Article Text

Extended report
Risk of diabetes among patients with rheumatoid arthritis, psoriatic arthritis and psoriasis
  1. Daniel H Solomon1,2,
  2. Thorvardur Jon Love1,
  3. Claire Canning2,
  4. Sebastian Schneeweiss2
  1. 1Division of Rheumatology, Department of Medicine, Brigham and Women's Hospital, Boston, Massachusetts, USA
  2. 2Division of Pharmacoepidemiology, Department of Medicine, Brigham and Women's Hospital, Boston, Massachusetts, USA
  1. Correspondence to Dr Daniel H Solomon, Division of Rheumatology, Department of Medicine, 75 Francis Street, PBB-B3, Boston, MA 02115, USA; dsolomon{at}


Objective To examine the risk of diabetes mellitus (DM) among subjects with rheumatoid arthritis (RA), psoriatic arthritis or psoriasis (PsA/PsO), compared with non-rheumatic controls.

Methods Study cohorts were assembled using linked healthcare utilisation data from British Columbia. All people with at least two diagnoses of RA or PsA/PsO were included and compared with a cohort of people without any known rheumatic disease. The outcome of interest was a diagnosis of new-onset DM, as defined by initiation of an antidiabetic drug. Incidence rates (IRs) per 1000 person-years and IR ratios were calculated and Cox regression models were constructed to determine the hazard ratio (HR) for diabetes by age, gender, systemic immunosuppressive drug and glucocorticoid use.

Results The study cohort comprised 48 718 subjects with RA, 40 346 with PsA/PsO and 442 033 without any rheumatic disease. The IR for DM among subjects with RA was 8.6 per 1000 person-years (95% CI 8.5 to 8.7), PsA/PsO 8.2 (95% CI 8.1 to 8.3) and for non-rheumatic controls 5.8 (95% CI 5.8 to 5.8). The adjusted HR for RA compared with non-rheumatic controls was 1.5 (95% CI 1.4 to 1.5) and 1.4 (95% CI 1.3 to 1.5) for PsA/PsO.

Conclusions RA and PsA/PsO appear to be associated with an increased risk of DM. The ability of potent antirheumatic treatments to reverse this trend warrants study.

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.


Cardiovascular disease (CVD) represents an important source of morbidity and mortality in several rheumatic diseases, including rheumatoid arthritis (RA) and psoriatic arthritis/psoriasis (PsA/PsO).1 2 Previous studies have debated the relative importance of traditional CVD risk factors versus rheumatic disease-specific factors.3 4 However, the dichotomy between different types of risk factors may be false. Inflammation appears intimately related to insulin resistance, dyslipidaemia and possibly, hypertension.5,,7

A number of studies show that insulin resistance is increased in RA. While there seems to be broad agreement about the relation between RA and insulin resistance, only two previous studies have focused on the risk of diabetes mellitus (DM) in RA. One very large study based on health insurance claims calculated an OR of 1.4 for DM among a cohort with RA compared with healthy controls.8 Another medical records study from a longitudinal cohort found no increased risk of DM.9 In contrast with RA, previous studies of PsA and PsO agree that there is an increased risk of DM, with the adjusted relative risks ranging from 1.2 to 1.6.10 11 In previous work, little attention has been paid to systemic immunosuppressive agents or topical glucocorticoid use.

We examined the risk of DM in population-based cohorts of patients with RA or PsA/PsO, paying close attention to age-specific and gender-specific risk, as well as to relevant drug use.


Data source

We studied three cohorts derived from the population-based insurance programme of British Columbia (BC), Canada. The pharmacy programme, PharmaNet, includes the name, dose and dispensed quantity for all prescription drugs dispensed in BC pharmacies. Up to 25 diagnoses for hospital discharges and one diagnosis for each medical service are recorded, with good specificity and completeness.12 Because all BC residents are covered for all medical services by the provincial Medical Services Plan, except for a small number of federal employees, and dispensing of drugs is recorded for all dispensed prescription drugs regardless of payer, the study sample is representative of BC's adult population (about three million in 2005).13

The appropriate institutional review board approved this protocol. Data use agreements are in place between the investigators and BC.

Cohort selection

The health insurance programmes of BC include all citizens and all their medical and pharmacy claims. The three cohorts encompassed RA, PsA/PsO and non-rheumatic disease controls. All subjects in the RA cohort had at least two visits for RA (ICD-9-CM 714.0) at least 1 week apart. Similarly, subjects with PsA/PsO were defined by at least two visits for PsA or PsO (ICD-9-CM 696.0 or 696.1) at least 1 week apart. The non-rheumatic disease controls had to have had no visit coded for RA, PsA/PsO or any other inflammatory rheumatic disease (ICD-9-CM 287.x, 446.x–447.x, 695.4, 710.x–713.x). We attempted to match five controls with each case based on calendar year of study entry. The controls entered the cohort on a doctor visit date.

Our study database spanned the period 1 January 1996 through 31 December 2006. All subjects could enter the cohort after qualifying for inclusion, that is, second diagnosis of RA or PsA/PsO or matched doctor visit for controls. We excluded subjects with a diagnosis of DM (ICD-9-CM 250.x) before their cohort entry date. Subjects were followed up until they experienced an outcome, died, left BC or follow-up ended (31 December 2006).

DM outcome

The outcome of interest was the diagnosis of DM or the use of drugs specific for DM. We did not have actual laboratory data; thus, the primary definition of DM required at least one prescription for a DM-specific drug. These included all insulin preparations, as well as oral agents. The secondary definition of DM required both a receipt of a DM-specific drug plus a diagnosis of DM (ICD 250.x). Similar definitions have been used in previous studies from Canada and found to have specificities of >90%.14 15 Sensitivity analyses considered the type of treatment started for DM, insulin or non-insulin.

Potential predictors of DM

We examined several potential predictors of DM using data from the 12 months before cohort entry, including age, gender, comorbid medical conditions, healthcare use, glucocorticoid use and systemic immunosuppressive drug use. Age was defined at cohort entry date. The count of comorbid medical conditions encompassed data from the 12 months before cohort entry and used the Romano adaptation of the Charlson index.16

Oral and topical glucocorticoids, as well as systemic immunosuppressive agents, were considered separately based on data from the 12 months before cohort entry (see online supplementary file for a list of all preparations).

Statistical analyses

We compared the baseline characteristics across the three cohorts. Incident DM was identified during follow-up and person-years calculated. This allows for estimation of an incidence rate (IR) for DM. The IRs were estimated for each cohort separately and then stratified by age, gender, systemic immunosuppressive drug use and glucocorticoid use. The IRs for RA and PsA/PsO were compared with those for non-rheumatic controls to calculate an IR ratio. These were also stratified according to age, gender and drug use. Finally, a Cox proportional hazard regression model was constructed to assess the adjusted HR of DM associated with RA, and PsA/PsO. These models were stratified according to age, gender and drug use.


From the total potential population of BC during the study period, 4 310 500 were potentially eligible. We identified 84 480 subjects with at least two diagnoses of RA (1.96%) and 73 909 subjects with PsA/PsO (1.71%). Further exclusions because of previous DM left 48 718 with RA, 40 346 with PsA/PsO and 442 033 non-rheumatic controls.

The baseline characteristics of the study cohorts are shown in table 1. There were substantial baseline differences in most of the characteristics. Table 2 shows the IRs for each cohort. The IR for diabetes among subjects with RA was 8.6 per 1000 (95% CI 8.5 to 8.7), for PsA/PsO 8.2 (95% CI 8.1 to 8.3) and for non-rheumatic controls 5.8 (95% CI 5.8 to 5.8). The IRs by age stratum demonstrate an increase with older age; as well, IRs were higher for men than women. The IRs were higher among people using systemic immunosuppressive agents than those not, likely reflecting a greater underlying disease burden. As expected, people using oral or topical glucocorticoids experienced higher IRs for DM. The IRs were very similar for the secondary definition of DM (see online supplementary file).

Table 1

Baseline characteristics of primary cohorts

Table 2

Diabetes incidence rates (per 1000 person years), age and gender stratified

The adjusted HRs are shown in figure 1. The HR for RA compared with non-rheumatic controls was 1.5 (95% CI 1.4 to 1.5) and 1.4 (95% CI 1.3 to 1.5) for PsA/PsO. In the adjusted Cox regression models, oral glucocorticoid use (HR=1.3, 95% CI 1.2 to 1.4) and topical (HR=1.3, 95% CI 1.1 to 1.4) were both associated with an elevated risk of DM. While the IRs increase with older age, the HRs are lower with older age in both genders (see figure 1B,C).

Figure 1

Cox proportional hazard ratio for diabetes mellitus across the age stratum for (A) total cohort; (B) women; (C) men. All models were adjusted for the variables in table 1.

Finally, we constructed Cox regression models among people without any use of oral glucocorticoids or topical glucocorticoids. The adjusted HRs did not change in these restricted models: RA (no oral glucocorticoid use HR=1.4, 95% CI 1.3 to 1.5; no topical glucocorticoid use HR=1.5, 95% CI 1.4 to 1.5) and PsA/PsO (no oral glucocorticoid use HR=1.4, 95% CI 1.3 to 1.5; no topical glucocorticoid use HR=1.4, 95% CI 1.3 to 1.4).


The relationship between DM and rheumatic diseases is of interest because of the well-documented increased risk of CVD in RA and PsA/PsO.1,,4 Inflammation has an important role in driving insulin resistance and metabolic syndrome.5 While substantial publications supports the relationship between insulin resistance and rheumatic diseases, there are surprisingly few data about rheumatic diseases and DM. We studied the IR of DM among subjects with either RA or PsA/PsO. Our results confirm an elevated relative risk for incident DM among subjects with PsA/PsO compared with non-rheumatic controls.8 10 11 The findings among subjects with RA were remarkably similar—elevated relative risk in both genders but decreasing risk with age. The elevated adjusted HRs seen among subjects not using oral or topical glucocorticoids suggests that this risk is not primarily an adverse effect of such treatments.

While substantial work on insulin resistance has been published, relatively few studies on the risk of DM in RA have been conducted. One study using a large insurance database found an increased risk (prevalence ratio 1.4).8 The second study to examine DM among subjects with RA used a much stronger study design. A longitudinal cohort of subjects with RA was assembled retrospectively from the Rochester Epidemiology Project. Diabetes was not the focus of this study, but in unadjusted results the authors found no increase in the risk of new-onset DM (RR=0.978). It is important to note that the IR calculated for DM in the study from the Rochester Epidemiology Project—7.9 per 1000 person-years—is very similar to the IR calculated in this study (8.6 per 1000 person-years) and both are in line with one previous estimate from a large RA cohort (7.6 per 1000 person-years, calculated from the date presented).17 Thus, the RR differs between our study and the study from the Rochester Epidemiology Project because of different IR estimates for the non-RA population—10.2 per 1000 person-years in Rochester Epidemiology Project versus 5.8 in our study.

Limitations of our study include a database that does not include actual blood glucose values, American College of Rheumatology classification criteria and lack of information on body mass index and family history of DM. Similar algorithms for DM have been previously tested and found to have positive predictive values of >90%.14 15 Our findings did not differ much when we ran analyses using slight variations on the coding algorithms for DM (data not shown). Our prevalence calculations for RA are consistent with the literature suggesting minimal misclassifcation.18 Additionally, body mass index among subjects with RA is similar to that of non-rheumatic controls.1

Our analyses need to be replicated, especially in RA cohorts. The current epidemiological analyses add important context to this ongoing area of study. People with RA or PsA/PsO and their healthcare providers should be aware of the potential link with DM. As these associations become better defined, regular DM screening may be called for among these rheumatic disease populations.


Supplementary materials

  • Web Only Data ard.2009.125476

    Files in this Data Supplement:


  • Funding Amgen supported this work through a research grant to Brigham and Women's Hospital. DHS is also supported by grants from the NIH (NIAMS K24 AR055989, NIAMS P60 AR 047782, NIDCR R21 DE 018750, NIAMS R01 AR 056215), the Arthritis Foundation and AHRQ.

  • Competing interest None.

  • Ethics approval This study was conducted with the approval of the Partners Healthcare Human Ethics Committee.

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