AI-Infused Post-market Safety Monitoring System Covering All Marketed Drugs and Medicinal Plants
Every year, nearly 100,000 older adults are forced to go to the emergency room because of an adverse drug event (ADE). In fact, between 8% and 14% of all emergency department visits are due to ADEs, half of these events are preventable. Post-market safety monitoring is essential to addressing this epidemic of harm. This invention combines Artificial Intelligence (AI) and a continuously-updated knowledge graph to constantly monitor the safety of prescription drugs across several data sources. The system also unique for monitoring plant products marketed for medicinal use (aka natural products, NPs). It provides users with real-time alerts on newly detected adverse event signals, and instantly generates plausible mechanistic explanations, making signal validation quick and simple and significantly enhancing post-market product safety monitoring.
Description
The system functions by continuously collecting and ingesting structured and unstructured data on drugs and NPs into a curated repository. An AI engine applies sophisticated algorithms, including temporal pattern-matching and recurrent neural networks (e.g., LSTM or GRU), to this data to detect novel safety signals (adverse events). Simultaneously, a knowledge graph is constructed and continuously updated using AI models, organizing complex relationships among prescription drugs, NPs, phytochemicals, biochemical pathways, human physiology, and adverse events. When a safety signal is detected, the system utilizes the knowledge graph to identify and rank plausible mechanisms—the biological or pharmacological processes that could explain the harm. Finally, a user interface presents a report specifying the adverse event signal and the identified mechanisms, which can also be summarized by a large language model (LLM) for clarity. The innovative aspects include the continuous, AI-driven updating of the knowledge graph from new literature and reports (including intelligent document processing of figures and tables document evidence), the use of advanced graph-based queries or machine-learning models (e.g., graph embeddings) to generate and rank potential mechanisms , and the use of transformer neural networks (LLMs) to generate user-facing summaries and mechanism insights. This approach provides a dynamic, evidence-based link between a safety signal and its potential root cause.Applications
- Drug Safety/Pharmacovigilance: Monitoring and alerting for adverse events related to new and existing prescription drugs.- Natural Product/Dietary Supplement Safety: Evaluating the general safety of botanicals, herbal supplements, and their constituents.
- Drug-Natural Product Interaction Screening: Identifying and explaining potentially harmful interactions when drugs and natural products are used concurrently.
- Regulatory Science: Providing systematic, evidence-based reports to regulatory bodies for assessing product risk and guiding action.
- Personalized Medicine/Clinical Decision Support: Assisting clinicians by providing mechanistic insights into potential adverse events for specific drug and NP combinations.
Advantages
- Real-Time Safety Signal Detection: Uses AI algorithms on temporally-organized data to detect novel adverse event signals, including those from complex NP-drug interactions.- Mechanistic Explanation Generation: Automatically constructs and updates a knowledge graph to identify one or more plausible biological or pharmacological mechanisms explaining a detected adverse event.
- Dynamic, Evidence-Based Knowledge: Maintains a continuously updated knowledge graph that incorporates the latest literature and reporting data, supporting evidence-based explanations.
- Enhanced Information Synthesis: Leverages transformer-based large language models (LLMs) to generate clear, user-facing summaries and mechanism insight reports.
- Improved Pharmacovigilance Efficiency: Provides an interactive user interface with alerts, case reports, and synthesized safety information to help users rapidly validate signals and track trends over time.