• Precisely determine clinical plan for in-patient at every stage of acute care
• Use EHR data between a patient’s time of presentation to hospital and admission to estimate probability patient dies during hospital stay
• Developed in 2019 by Duke Institute of Innovation Health using machine learning; vetted using prospective evaluation

See paper describing this model.

Live in Duke university health and embedded into their work-flow to improve clinical decisions as early as during in-patient admission

Robust development using rich training, validation, and (prospective and external) evaluation data:

• 75,247 hospital admissions in total
• Training and validation using 43,180 hospitalizations from one hospital
• Externally validated using 26,794 hospitalizations from three other hospitals
• Prospective (real-time) validation using 5,273 hospitalizations over two-month period