Humana reduces emergency hospital admissions with AI-powered patient risk prediction

Humana, a leading US healthcare insurer, partnered with Microsoft Research to leverage AI and the Microsoft Cloud for Healthcare to proactively identify members at high risk for emergency hospital admissions. Traditionally focused on in-patient care and remote monitoring, Humana shifted toward integrating clinical data and key patient event triggers to develop advanced predictive models. These models use neural networks, tree-based models, and deep learning, unified on the Microsoft Cloud, to capture nuanced patient health dynamics. By combining existing single-focus predictive models with structured patient data, the collaboration resulted in over 20% improvement in model precision. Importantly, this was accomplished with strict adherence to data privacy using de-identified information. Enhanced model accuracy allows care teams to act earlier with personalized care plans, helping reduce readmissions and optimize care delivery. The project highlights the impact of precise AI deployment in transforming patient outcomes and reducing healthcare system costs. Humana partnered with Microsoft Research to develop a multivariable AI model integrating data from Humana’s 4.9 million Medicare Advantage members. The research addressed sample imbalance and precision issues through novel deep learning techniques. Resulting improvements in model precision directly enable earlier care team interventions. The effort positions Humana for further analytics-driven care enhancements as models continue to evolve.