This invention is a research tool that uses AI to monitor and improve machine learning models for predicting adverse events and preventing medication errors in health systems. Its most significant advantage is a custom generative AI model that can identify issues in machine learning models and automatically initiate improvements.
Description
The technology integrates four main components to help health systems prevent medication errors. First, a dashboard allows users to track patients at risk for adverse events like falls, fractures, and strokes. Second, a large-scale data analytics platform processes electronic health record (EHR) data, improving its quality and preparing it for analysis. Third, a machine learning specification engine is used by health system stakeholders to rapidly create and apply adverse event risk models using the processed EHR data. Finally, a custom fine-tuned generative AI model identifies issues such as model drift or bias within the machine learning models and automatically triggers improvements via an agent that communicates with the machine learning specification engine. This process leads to a newly evaluated model that, if it meets performance criteria, becomes the new default risk prediction model. The generative AI model takes on the role of a clinical expert to evaluate the plausibility of the model's predictions and reason about potential improvements.
Applications
- Hospitals and health systems
- Medical data analytics companies
- Healthcare technology providers
- Electronic health record (EHR) software developers
Advantages
- Novel AI Application: The system uses a custom fine-tuned generative AI to both identify issues in machine learning models and initiate improvements, a feature that is considered completely novel.
- Proactive Error Prevention: It helps health systems avoid preventable adverse events by identifying and tracking high-risk patients.
- Automated Model Improvement: The generative AI can automatically initiate changes to the machine learning model specification, leading to a freshly evaluated and improved model.
- Rapid Development: The machine learning specification engine/UI is designed for rapid iteration, evaluation, and improvement of adverse event risk models.
- Actionable Insights: The system provides concise and informative summaries of patient data, which assists users in assessing the need for intervention.
Invention Readiness
The technology is at the prototype exists stage of development. It is based on a concept that has been fully defined and relies on a specific ecosystem that requires significant know-how to implement.
IP Status
Patent Pending
