Identifying Adverse Drug Withdrawal Events at Scale Using Electronic Clinical Data: Development, Adjudication, and Practical Challenges

BACKGROUND: There are no scalable, automated methods to identify adverse drug withdrawal events (ADWE) in deprescribing studies. We describe the development of a scalable approach to identifying potential ADWE using electronic clinical data, characterize the ADWE identified, and discuss complexities encountered. METHODS: We identified potential ADWE through two steps: (1) identifying intentional medication discontinuations; (2) applying diagnostic and procedure codes and clinical parameters suggestive of possible ADWE in specified risk timeframes. Discontinuations of five medication classes (oral diabetes medications, statins, bladder antimuscarinics, antihypertensives, and anticoagulants/antiplatelet agents) for adults age ≥ 65 enrolled with Kaiser Permanente Colorado (KPCO) were identified through text search and medication order data and confirmed through chart review. Electronically identified potential ADWE were manually adjudicated; notes made during adjudication were analyzed using qualitative content analysis to classify possible ADWE and characterize associated features. RESULTS: Within specific drug groups, between 51% (39/76 bladder antimuscarinics) and 81% (104/128 antihypertensives) were confirmed as intended discontinuations. Manual adjudication across all drug groups classified 60 of 293 events as possible ADWE, 23 probable, and 2 definite; 1 major ADWE resulted in hospitalization. Adjudication illuminated 5 challenges in classifying ADWE: (1) Causal Inference: unclear whether potential ADWE had a clear causal link to medication discontinuation given medical complexity; (2) Timeframe: event occurred outside expected risk window; (3) Contraindication: medication was intentionally held due to acute contraindication; resulting clinical changes were expected; (4) Terminology: medication had never been effective; symptoms were present both before and after discontinuation, and did not recur as a result of stopping drug; and (5) Intent: observed clinical change was intended or expected outcome of stopping medication. CONCLUSIONS: Many potential ADWE may be expected or intended sequelae of appropriate clinical care, or unavoidable. Future research should define and validate ADWE. Broader terminology is required to avoid mislabeling expected consequences as ADWE.