Explainable Artificial Intelligence for Pharmacovigilance: What Features Are Important When Predicting Adverse Outcomes?

Machine learning can help identify which medications increase risk of serious health events: The study used AI models to analyze health records of over 65-year-olds and successfully predicted which individuals were at risk of Acute Coronary Syndrome (ACS), a serious heart condition, based on their medication and medical history.

Explainable AI makes drug safety monitoring more transparent: By using XAI techniques, researchers could identify which specific drugs contributed to adverse outcome predictions—for example, finding that rofecoxib and celecoxib (pain medications) had measurable associations with ACS risk—rather than having a "black box" model that gives predictions without explanation.

This approach could modernize pharmacovigilance systems: Traditional drug safety monitoring relies on voluntary reporting of side effects, which is often incomplete. This study demonstrates that analyzing existing health records with explainable AI could provide a more systematic way to detect medication risks and protect patients.

The method works with real-world patient data: The researchers used actual linked health datasets from Western Australia spanning 1993-2009, showing this approach is feasible with the types of records that healthcare systems already maintain, making it potentially practical to implement.