Researchers Pioneer Innovative AI for Forecasting Stroke Prevention Treatments
Researchers at The Ohio State University have pioneered an innovative artificial intelligence system capable of forecasting the most effective stroke prevention treatments for patients with heart disease. Its predictive accuracy eclipses that of conventional methodologies by emulating randomized clinical studies.
Pre-loaded with a comprehensive assembly of health care claims, the AI system refines its forecasting abilities tailoring to specific medical conditions. Consequently, it can identify the optimal course of treatment for a patient based on their individual profile. During the unveiling, the researchers reported that their system outperformed seven competing models and aligned with the treatment choices of four actual clinical studies.
Transforming Clinical Trials and Customized Patient Treatment
Ping Zhang, an associate professor at Ohio State with expertise in both computer science and engineering as well as biomedical informatics, stressed the unique qualities of the model. Zhang stated, “No current algorithm can perform this task. Our method improved results by 7% to 8% beyond other models. While other algorithms might reach similar conclusions, they cannot replicate the precise outcomes of randomized clinical trials. Ours can.”
The ambitions of the research group go beyond the mere duplication of results from clinical trials. They aspire to streamline the clinical trial process and customize treatments by proposing fewer, yet more promising, drug candidates for research. This could significantly cut down the time and financial investment typically necessary for conventional research approaches.
Named CURE (CaUsal tReatment Effect estimation), this AI innovation integrates patient information with biomedical knowledge graphs, enhancing its predictive accuracy. Ruoqi Liu, a PhD candidate in computer science and engineering, explained, “The model is pre-trained on widespread datasets without being confined to specific treatments. We then fine-tune the pre-trained model on small-scale, task-oriented datasets, enabling rapid adaptation across various tasks.”
Dr. Zhang projects that in the near future, physicians could utilize algorithms strengthened by the electronic health records of millions to form a “digital twin” of a patient, facilitating individualized treatment recommendations. “This model isn’t just optimism: It’s based on comprehensive data and foundational model AI. We can assert with justified confidence which treatment approach is superior,” Zhang remarked.
This study was unveiled in the May 1, 2024 issue of the journal Patterns, with funding from the National Institutes of Health. Among the co-authors focusing on CURE and its knowledge graph element, KG-TREAT, are Pin-Yu Chen of IBM Research and Lingfei Wu of Anytime AI.
