UnityPoint Health

UnityPoint Health struggled with inaccurate patient discharge estimates, with an estimated fill rate below 60% and fewer than a third of patients having an estimated discharge date within the first few days of admission. Manual data entry processes created significant work burden for nursing staff without producing reliable results. Long encounters were progressively harder to predict, undermining discharge planning and capacity management.

UnityPoint Health developed a real-time length of stay (LOS) predictive model trained on 120,000 encounters over two years using 225 data elements from the Health Catalyst Data Operating System (DOS™) platform. The model runs daily at 0700 and is deployed directly into the EHR workflow, providing automated discharge estimates to clinicians. A three-month silent pilot allowed refinement and clinical validation before broad rollout, accompanied by stakeholder education for frontline staff, physicians, and executives.

Over 1.5 years, the LOS predictive model paired with other LOS initiatives reduced expenses by $41M and eliminated 38,000 excess LOS days. Nursing labor savings reached 5,000 hours annually by eliminating manual data entry. Prediction accuracy increased by 40%, and the number of patients with an estimated LOS within 24 hours of admission increased fourfold.

• Engaging physicians with comparative literature and using clinical terminology when presenting model features was critical for earning trust and adoption.
• Running the model silently for three months before launch generated validation data and allowed refinement without disrupting workflows.
• Integrating predictions directly into existing EHR workflows—rather than requiring separate tools—drove consistent daily use by multidisciplinary care teams.