Automatic Estimation of the Most Likely Drug Combination in Electronic Health Records Using the Smooth Algorithm:...

Area of Science:

• Pharmacovigilance and Pharmacoepidemiology
• Health Informatics
• Computational Pharmacology

Background:

• Electronic health records (EHRs) enable automated pharmacovigilance and pharmacoepidemiology.
• Characterizing complex drug therapies (2+ drugs) in EHRs remains a significant challenge.
• Existing methods require improvement for accurate drug exposure analysis.

Purpose of the Study:

• To develop and describe a novel algorithm for automated therapy characterization.
• The algorithm aims to identify single and combination drug therapies over time.
• To enhance the analysis of drug exposure in EHR data.

Main Methods:

• Developed the 'smooth' algorithm using the Information System for Research in Primary Care EHR platform.
• Algorithm utilizes moving averages and a flexible smoothing parameter (Wt).
• Evaluated Wt impact via simulation and assessed clinical performance through a validation study comparing it to a traditional method.

Main Results:

• The Wt parameter effectively smoothed raw EHR data, with minimal changes (<5%) to total patient data.
• The smooth algorithm demonstrated significantly improved performance over traditional methods in validation studies.
• Performance improvements were consistent across various pharmacological scenarios.

Conclusions:

• The smooth algorithm offers an automated, standardized, and simplified approach for processing drug exposure data in EHRs.
• It is generalizable across most medications and aids in detecting treatment modifications.
• Facilitates improved drug safety monitoring and pharmacoepidemiological research.