Responsible AI for Predicting Delayed Hospital Discharge Among Older Adults: Development and Evaluation Study for...

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Effective hospital bed capacity management is critical due to increasing demands and limited resources.
• Delayed hospital discharges negatively impact patient outcomes, system efficiency, and strain healthcare resources.
• Predicting discharge delays enables timely interventions to mitigate these negative effects.

Purpose of the Study:

• To develop predictive analytics for delayed discharges in older adults using explainable machine learning.
• To enhance transparency and interpretability of predictive models.
• To integrate fairness metrics to address potential algorithmic biases.

Main Methods:

• Utilized longitudinal data from Ontario, Canada, spanning over two decades.
• Applied extreme gradient boosting and logistic regression models to predict delayed discharges within 90 days post-acute care.
• Assessed model performance using AUC, calibration, and clinical utility, while evaluating fairness across demographic and residential factors. Employed global and local explainability techniques.

Main Results:

• The extreme gradient boosting model achieved an AUC of 0.82, outperforming logistic regression.
• Identified functional/cognitive declines and regional disparities as key predictors of high-risk discharges.
• Bias mitigation strategies improved fairness, highlighting trade-offs between accuracy, fairness, and interpretability.

Conclusions:

• Responsible AI holds significant potential for healthcare, balancing predictive accuracy, equity, and interpretability.
• The study identified systemic gaps and provided insights for improving discharge planning and resource optimization.
• Actionable insights can lead to more equitable care delivery and efficient healthcare systems.