A Structured, Auditable Taxonomy for Pharmacist-Identifiable Medication-Related Risk Signals in Routine Admission Data: A Retrospective Medication Safety Audit in a High-Risk Inpatient Cohort.

Background Hospital pharmacists collect rich admission data, i.e., medication history, brought-in medicines, prescription and laboratory records, and vital signs, yet the medication-related problems these data contain are surfaced by ad hoc, non-auditable judgment. Existing drug-related problem classifications code a problem after it has been recognized; they do not specify which admission data element to examine or which decision rule to apply. The prespecified aims of this study were to develop a taxonomy of pharmacist-identifiable medication-related risk signals with explicit operational definitions and data sources, and to apply it, as a feasibility application in a very small cohort, to routine admission data from patients who developed in-hospital ischemic stroke. A four-stage pharmacist-led assessment framework was derived post hoc as a secondary output. Methodology In this single-center retrospective medication safety audit (Strengthening the Reporting of Observational Studies in Epidemiology), 12 signal categories, each with an operational definition and data source, were fixed before record review and mapped to existing classifications. The taxonomy was applied to 17 consecutive patients with in-hospital ischemic stroke (August 2020-September 2023). Two pharmacists reviewed the records jointly and classified each case by consensus, counting only signals documented before the index date. Analysis was descriptive; category-level counts with explicit denominators were the primary quantities. Results All 17 cases could be classified against all 12 categories from admission data. Category-level counts were: poor adherence or medication management, 10 of 17; residual medication discrepancy, 5 of 17; treatment self-discontinuation, 3 of 17; insulin instruction mismatch, 2 of 17; suboptimal blood pressure control, 9 of 13 treated hypertensive patients; and suboptimal glycemic control, 5 of 8 patients with diabetes. As a secondary summary, 14 of 17 patients had at least one signal. Inter-rater reliability, content validity, and pharmacist workload were not measured. Conclusions In this 17-patient feasibility application, a predefined, operationally defined taxonomy could be applied to data pharmacists already collect. It yields a potentially feasible structured and auditable framework using routinely available admission data, not a validated or demonstrably low-burden tool. Because no comparator was studied, the counts are neither specific to nor predictive of stroke; independent prospective evaluation of reliability, validity, and workload is required.