Background: Registries and biobanking cohort studies often collect self-reported medication data from participants. The inherent potential for recall bias means validation of self-reports are important for maximising the research utility and accuracy of these data for use in pharmacoepidemiology, health economics and precision medicine outputs.
Methods: We analysed concordance between Australian reference-standard prescription medication dispensing data (Pharmaceutical Benefits Scheme; PBS; 2011-2018), and questionnaire self-reports of current disease-modifying antirheumatic drug (DMARD), corticosteroid, anti-inflammatory and analgesic use from participants in the Australian Rheumatology Association Database (ARAD) cohort (2012-2018). Estimates of sensitivity, positive predictive value (PPV) and inter-rater reliability (Kappa statistic; κ), with 95% confidence intervals (CI) were produced for each medication, matched at the 7-digit Anatomical Therapeutic Chemical (ATC) or PBS item code level, over four fixed time-windows of dispensing data prior to questionnaire completion (30 days, 90 days (Table 1), 6 months and 12 months).
Results: In our study population (n=3282; 67% female; 63.9% rheumatoid arthritis) self-reports of biologic and targeted-synthetic DMARDs with shorter dosing intervals (i.e. daily to 8-weekly) showed very good/excellent agreement in the primary 90-day analysis (ranges: κ 0.85-0.94/PPV 0.83-0.97/sensitivity 0.84-0.96), followed by conventional synthetic DMARDs (κ 0.41-0.86/PPV 0.45-0.86/sensitivity 0.87–1.00), prescription-only non-opioid analgesics (κ 0.65-0.78/PPV 0.6-0.72/sensitivity 0.72-0.86), oral prednisone/prednisolone (κ 0.72/PPV 0.78/sensitivity 0.8) and prescription-only opioid analgesics (κ 0.49-0.7/PPV 0.43-0.71/sensitivity 0.51-0.72).
Conclusions: Self-reports of prescription-only medications that are dispensed regularly at shorter intervals showed very good/excellent agreement with 90 days of retrospective pharmaceutical claims data. Medications with longer dosing intervals, intermittent use, larger dispensing amounts and/or fractional dosing showed moderate to very good agreement, which improved with wider retrospective time windows of claims data (6 – 12 months). Expectedly, mixed prescription/over the counter medicines were difficult to validate with this method. These results suggest ARAD participants accurately self-report their current rheumatology-related medications via longitudinal patient-reported surveys.