Background Whether depression makes RA worse is unclear because active RA can itself cause depression. The issue is even less clear when cross-sectional studies are performed, for there is no way to determine the direction of causality, and patient self-reports are also influenced by depression. Longitudinal evaluations employing a ?hard? outcome such as mortality can help overcome these problems.

Objectives To understand the correlates of depression in RA and to determine whether depression itself is predictive of mortality in RA.

Methods We studied 1500 consecutive RA outpatients during a 20 year period beginning in 1981. 1483 patients completed 27,773 depression questionnaires from the Arthritis Measurement Impact Scales (AIMS). The relation of depression to other variables was studied with generalised estimating equations (GEE). For ease of interpretation, depression was considered present when the AIMS score was 4 or greater in the GEE analyses. The predictability for mortality of depression, and other variables obtained 1) during the first 2 years of follow-up and 2) over the entire course of RA was examined using Cox regression (Cox) models.

Results Depression was related to age < 65 years (odds ratio 1.3), non-white ethnic origin (OR 1.9), and decreased family income, but was unrelated to sex or marital status. Compared to university graduates, secondary school graduates and non-graduates had increased risk for depression, OR 1.4 and 2.1 respectively. In multivariate models, education, income and ethnic origin were independently related to depression. In clinical models, ESR, white cell count, joint count, grip strength, patient global severity, pain and HAQ (HAQ OR 2.33 (95% CI 2.11, 2.59, p < 0.001) were also strongly related to depression. In univariate Cox models, 2-year depression was predictive of mortality (z = 3.55 p < 0.001), and remained predictive even when ESR, joint count, WBC and grip strength were added as 2-year covariates (z = 2.00 p = 0.046). However, when HAQ disability, patient global, or pain was added to the model, depression was no longer significant (p = 0.388). The same results were noted when the 2-year depression scores were modelled with full covariates (from first through last observation). Finally the hazard ratio (HR) for the effect of a depression score of 4 or greater over the full course of the disease was 1.33 (95% CI 1.09, 1.63) after adjusting for all covariates. This result was maintained after also controlling for blood pressure, BMI, prednisone use and comorbid illnesses.

Conclusion Depression during the first 2 years of clinical care is associated with increased RA mortality, and is also associated when laboratory or physician associated covariates are added. The effect of depression on mortality is strengthened by follow-up, and this association (HR 1.33) is maintained even after the inclusion of all covariates. These data suggest that depression has a small but definite association with mortality in RA. Even so, it is possible that additional medical factors that were not recognised by our models might be causally related to depression and therefore to mortality.