Methodological challenges when evaluating potential off-label prescribing of drugs using electronic health care databases
Cainzos-Achirica, M., Varas-Lorenzo, C., Pottegård, A., Asmar, J., Plana, E., Rasmussen, L., ... Pladevall-Vila, M. (2018). Methodological challenges when evaluating potential off-label prescribing of drugs using electronic health care databases: A case study of dabigatran etexilate in Europe. Pharmacoepidemiology and Drug Safety. DOI: 10.1002/pds.4416
PURPOSE: To report and discuss estimated prevalence of potential off-label use and associated methodological challenges using a case study of dabigatran.
METHODS: Observational, cross-sectional study using 3 databases with different types of clinical information available: Cegedim Strategic Data Longitudinal Patient Database (CSD-LPD), France (cardiologist panel, n = 1706; general practitioner panel, n = 2813; primary care data); National Health Databases, Denmark (n = 28 619; hospital episodes and dispensed ambulatory medications); and Clinical Practice Research Datalink (CPRD), UK (linkable to Hospital Episode Statistics [HES], n = 2150; not linkable, n = 1285; primary care data plus hospital data for HES-linkable patients).
STUDY PERIOD: August 2011 to August 2015. Two definitions were used to estimate potential off-label use: a broad definition of on-label prescribing using codes for disease indication (eg, atrial fibrillation [AF]), and a restrictive definition excluding patients with conditions for which dabigatran is not indicated (eg, valvular AF).
RESULTS: Prevalence estimates under the broad definition ranged from 5.7% (CPRD-HES) to 34.0% (CSD-LPD) and, under the restrictive definition, from 17.4% (CPRD-HES) to 44.1% (CSD-LPD). For the majority of potential off-label users, no diagnosis potentially related to anticoagulant use was identified. Key methodological challenges were the limited availability of detailed clinical information, likely leading to overestimation of off-label use, and differences in the information available, which may explain the disparate prevalence estimates across data sources.
CONCLUSIONS: Estimates of potential off-label use should be interpreted cautiously due to limitations in available information. In this context, CPRD HES-linkable estimates are likely to be the most accurate.