, hydroxychloroquine, NSAIDs, and glucocorticoids included start and stop dates to capture therapeutic segments during the observation period. Continuous medication use was defined as a discontinuation gap of <3 months.
Surrogate measures to approximate RA severity included the presence of rheumatoid factor (RF; by latex agglutination method, positive at ≥14 IU/ml [Roche Diagnostics]) and anti–cyclic citrullinated peptide antibodies (anti-CCP; by enzyme-linked immunosorbent assay, positive at ≥20 units [Inova Diagnostics]) (18). Erythrocyte sedimentation rate (ESR; by Westergren method) was used as a surrogate measure to approximate RA activity; C-reactive protein level was not included because the majority of patients (64%) had no data for this variable. Only results performed at a Geisinger Health System facility were used in this study; the results of laboratory tests done elsewhere are scanned into the electronic health records and were not available for analysis.
In order to identify potential incident cases of DM, a broad composite definition (any one of the following: ICD-9 code 250 [DM], random serum glucose level of ≥200 mg/dl, HbA1c level of ≥6.5%, or ever use of any hypoglycemic/antidiabetic medications) was used initially. Chart reviews then were carried out for those patients by 2 of the authors (AB, JLA) in a more rigorous process to establish the definitive diagnosis of incident DM: either using the 2010 American Diabetes Association (ADA) criteria or, when laboratory data to validate the 2010 ADA criteria were not available (n = 11), the definitive diagnosis of incident DM was based on a physician-coded diagnosis (rheumatologist note stating a new diagnosis of DM or DM [ICD-9 code 250] listed as a visit diagnosis by a physician) and concomitant use of hypoglycemic medications. The 2010 ADA criteria are as follows: 1) HbA1c level of ≥6.5%, 2) fasting plasma glucose level of ≥126 mg/dl (7.0 mmoles/liter), 3) 2-hour plasma glucose level of ≥200 mg/dl (11.1 mmoles/liter) during an oral glucose tolerance test, or 4) a random plasma glucose level of ≥200 mg/dl (11.1 mmoles/liter) in a patient with classic symptoms of hyperglycemia or hyperglycemic crisis (19). Of the original 221 cases of incident DM identified with the broad definition, 91 cases fulfilled the more stringent criteria.
Data are presented as means with SDs or medians with interquartile ranges (IQRs) for continuous variables, and frequencies for categorical variables. Patients were classified as ever or never TNFα inhibitor users for the observation period for the purposes of describing the cohort. Baseline characteristics of these groups were compared by either Pearson's chi-square test for categorical variables and 2-sample t-test or Wilcoxon's rank sum test for continuous variables, as appropriate. The estimated DM incidence rates per 1,000 person-years, stratified by time on or off TNFα inhibitors, were compared in a Poisson regression model. The incident rate ratio was calculated to describe the association of TNFα inhibitor use and incidence rate of DM. Additionally, time until the diagnosis of DM was modeled using the Cox proportional hazards regression model. Patients that did not develop DM during the followup periods were censored at their last visit plus 6 months or December 31, 2009, whichever came first. To take full advantage of the longitudinal laboratory and medication data, TNFα inhibitor use was parameterized as a time-dependent variable in the analysis utilizing the start and stop dates of TNFα inhibitor use. A gap of <3 months of TNFα inhibitor use was considered continuous usage. To estimate an unbiased association between TNFα inhibitor use and DM, a series of models was fit as: 1) unadjusted; 2) adjusted for age at RA diagnosis, sex, and race; 3) additionally adjusted for BMI; 4) additionally adjusted for the presence of RF, anti-CCP antibodies, and ESR; and 5) additionally adjusted for the following medications: NSAIDs, glucocorticoids, hydroxychloroquine, and methotrexate. These medications were chosen because they have a potential impact on blood glucose levels (20–22). Each sequential model controlled for variables in the prior model. Variables collected longitudinally (BMI, RF, anti-CCP, ESR, and all medications) were treated as time dependent in the models. Results of the time-varying TNFα inhibitor exposure variable are reported as hazard ratios (HRs) and 95% confidence intervals (95% CIs). Kaplan-Meier curves were used for distribution of the time until development of DM between ever and never users of TNFα inhibitor.
To avoid excluding patients with unavailable laboratory test results, indicator variables were created to indicate the presence and absence of these tests. The indicator variables were added to the models and interacted with corresponding results variables to create conditional variables in the regression models. Therefore, the effect of ESR on the development of DM was determined only among those with at least one ESR measurement. P values less than 0.05 were considered statistically significant. The analysis was performed using SAS software, version 9.2. The study was approved by the Geisinger Health System Institutional Review Board.
A total of 1,881 patients with incident RA were identified through the electronic health records. Patients with prevalent DM were excluded (n = 294). Excluded patients were more likely to be male (39% versus 28%), be older, and have a higher median BMI and ESR. They were less likely to have ever used TNFα inhibitors (26% versus 33%), NSAIDs (61% versus 72%), and methotrexate (55% versus 63%). Of the 1,587 incident RA patients without DM that were included, 522 were ever and 1,065 were never TNFα inhibitor users.
Baseline patient characteristics overall and by TNFα inhibitor use status are shown in Table 1. Overall, 72% of patients were women and 95% were white, with a median age at RA diagnosis of 57 years (IQR 46–68 years) and a median baseline BMI of 28.8 kg/m2 (IQR 25.1–33.1 kg/m2). RF and anti-CCP were positive in 77% and 58%, respectively (of the 71% and 50% of patients in whom these tests were captured in the electronic health record, respectively).
|Overall (n = 1,587)||Never (n = 1,065)||Ever (n = 522)||P†|
|Female sex, no. (%)||1,150 (72)||775 (73)||375 (72)||0.697|
|White race, no. (%)||1,507 (95)||1,014 (95)||493 (94)||0.512|
|Positive anti-CCP (50% known), no. (%)||452 (58)||283 (55)||169 (62)||0.046|
|Positive RF (71% known), no. (%)||871 (77)||568 (75)||303 (81)||0.027|
|Medication use (ever), no. (%)|
|Hydroxychloroquine||609 (38)||402 (38)||207 (40)||0.463|
|NSAIDs||1,139 (72)||745 (70)||394 (76)||0.022|
|Glucocorticoids||1,350 (85)||865 (81)||485 (93)||< 0.001|
|Methotrexate||993 (63)||592 (56)||401 (77)||< 0.001|
|Age at RA diagnosis, median (IQR) years||57 (46–68)||60 (49–71)||51 (43–60)||< 0.001|
|Median BMI (88% known), median (IQR) kg/m2||28.8 (25.0–33.1)||28.3 (24.8–32.2)||29.5 (25.5–35.2)||< 0.001|
|Maximum BMI (88% known), median (IQR) kg/m2||30.7 (26.5–36.3)||30.2 (26.1–35.2)||31.8 (27.6–38.0)||< 0.001|
|BMI closest to RA diagnosis (88% known), median (IQR) kg/m2||28.6 (24.9–33.3)||28.4 (24.8–32.8)||28.9 (25.4–33.9)||0.078|
|Maximum CRP level (36% known), median (IQR) mg/liter‡||9.0 (3.0–23)||7.8 (2.3–20)||9.7 (3.6–25)||0.047|
|Maximum ESR (78% known), median (IQR) mm/hour||31 (17–52)||30 (16–50)||33.5 (20–55)||0.011|
The patients in the TNFα inhibitor user group had a lower median age and a higher median baseline BMI compared to the never users. They were also more likely to be RF or anti-CCP positive, and they had higher maximum ESR levels than the never users. Additionally, they had a higher likelihood of being treated with NSAIDs, glucocorticoids, or methotrexate.
Incident DM was found in 91 patients (5.7%). Characteristics of patients who developed DM are shown in Table 2.
|Overall (n = 91)||TNFα inhibitor use†||P‡|
|Never (n = 75)||Ever (n = 16)|
|Female sex, no. (%)||65 (71)||53 (71)||12 (75)||0.999|
|White race, no. (%)||83 (91)||70 (93)||13 (81)||0.141|
|Positive anti-CCP (42% known), no. (%)||25 (66)||20 (65)||5 (71)||0.999|
|Positive RF (82% known), no. (%)||62 (83)||52 (83)||10 (83)||0.999|
|Medication use (ever), no. (%)|
|Hydroxychloroquine||34 (37)||28 (37)||6 (38)||0.999|
|NSAIDs||68 (75)||56 (75)||12 (75)||0.999|
|Steroids||81 (89)||67 (89)||14 (88)||0.999|
|Methotrexate||60 (66)||49 (65)||11 (69)||0.999|
|Age, median (IQR) years|
|At RA diagnosis||59 (51–71)||62 (54–72)||50 (47–59)||0.002|
|At DM diagnosis||63.0 (54–74)||66 (57–75)||54 (52–62)||0.006|
|Median BMI (90% known), median (IQR) kg/m2||33.9 (29.4–38.3)||33.5 (29.2–38.1)||35.2 (31.1–38.7)||0.512|
|Maximum ESR (82% known), median (IQR) mm/hour||45 (31–68)||45 (31–67)||52 (26–79)||0.834|
The median followup time (from RA diagnosis date to DM diagnosis date or censor date) for the ever and never TNFα inhibitor users was 44.9 months (IQR 23.7–73.0 months) and 37.1 months (IQR 16.3–65.1 months), respectively (P < 0.001).The median duration of TNFα inhibitor exposure was 16.9 months (IQR 7.6–35.3 months).
Figure 1 shows Kaplan-Meier curves for the probability of developing DM by TNFα inhibitor use status. The difference between these curves was statistically significant (log rank P < 0.001), with TNFα inhibitor users having a lower risk of developing DM.
Among the 91 cases of incident DM, 16 were ever and 75 were never users, yielding incidence rates of 8.6 and 17.2 per 1,000 person-years, respectively (Poisson regression model, relative risk 0.50 [95% CI 0.25–0.99, P = 0.048]).
Time-varying Cox proportional hazards regression models were used to adjust for the following covariates: sex, age, BMI, RF and anti-CCP, ESR, and use of NSAIDs, glucocorticoids, hydroxychloroquine, and methotrexate (Table 3). The unadjusted model yielded an HR of 0.52 (95% CI 0.26–1.03, P = 0.060) for comparing current to not current use of TNFα inhibitors. In the fully adjusted model, the HR for developing DM was 0.49 (95% CI 0.24–0.99, P = 0.049) for the TNFα blocker users compared to never users.
|Unadjusted||Adjusted for sex, age, race||Additionally adjusted for BMI (kg/m2)||Additionally adjusted for RF, anti-CCP, ESR||Additionally adjusted for NSAIDs, GC, HCQ, MTX|
|HR (95% CI)||0.52 (0.26–1.03)||0.59 (0.29–1.19)||0.57 (0.28–1.16)||0.54 (0.26–1.09)||0.49 (0.24–0.99)|
This study demonstrates that the use of TNFα inhibitors in an inception RA cohort at the Geisinger Health System was associated with a 51% reduction in risk of developing DM. These findings are novel and biologically plausible.
The literature addressing the burden of DM in patients with RA is not straightforward. While Han et al found a significantly higher prevalence (10.4% versus 7.6%) of type 2 DM in patients with RA when matched with controls in a large database of approximately 28,000 RA patients, the lack of information on confounders such as RA disease duration, severity, activity, and medication use patterns limited the interpretation of their findings (23). Others have reported comparable prevalence rates of DM in patients with RA and nonrheumatic comparator groups (24–26) or patients with osteoarthritis (27). However, a more recent study by Solomon et al used health care utilization data to compare the incidence of DM in a cohort of RA patients with nonrheumatic controls and found a higher incidence of DM in RA patients compared with nonrheumatic controls (adjusted HR 1.5, 95% CI 1.4–1.5) (28).
Patients with RA have been found to have altered glucose metabolism and IR when compared with healthy controls, as shown initially by Svenson et al (6, 29), who also studied IR in RA patients and found greater IR (estimated by the quantitative insulin sensitivity check index [QUICKI]) in RA versus osteoarthritis patients. A later study (30) that included 94 patients with RA showed that patients with high-grade inflammation (estimated by high-sensitivity C-reactive protein level) were more insulin resistant (measured by homeostatic model assessment of IR [HOMA-IR]) than patients with low-grade inflammation, while homeostatic model assessment of beta cell function was similar in the 2 groups.
TNFα is secreted in excessive quantities by adipocytes in obese patients with IR (31). It is also known to decrease the tyrosine kinase activity of the insulin receptor, thus resulting in decreased glucose uptake by skeletal muscle (32) and consequent IR. It increases lipolysis in adipose tissue with a resultant increase in the plasma free fatty acid concentration, further contributing to IR (33). Furthermore, TNFα increases the production of leptin, an inflammatory chemokine produced by the adipose tissues that reduces the insulin secretion and promotes IR (34). TNFα also down-regulates adiponectin, an adipokine that promotes insulin sensitivity (35, 36). Infusing healthy humans with TNFα results in IR in skeletal muscles, as measured by the euglycemic insulin clamp (37). Isolated case reports describe the development of hypoglycemia or unstable DM after infusion of etanercept in patients with RA and DM (9, 38).
Several studies have suggested that TNFα inhibition may favorably alter glucose metabolism in patients with RA. TNFα blockade with infliximab immediately (120 minutes from the initiation of infusion) and dramatically reduced serum insulin levels and improved IR in 27 RA patients (10). Another study by Tam et al (11) of 19 patients with active RA treated with infliximab for 14 weeks demonstrated a decrease in IR as measured by the HOMA-IR, while a smaller trial by Huvers et al has shown an increase in insulin sensitivity in 5 infliximab-treated patients observed over 6 weeks (39).
A 24-week prospective study by Seriolo et al (12) showed that infliximab and etanercept resulted in progressive reduction of IR as reflected by reduction of HOMA-IR and QUICKI indexes, respectively, in a cohort of 38 RA patients (20 treated with etanercept, 18 received infliximab). The authors also found correlations between improved IR and a decrease in disease activity, with no difference found between the effects of etanercept and infliximab on glycemia (13).
A prospective study by Kiortsis et al (15) on 45 patients with RA and ankylosing spondylitis documented no changes in the HOMA-IR and QUICKI with infliximab treatment over a 1-year observation period, but noted that in the tertile of patients with the highest IR, a significant reduction in the HOMA-IR and an increase in the QUICKI were seen. In contrast, no significant changes in IR were noted by Rosenvinge et al in a small study that included 9 patients with RA after 8 weeks of treatment with adalimumab (8).
This study had several limitations. Because this was an observational study rather than a randomized clinical trial, the patients in the anti-TNF user group were somewhat different than the never users. Specifically, the patients in the TNFα inhibitor user group were younger and had a higher median baseline BMI than the never users. They were also more likely to be RF or anti-CCP positive, and they had higher median maximum ESRs and C-reactive protein levels than the never users. Additionally, they had a higher likelihood of being treated with NSAIDs, glucocorticoids, or methotrexate, and were less likely to have been treated with hydroxychloroquine. Although this is a large observational study with well-characterized patients and physician-entered diagnoses, laboratory data, and medications, it is subject to confounding by indication. This type of error can be minimized but not entirely eliminated by the use of multivariate regression models. It is reassuring, however, that the patients receiving TNFα inhibitors had more risk factors for the development of DM than the never users. Specifically, they had higher baseline BMI, higher RF and anti-CCP antibody positivity, higher maximum ESR levels, and more frequent use of glucocorticoids; nevertheless, they had a lower incidence of DM than the never users.
We used a broad definition of prevalent DM at baseline to exclude patients with preexisting DM from the observational study cohort. We therefore are confident that we did not include any patients with prevalent DM.
Our population was homogenous, thus limiting the applicability of our findings to other patient populations. These results are consistent with the only other study that looked at the association of TNFα inhibitor use and the development of DM in RA patients, although this study also included patients with psoriasis (40). In a retrospective cohort of 121,280 patients with RA or psoriasis, Solomon et al reported that TNFα inhibitor use was associated with a reduced risk of DM (HR 0.62, 95% CI 0.42–0.910) after adjustment for diagnosis, age, sex, steroid and disease-modifying antirheumatic drug use, and markers of general health.
In conclusion, this study demonstrated a 51% reduction in the risk of developing DM in RA patients treated with anti-TNF agents. While further studies are needed to confirm these findings, our results suggest that drugs inhibiting the biologic effect of TNFα have the potential not only to improve RA activity but also to reduce the burden of DM in these chronically ill patients at high risk for cardiovascular disease. This protective association between the use of anti-TNF agents and DM risk should be considered when weighing the potential risks and benefits of these biologic agents in the treatment of patients with RA.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Bili had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Antohe, Bili, Kirchner, Dancea, Wasko.
Acquisition of data. Antohe, Bili, Morris, Dancea.
Analysis and interpretation of data. Antohe, Bili, Sartorius, Kirchner, Dancea, Wasko.
Amgen/Wyeth provided financial support for this study through an investigator-initiated grant. The companies had no involvement in the study design, data collection, data analysis, or writing of the manuscript. Neither the content nor the publication of this study was contingent upon the companies' approval.