Data Mining for Drug-Drug Interactions and More

Background: In the United States alone, there are more than 19,000 approved drug products. The number of potential drug combinations that patients use in the real world is exponentially larger. Real-world data (RWD; e.g., medical and pharmacy data collected in the course of routine clinical care and billing) offer an important opportunity to efficiently identify which combinations may result in clinically relevant drug-drug interactions (DDIs) via semi-automated, high-throughput data mining. However, such data mining must account for numerous sources of bias (e.g., multiple dimensions of confounding or heightened opportunity for immortal time bias) that are particularly salient in the setting of exposures defined by the use of two or more drugs.

Objectives: To describe and discuss opportunities, challenges, and potential solutions for rigorous data mining for DDIs. Participants will leave the session with a better understanding of the principles and techniques of data mining, as well as considerations specific to data mining for DDIs.

Description: This symposium will convene panelists from academia and industry who are at the forefront of DDI data mining methodology using RWD. Dr. Gagne will motivate the need for data mining for DDIs, describe general principles of data mining, and highlight additional challenges specific to data mining for DDIs in RWD (15 mins). Dr. Miano will describe the use of semi-automated, high-throughput self-controlled case series studies to identify DDI signals while controlling for confounding; and the use of semi-Bayesian adjustment to correct for multiple testing (20 mins). Dr. Donneyong will discuss challenges and opportunities in identifying DDIs involving more than two drugs using a novel approach that combines data mining, pharmacoepidemiology, and clinical pharmacology (20 mins). Dr. Bykov will discuss examples of follow-up etiologic studies that aim to confirm or refute DDI signals from case-crossover-based DDI data mining, highlighting the importance of (1) externally validating DDI signals generated from data-mining studies; (2) evaluating and striving to improve data mining approaches; and (3) contextualizing the burden of false positives (20 mins). Lastly, the presenters will facilitate an interactive discussion with the audience and provide their visions for future directions in the broader field of data mining and DDI data mining specifically (15 mins). While this symposium focuses on DDIs, it will be useful to any ISPE member interested in the fundamentals of data mining.