Overtriage, Undertriage, and Value of Care after Major Surgery: An Automated, Explainable Deep Learning-Enabled...

Area of Science:

• Healthcare Management
• Medical Informatics
• Surgical Outcomes

Background:

• Overtriaging low-risk patients to ICUs leads to low value of care.
• Undertriaging high-risk patients to general wards increases mortality and morbidity.
• Automated systems can improve postoperative patient classification.

Purpose of the Study:

• To test the reproducibility of an automated postoperative triage classification system.
• To generate an actionable and explainable decision support system for patient triage.
• To evaluate the impact of triage accuracy on patient outcomes and healthcare costs.

Main Methods:

• A longitudinal cohort study at two university hospitals.
• An explainable deep learning model used EHR data for triage.
• Nearest neighbor algorithms identified risk-matched controls for outcome comparison.

Main Results:

• Overtriage (5.1% of ICU admissions) resulted in higher costs and lower value of care.
• Undertriage (12.0% of ward admissions) was linked to increased mortality, morbidity, and lower value of care.
• The automated triage system demonstrated reproducibility across institutions.

Conclusions:

• Automated postoperative triage systems are reproducible and can enhance clinical decision-making.
• Accurate patient triage is crucial for optimizing value of care and patient outcomes.
• Addressing both overtriaging and undertriaging is essential for improving surgical patient management.