Abstract
OBJECTIVE: To assess agreement between prescription claims data and self-reported medication use via longitudinal questionnaires in the Australian Rheumatology Association Database inflammatory arthritis cohort and to identify predictors of discordant self-reports. METHODS: Agreement was determined between longitudinal questionnaire self-reports (2012-2023) of disease-modifying antirheumatic drug (DMARD), glucocorticoid, anti-inflammatory, and analgesic use and Australian reference standard prescription medication dispensing data (Pharmaceutical Benefits Scheme) using Cohen's kappa, sensitivity, positive predictive value (PPV), and negative predictive value (NPV). Analyses were repeated using four look-back windows of dispensing data (1, 3, 6, and 12 months) before each questionnaire to characterize variations in agreement metrics at the individual medication level. Predictors of discordant self-reports were explored using multivariable logistic regression. RESULTS: In our study population (N = 3,407, 67% female, 93.7% White, 64.4% with rheumatoid arthritis, 18.5% with psoriatic arthritis, 14.6% with ankylosing spondylitis, 2.6% with juvenile idiopathic arthritis), agreement with prescription claims data was substantial to high for DMARDs (κ 0.67-0.95, sensitivity 0.69-0.96, PPV 0.64-0.96, NPV 0.9-1), substantial for prescription-only nonopioid analgesics and oral prednisolone/prednisone (κ 0.66-0.80, sensitivity 0.65-0.88, PPV 0.68-0.77, NPV 0.93-1), and moderate to substantial for prescription-only opioid analgesics (κ 0.48-0.7, sensitivity 0.57-0.74, PPV 0.36-0.69, NPV 0.94-1). A 3-month look-back window optimized agreement for most medications, whereas 6- and 12-month windows further improved agreement for specific drugs. No consistent predictors of discordant self-reports were identified, though greater self-rated disability severity and poorer overall health showed the most consistent associations with discordance. CONCLUSION: Agreement between self-reported and pharmaceutical claims data was moderate to high. Poorer overall health and disability may impact accuracy of medication self-report.