A method for controlling complex confounding effects in the detection of adverse drug reactions using electronic health records.

摘要: Objective Electronic health records (EHRs) contain information to detect adverse drug reactions (ADRs), as they comprehensive clinical information. A major challenge of using involves confounding. We propose a novel data-driven method identify ADR signals accurately by adjusting for confounders. Materials and methods focused on two serious ADRs, rhabdomyolysis pancreatitis, used in 264 155 unique patient records. identified an established criteria, selected potential confounders, then penalized logistic regressions estimate confounder-adjusted associations. reference standard was created evaluate compare the precision proposed four others. Results Precision 83.3% 60.8% pancreatitis when method, we several safety that are interesting further review. Discussion The effectively estimated associations after confounders. main cause error probably due nature dataset substantial number patients had single visit only and, therefore, it not possible determine correctly appropriate sequence events them. It is likely performance will be improved with use EHR data more longitudinal records. Conclusions This effective controlling confounding, resulting either higher or similar compared comparators, has ability provide insight into confounders each specific medication–ADR pair, can easily adapted other systems.