OBJECTIVES: Rheumatoid arthritis (RA) confers both significant short- and long-term disability, and substantial background health care costs, yet cost data are scarce. We modeled disability in RA using covariates also available in routine, primary care data from which total health care cost were estimated. METHODS: In modeling disability, biologic-naïve RA cases from the UK biologics register were selected.  General Linear Modeling (GLM) was used to predict HAQ score from demographics, co-morbidity, polypharmacy, RA duration, BMI, smoking status and disease activity. Health care costs were estimated from similar incident cases with >=1 year’s observation selected from the THIN database.  Annual costs included prescriptions, consultations, and investigations from primary care and hospital episodes. An alternatve generalised linear model assumed a Poisson distribution, and a log-link explored the association between health care costs and predicted HAQ.  RESULTS: Disability was modeled using 3,924 cases, of whom 75% were female with an average age of 58 years (sd 14) and an average baseline HAQ of 1.80 (sd 0.72). The optimal HAQ model included gender, RA duration, number prior DMARDs, number of current non-DMARD drugs, logCRP, smoking status and stroke history.  The r2 value was 0.27, while mean prediction error was 0.15 (sd 0.30). Costs were assessed in 3,378 cases with similar gender balance but slightly older (mean 64 years). Their predicted HAQ scores ranged from 0.48 to 3.00, with a mean of 1.60 (sd 0.40). The mean total cost of care in this population was £1,960 per annum (sd, £1,864), with prescriptions the largest component (43%), followed by outpatient attendances (18%), GP consultations (16%), and hospital admissions (13%). Predicted total costs (GBP2007) ranging from £650 to £10,017 were calculated from EXP^(5.949+(Age*0.002)+(pHAQ*0.675)+(Age*pHAQ*0.003)). CONCLUSIONS: Total health care costs for RA patients increased exponentially across the range of estimated disability. Exclusion of support-service costs suggests these values may be conservative.