A super‐combo‐drug test to detect adverse drug events and drug interactions from electronic health records in the era of polypharmacy

Abstract

Pharmacoinformatics research has experienced a great deal of successes in detecting drug‐induced adverse events (AEs) using large‐scale health record databases. In the era of polypharmacy, pharmacoinformatics faces many new challenges, and two significant challenges are to detect high‐order drug interactions and to handle strongly correlated drugs. In this article, we propose a super‐combo‐drug test (SupCD‐T) to address the aforementioned two challenges. SupCD‐T detects drug interactions by identifying optimal drug combinations with increased AE risks. In addition, SupCD‐T increases the statistical powers to detect single‐drug effects by combining strongly correlated drugs. Although SupCD‐T does not distinguish single‐drug effects from their combination effects, it is noticeably more powerful in selecting an individual drug effect in the multiple regression analysis, where confounding justification between two correlated drugs reduces the power in testing the individual drug effects on AEs. Our simulation studies demonstrate that SupCD‐T has generally better power comparing with the multiple regression analysis. In addition, SupCD‐T is able to select meaningful drug combinations (eg, highly coprescribed drugs). Using electronic health record database, we illustrate the utility of SupCD‐T and discover a number of drug combinations that have increased risk in myopathy. Some novel drug combinations have not yet been investigated and reported in the pharmacology research.

Random control selection for conducting high‐throughput adverse drug events screening using large‐scale longitudinal health data

“…Pharmacoinformatic studies have also identified ADE signals from longitudinal health data including electronic health record (EHR) data and administrative claims data. 15 , 16 , 17 Unlike SRS, EHR data and administrative claims data contain individual level and longitudinal information, including clinical outcomes (i.e., diagnoses) and medications (i.e., pharmacy prescriptions/claims). Compared to SRS, EHR data and administrative claims data contain more population groups (i.e., individuals without any ADE), variables (i.e., health conditions other than ADE), and detailed temporal information.…”

“…Pharmacoinformatic studies have also identified ADE signals from longitudinal health data including electronic health record (EHR) data and administrative claims data 15–17 . Unlike SRS, EHR data and administrative claims data contain individual level and longitudinal information, including clinical outcomes (i.e., diagnoses) and medications (i.e., pharmacy prescriptions/claims).…”