Mining association patterns of drug-interactions using post marketing FDA's spontaneous reporting data

Area of Science:

• Pharmacovigilance and Drug Safety
• Data Mining and Machine Learning
• Clinical Pharmacology

Background:

• Pharmacovigilance (PhV) is crucial for public health and clinical research, aiming to detect novel adverse drug events (ADEs).
• Drug interactions (DI) contribute significantly to unexpected ADEs, necessitating early detection.
• Spontaneous reporting systems (SRS) are primary data sources for post-marketing PhV.

Purpose of the Study:

• To identify drug interaction (DI) signals within spontaneous reporting systems (SRS).
• To present an optimized, tailored data mining algorithm named "hybrid Apriori" for DI signal detection.
• To assess the algorithm's ability to extract significant drug interaction-adverse event (DIAE) patterns.

Main Methods:

• Applied a modified association rule mining (ARM) approach using the hybrid Apriori algorithm to the U.S. FDA's FAERS database.
• Extracted drug interaction-adverse event (DIAE) patterns and assessed them using a three-element taxonomy and performance metrics.
• Employed logistic regression (LR) to quantify interaction magnitude and direction, controlling for co-medication effects.

Main Results:

• The hybrid Apriori algorithm identified 2933 interacting DIAE patterns, including 1256 serious cases.
• The method demonstrated high performance with an average precision of 85%, negative predictive value of 80%, sensitivity of 81%, and specificity of 84%.
• Results confirmed the algorithm's capability to detect serious interacting DIAEs and provided statistical context against spurious findings via LR modeling.

Conclusions:

• The developed hybrid Apriori method effectively detects drug interaction-adverse event (DIAE) signals from SRS data.
• The algorithm shows proficiency in identifying rare adverse drug reactions (ADRs).
• This approach enhances the timely detection of potential drug safety issues from real-world data.