Standard-of-care vs. machine learning-recommended discharge destinations for geriatric surgical inpatients: algorithm...

Area of Science:

• Geriatric Medicine
• Surgical Outcomes
• Artificial Intelligence in Healthcare

Background:

• Older adults undergoing surgery face high risks due to age-related vulnerabilities.
• Comprehensive Geriatric Assessment (CGA) aids discharge planning but faces resource limitations.
• Individualized discharge destination recommendations are crucial for geriatric surgical patients.

Purpose of the Study:

• To evaluate if machine learning (ML) can support individualized discharge destination recommendations for geriatric surgical inpatients.
• To develop and validate an ML model using electronic health record and CGA data.
• To compare ML model performance against standard-of-care discharge decisions.

Main Methods:

• An AdaBoost classifier was developed and validated using data from 169 geriatric surgical patients (age ≥70).
• The model predicted four discharge destinations: home, acute geriatric care, rehabilitation, or nursing home.
• Performance was assessed using accuracy, ROC curves, and calibration, with a fivefold cross-validation framework.

Main Results:

• The ML model achieved 82% accuracy and an ROC AUC of 0.94, significantly outperforming standard-of-care decisions.
• Excluding uncertain predictions increased accuracy to 85%.
• Key predictors included Barthel Index, Clinical Frailty Scale, age, and medication count.

Conclusions:

• Machine learning offers a proof-of-concept for effectively supporting individualized discharge planning in geriatric surgery.
• The ML model demonstrated superior performance compared to standard discharge planning methods.
• Ongoing trials are evaluating the real-world clinical impact of this ML-based approach.