A Novel AI Tool Revolutionizing Stroke Prevention for Heart Disease Patients
A novel AI tool, crafted by specialists at The Ohio State University, is making strides in forecasting effective stroke prevention methodologies for heart disease patients. This innovative AI platform is set to enhance clinical trial processes and tailor patient treatment with greater precision.
Exploiting a comprehensive set of healthcare claim data from various sources, including employers, hospitals, and health insurance plans, the AI leverages a foundational model strategy similar to that used by generative AI technologies such as ChatGPT. This sentiment is echoed by the senior author Ping Zhang, who holds dual roles as an associate professor in computer science and engineering and biomedical informatics. Zhang emphasizes the uniqueness of the AI by saying no other existing algorithms can match its capabilities. The AI’s ability to process general data and apply it to specific health conditions and treatments allows it to replicate the effects of randomized clinical trials without incurring the same significant costs and delays traditionally involved in such research.
AI Proves Successful in Clinical Trials
As published in the journal Patterns on May 1, 2024, the AI surpassed seven current methodologies and correlated precisely with the outcomes of four real-world clinical trials. Surpassing its peers, the AI demonstrated a performance increase of 7% to 8%, providing treatment guidance that closely reflects clinical trial recommendations. Its goal is not to replace, but rather to accelerate and complement clinical research.
Efficiency and the Pre-Training Phase
The study’s lead author, Ruoqi Liu, a doctoral candidate in Zhang’s lab, refers to the AI, named CURE (CaUsal tReatment Effect estimation), as a tool that could potentially aid clinicians in streamlining their search for the most effective treatments prior to initiating focused clinical trials.
At the heart of its efficiency is its pre-training phase, which utilizes an enormous range of unlabeled data from real-world scenarios, fostering its adaptability to different diseases and treatments. MarketScan Commercial Claims and Encounters data from 2012-2017 provided over 3 million patient records for the study, including thousands of medical codes. The development of this model also incorporated innovative gap-filling techniques and the use of knowledge graphs that significantly bolstered the predictive capabilities of CURE.
Digital Twin and Clinical Decision-Making
By initially ingesting a large body of general data and then refining its focus with disease and treatment-specific details, the model can predict outcomes across various treatment options. It essentially crafts a “digital twin” of the patient, which is then used to inform and enhance clinical decision-making.
Laying a Path to Future Patient Treatment
Zhang’s team is hopeful that, with FDA endorsement, an AI tool could emerge that utilizes electronic health records of millions of patients to inform and guide clinicians in their decision-making process. Zhang emphasizes the goal of empowering physicians with confidence in prescribing the best treatment pathways through the integration of massive datasets and foundational AI technology.
The project received financial backing from the National Institutes of Health and was strengthened by collaborations with experts such as Pin-Yu Chen from IBM Research and Lingfei Wu from Anytime AI. With regulatory support on the horizon, this AI breakthrough has the potential to play an essential role in the future of patient treatment.
