Hartford HealthCare cuts length of stay by 5% with H2O AI discharge prediction platform

Less experienced clinicians often take a more cautious approach to discharge readiness, translating into longer stays and more avoidable patient days across the system. The cost extended beyond financial impact — every extra day increased patient risk of infections, falls, and deconditioning, while slowing throughput for the next patient waiting for a bed.

Hartford HealthCare partnered with MIT applied mathematician Dimitris Bertsimas to co-create H2O (Holistic Hospital Operations), a machine learning analytics platform that analyzes de-identified patient data to predict discharge readiness. The tool was initially standalone, then embedded directly into the Epic EHR based on physician feedback, and is used alongside unit-based progression rounds where EDD and AI prediction discrepancies surface fixable barriers.

Nearly every admitted medical patient now receives a discharge readiness prediction within 24 hours of admission. Combined with standardized progression rounds and physician-owned expected discharge dates, Hartford HealthCare achieved approximately a 5% reduction in overall length of stay versus the pre-implementation period, with greater accountability to planned discharge dates.

• Embedding AI directly into the existing EHR (Epic) was critical to adoption — a standalone app had lower usability and impact.
• Pairing AI predictions with disciplined operational routines (standardized rounding, physician-owned EDDs) drives outcomes; AI alone is insufficient.
• Involving physician leaders from day one in both design and daily workflows builds clinical trust and keeps final decisions appropriately human.