Medal: a patient similarity metric using medication prescribing patterns [article]

Arturo Lopez Pineda, Armin Pourshafeie, Alexander Ioannidis, Collin McCloskey Leibold, Avis Chan, Jennifer Frankovich, Carlos D. Bustamante, Genevieve L. Wojcik
2019 medRxiv   pre-print
Patient similarity offers an exciting potential to improve our understanding of treatment patterns. Medication history is a valuable source of information given the clinical considerations taken during the physician's prescription process. However, available similarity methods do not consider timeliness of event occurrence in the longitudinal care of patients. In this observational cohort study, we propose an event alignment algorithm, Medal, which uses a dynamic programming approach for
more » ... e alignment of medication histories. To test our algorithm, we obtained the medication usage from a cohort of patients with pediatric acute-onset neuropsychiatric syndrome (PANS). After calculating a distance metric with Medal, we compute a hierarchical clustering and explore the most appropriate number of clusters. We identified four clusters in PANS with distinct medication usage histories, driven primarily by penicillin. We foresee that our algorithm could be used to identify clusters in other syndromes treated with multiple medication regimens.
doi:10.1101/19004440 fatcat:3o7go6bmivh6dmgxhy7smolngu