Parkview Health

Care managers at Parkview Health were using a reactive approach to discharge planning, waiting to assess post-acute care needs until after therapist evaluations were complete. A manual 4-question scoring tool proved inaccurate due to subjective interpretation. Delays in identifying post-acute care needs meant patients spent unnecessary extra days in the hospital waiting for placement, exposing them to infection risk and reducing bed availability for more acute patients.

Parkview Health's care management director and data science team built a custom predictive model trained on four years of historical discharge data from Epic's Caboodle data warehouse. Deployed via Epic's cognitive computing platform (Nebula), the model predicts within the first 24 hours of admission whether a patient will need post-acute care. Care managers sort their daily patient lists by this score to prioritize early conversations, therapy evaluations, and facility coordination for high-risk patients.

In the first year after deploying the model in March 2023, care managers eliminated 2,450 excess hospital days and reduced average length of stay by 0.54 days per flagged patient, saving $7.5 million. Patients flagged as high-risk were approximately 4.5 times more likely to actually discharge to post-acute care, validating model accuracy. The approach also improved care team alignment and reduced mixed messaging to patients and families about discharge plans.

• Early identification within 24 hours of admission gives care managers a multi-day head start on coordination tasks that previously didn't begin until day three or four.
• Embedding predictive scores into daily huddles creates consistent messaging across the care team, reducing patient and family confusion.
• Starting with a validated score threshold that maximizes true positives while minimizing false positives is critical to clinician adoption and trust.