Examining Baseline and 1 Year Predictors of Radiographic Progression over 5 Years Using Mixed Effects Negative Binomial Models: Results from the Early Rheumatoid Arthritis Study (ERAS) Cohort

Background Radiographic damage is regarded as one of the most important outcomes in Rheumatoid Arthritis (RA). The Larsen score is one common method of scoring radiographic damage. While presence of erosions by 3 years in patients with RA is common, there are large numbers of patients with zero scores in early disease, which results in skewed distributions of Larsen data. To date, no model has been developed to examine predictors for radiographic progression over 5 years, which also accounts for the non-normal distribution of Larsen data, and also models all available data over a 5-year period. Objectives To identify potential predictors of radiographic progression over 5 years in early RA, using standard clinical and laboratory variables recorded at baseline and 1 year. Methods The analysis involved data from the Early Rheumatoid Arthritis Study (ERAS, n=1465), a UK inception cohort of patients with early RA. The analysis was restricted to a sub-group of patients with digitised radiographs scored with Larsen for first 5 years of follow-up (Larsen data was available for 76%, 70%, 71%, 68%, 36% and 55% of patients at baseline, 1, 2, 3, 4, and 5 year respectively). A mixed effects negative binomial regression model examined predictors of radiographic progression over the first 5 years of disease. Two models were run. Both models included sex, age at presentation and rheumatoid factor. One model included BMI, ESR, HAQ, DAS, haemoglobin (HB) and time to first RA visit from symptom onset at baseline, the other included the same variables at 1-year and time to first DMARD from symptom onset. Results Of the 1465 patients recruited into ERAS, 1031 patients (contributing 4908 observations, with an average of 4.8 observations per patient) were included in the baseline analysis, and 1005 (contributing 4818 observations, with an average of 4.8 observations per patient) were included in the 1-year analysis. There was a significant progression of Larsen scores over 5 years, indicating an average increase of 46% in Larsen for every 1 year increase (IRR 1.46, P