Responsible AI for Predicting Delayed Hospital Discharge Among Older Adults: Development and Evaluation Study for Balancing Accuracy, Equity, and Explainability

BACKGROUND: Amid growing demands and constrained health care resources, effective hospital bed capacity management is crucial. Delayed hospital discharge, where patients remain in the hospital beyond the need for acute care, strains resources, affects patient outcomes, and reduces system efficiency. Predicting such delays facilitates early interventions to avert them and alleviate burdens on patients, care partners, hospitals, and the broader health care system. OBJECTIVE: This study aimed to develop comprehensive predictive analytics for delayed discharges among older adults using explainable machine learning to boost transparency and interpretability, while integrating fairness to mitigate algorithmic biases. METHODS: Leveraging longitudinal data from over 2 decades in Ontario, Canada, we applied extreme gradient boosting and logistic regression models to predict delayed discharges within 90 days post-acute care. Data preprocessing included a 2-year look-back for clinical histories and balanced sampling to address class imbalance. Model performance was assessed via area under the receiver operating characteristic curve, calibration, and clinical utility. Fairness was evaluated across sex, urban or rural residence, and residential instability using several threshold-free metrics. Explainability was examined at the global model level (via partial dependence plots and permutation feature importance) and locally (via Shapley Additive Explanations, breakdown, and ceteris paribus methods), with principal component analysis used to cluster key features for high-risk patients. RESULTS: The extreme gradient boosting model outperformed logistic regression, achieving an area under the receiver operating characteristic curve of 0.82 on the test set, with acceptable within-group and cross-group ranking fairness across subgroups. Explainability clustering analyses identified functional and cognitive declines (eg, care support needs, dementia, and mobility issues) and regional disparities as primary drivers of high-risk predictions. Bias mitigation improved calibration parity, especially when stratifying by residential instability, underscoring the trade-offs policymakers must weigh between accuracy, fairness, and explainability. CONCLUSIONS: This study demonstrates the potential of responsible artificial intelligence in health care, emphasizing the need to balance predictive accuracy, equity, and interpretability. It uncovers systemic gaps and offers actionable insights for enhanced discharge planning, resource optimization, and equitable care delivery.