P215: Transforming Pharmacovigilance with AI: Revolutionizing Safety Monitoring for Better Patient Outcomes
Poster Presenter
Teresa Zhang
Safety Scientist
Beone Medicines United States
Objectives
The objective of this study is to compare the use of an internally developed Artificial Intelligence (AI) tool for narrative summarization, drug interaction analysis, and differential diagnosis with manual processes to assess the tool’s potential to enhance decision-making in pharmacovigilance.
Method
In this pilot study, safety scientists and physicians evaluated the use of an AI tool for Individual Case Safety Report (ICSR) summary and assessment. Factors such as steps taken, number of external medical resources consulted, and time spent on case summary and analysis were assessed.
Results
Preliminary comparisons suggest that the AI tool powered by a commercial Large Language Model (LLM) may improve decision-making, support analytical thinking and boost productivity in pharmacovigilance. The tool assisted safety scientists and physicians in the areas of automating patient narrative summaries and CTCAE Grading, assessment of drug interactions, and identifying differential diagnoses. In the pilot study, ICSR analysis was 50-80% faster with the AI tool compared to the time needed using manual processes. Additionally, less than 40% of reports generated by the tool needed additional details added by the safety scientist or physician. The steps necessary for ICSR analysis were streamlined with use of the tool versus without, and fewer external medical resources needed to be consulted to make an assessment.
Conclusion
While human oversight is needed to ensure accuracy of AI-generated information, the tool was able to provide concise narrative summarization, accurate CTCAE Grading based on information within the narrative, analysis of infrequent or less well-known drug interactions, and provide multiple differential diagnoses which were supported with clinical rationale on complex cases. The results of this pilot study demonstrated that this internally developed AI tool that leverages current AI models has the potential to enhance pharmacovigilance processes, leading to faster, more accurate, and scalable solutions for enhanced patient safety.
