Recently, I2DB students Levi Kaster, Nicholas Hadas, Guntaas Shergill and Isaac Kyeremateng competed in the Creating AI-Enabled All-Health Team Data Fabric (CAIDF) Hackathon, hosted at the University of Illinois Chicago and supported by University of Missouri-Columbia, University of Iowa, Loyola University Chicago, TackleAI, and Microsoft. Their presentation, “Predicting 30-day Fall Readmission Through LLM-Enhanced Predictive Modeling,” was named the top project in predictive analytics and will receive a $5,000 prize.
In just over 24 hours, the team developed 30-day fall re-admission predictive models for multiple sites, incorporating LLM-extracted features from clinical notes to enhance performance and interpretability. Additionally, they built an LLM-powered interactive dashboard to display patient-level model results. The dashboard included key patient demographics, social determinants of health, predicted re-admission risk, and a toggle to show the biggest factors influencing re-admission risk, as well as an AI-generated patient summary suggesting a plan of care and highlighting additional import factors.