Statistical analysis
Classic descriptive statistical methods were used to evaluate the distribution of continuous and categorical variables. Multivariable linear regression models were performed to identify the day of PM10, PM2.5, NO2 and O3 independently associated with DAS28, SDAI, GH, PhGA, PaGA, TJC and SJC, adjusting for radiological damage, smoking habits (three categories), seropositivity for rheumatoid factor and/or ACPA (yes/no), therapy with Disease Modifying anti-Rheumatic Drugs (DMARDs) (no DMARDs, conventional synthetic - csDMARDs, targeted synthetic - tsDMARDs, biological - bDMARDs), use of steroids, age at examination, and disease duration. The potential effect modification of the therapy was investigated adding an interaction term between pollutants and therapy in each model. When the interaction term resulted significant (p <0.05), the association between pollutant and outcome was investigated in each subgroup of therapy. β coefficients were reported for 10 µg/m3 increments of PM10, PM2.5, O3 and NO2 concentrations. As the distribution of the outcome variables (DAS28, SDAI, GH, PhGA, PaGA, TJC and SJC) were skewed, they were log-transformed. All statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, NC, USA). The data underlying this article will be shared on reasonable request to the corresponding author.