Deep Learning

Fillmore and his collaborators have developed deep learning architectures, specifically transformer models, to analyze longitudinal electronic health records for the early detection of high-risk clinical events. This research demonstrates that sequential modeling of diagnostic and medication trajectories can identify patients at risk for pancreatic cancer up to three years before diagnosis and provide dynamic risk assessments for cancer-associated thromboembolism. These findings suggest that deep learning can surpass traditional static scoring systems to improve targeted surveillance and preventative intervention in oncology.