AI-driven Predictive Analytics for Hospital Resource Allocation to Optimize Patient Flow and Care Efficiency

Abstract

Background: Bottlenecks in patient flow across the Emergency Department, wards, operating rooms and Intensive Care Unit reduce efficiency. This study set out to map predictive analytics for hospital resource allocation and summarise the associated outcomes and implementation barriers. Methods: A scoping review of empirical studies was conducted by retrieving relevant literature from PubMed, Web of Science and IEEE Xplore. We screened studies using the Population-Concept-Context criteria and charted the data using a six-field framework covering operational use case, data inputs, methods, decision linkage, validation, and outcomes. Results: Twenty studies were included. Most of these were single-site retrospective studies that used Electronic Health Records and systems such as admission, discharge and transfer, bed management systems, and scheduling systems. Time-series forecasting was predominant in bed, census, and demand prediction, whereas supervised machine learning was predominant in admission or crowding prediction, task waiting times, and procedure duration. Probabilistic forecasting was less common. Prescriptive optimisation was rare, mainly taking the form of threshold-based triggers. While most studies reported better prediction metrics than baselines, few assessed operational key performance indicators such as waiting time, boarding, length of stay, throughput, or costs. Reporting on deployment was inconsistent due to constraints including missing or delayed data, interoperability issues, changes during shocks, and limited governance and monitoring. Equity assessments were also rarely reported. Conclusions: Evidence supports prediction, but not consistent operational impact. Future studies should therefore standardise KPIs, strengthen causal evaluations, validate results across sites and over time, monitor drift, and report on fairness and governance.