Predicting onward care needs at admission to reduce discharge delay using explainable machine learning.

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Social Care Research

Background:

• Timely identification of patients requiring social care is crucial for efficient hospital discharge.
• Delays in social care referral can lead to prolonged hospital stays.

Purpose of the Study:

• To introduce and evaluate an explainable machine learning (ML) model for early identification of social care needs upon hospital admission.
• To compare ML model performance with clinician predictions and explore hybrid model potential.

Main Methods:

• Development and training of an explainable ML model using routinely collected hospital data (2017-2023).
• Evaluation of model performance using Area Under the Receiver Operating Characteristic curve (AUROC).
• Comparison of ML predictions with clinician assessments for various care needs.

Main Results:

• The ML model achieved high performance (AUROC = 0.915), comparable to clinician predictions.
• ML and clinicians demonstrated complementary strengths in identifying different types of care needs.
• A hybrid model combining ML and clinician predictions yielded superior accuracy (AUROC = 0.936).

Conclusions:

• Explainable ML offers a valuable tool for early detection of social care needs, supporting clinical decision-making.
• Hybrid human-in-the-loop systems integrating ML and clinicians can optimize early predictions for onward social care requirements.