Background Use of anti-TNF biologic agents in rheumatoid arthritis has been linked to an increased risk of infections. Both administrative and clinical datasets have limitations with case identification which may affect conclusions.
Objectives To evaluate three different methods of case definition of serious infections in Rheumatoid Arthritis (RA) patients treated with anti-Tumor Necrosis Factor (anti-TNF) therapy and how they have an impact on incidence and predictors of developing a serious infection.
Methods RA patients treated with anti-TNF therapy between January 2004 and March 2009 were followed prospectively in the Alberta Biologics Pharmacosurveillance Program to assess treatment efficacy and adverse events. Clinical and self-reported data were linked with provincial health care administrative databases. Infections were identified in three ways: method A) by using relevant ICD 10 diagnosis codes from the hospitalization database; method B) by using relevant ICD 10 diagnostic codes from the hospitalization or acute care databases; and method C) by using self-reported hospitalization for infection data (nurse verified). We used multivariate Cox-regression to assess independent predictors of the risk of infection. Candidate variables included steroid use, DAS28 remission, biologic switch, baseline variables (HAQ, DAS, Disease duration, Age, Sex, Marital status, Education and Smoking) and Self-Reported Co-morbidity Questionnaire Score.
Results The whole cohort consists of 1,086 patients (70% female, mean age of 54 years) with a mean follow-up time of 2.3 years. Nine hundred and forty three patients with at least 3 months of administrative data were either first biologic users (n=731) or biologic switchers (n=212). Using the three definitions of serious infections as our outcome variable: method A identified 39/943 patients (4%), method B identified 155/943 patients (16%), and method C identified 169/943 patients (18%) who had a serious infection. The most common serious infections in decreasing order of frequency was pneumonia, cellulitis, septicemia, and septic arthritis. Using model A, any use of steroids predicted a RR of 3.29 for serious infections and post-secondary education level was protective (RR 0.33). Using model B the predictors were any steroid use (RR 1.66) and disease duration (RR 1.02). Using model C the predictors were any steroid use (RR 1.85), biologic switch (RR 1.58), post-secondary education (RR 1.69), age RR (1.01), and smoking (RR 2.5).
Conclusions In our analysis any use of steroid was a consistent predictor of serious infection even taking into account disease severity as measured by baseline HAQ. Achieving a DAS28 remission was not associated with a lower risk of infection. Using method C, switching biologics and current smoking also predicted infections. Post secondary education was protective in the method A analysis but predictive in the method C analysis which may relate to likelihood of self-reporting. Balancing comprehensive searching of administrative databases with validated self-reported methods, that is, analyzing the differences in sensitivity and specificity of the methods will vary results but also help identify consistent predictors of infection.
Disclosure of Interest None Declared