Adverse drug effect detection.

Area of Science:

• Pharmacovigilance
• Health Informatics
• Data Science

Background:

• Electronic patient records offer rich, underutilized data for real-world medicine use analysis.
• Drug safety signal detection from longitudinal observational data is a growing area of interest.

Purpose of the Study:

• To propose and evaluate novel algorithms for adverse drug effect discovery using electronic patient records.
• To assess the performance of these algorithms against state-of-the-art methods and a baseline.

Main Methods:

• Developed a likelihood ratio model and a Bayesian network model for adverse drug effect discovery.
• Utilized a simulated Observational Medical Outcomes Partnership (OMOP) dataset with predefined adverse drug effects for evaluation.
• Compared the proposed algorithms' performance with the Bayesian confidence propagation neural network and a chi-square baseline.

Main Results:

• The proposed likelihood ratio and Bayesian network models demonstrate comparable performance to the Bayesian confidence propagation neural network.
• Combining the three algorithms (likelihood ratio, Bayesian network, and BCPNN) yields superior results due to solution diversity.
• The novel pattern discovery method improved upon the standard chi-square algorithm by 23.83% on the simulated OMOP dataset.

Conclusions:

• The proposed algorithms are effective for adverse drug effect discovery from electronic patient records.
• Combining diverse algorithmic approaches enhances the reliability of drug safety signal detection.
• These methods offer a valuable tool for improving patient safety through better understanding of real-world drug effects.