Reducing Delayed Hospital Discharges Through AI-Driven Documentation and Machine Learning–Based Predictive Models: Impacts on Patient Flow, Length of Stay, and Operational Efficiency

Abstract

Delayed hospital discharges (“bed-blocking”) occur when patients remain in inpatient beds despite being medically fit to leave, leading to unnecessary occupancy, reduced bed turnover, and emergency department (ED) overcrowding. These delays, affecting up to 1 in 7 NHS beds, increase exposure to hospital-acquired infections, functional decline, and psychological stress, while burdening staff and inflating healthcare costs. Causes include post-acute placement delays, diagnostic bottlenecks, medication reconciliation issues, and communication gaps between care teams and community services. This study explores the potential of artificial intelligence (AI) and machine learning (ML) to improve discharge efficiency. A feasibility evaluation compared ChatGPT-4–generated discharge summaries, based on anonymized clinical notes, with those written by junior doctors. General Practitioners, blinded to authorship, rated AI summaries as acceptable in 100% of cases versus <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$92 \%$</tex> for doctors, with both meeting <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\sim 97\%$</tex> of the minimum dataset requirements. McNemar's test and confidence interval analysis supported a significant improvement in acceptability for AI-generated documentation. Additionally, real-time ML models using electronic health record data predicted likelihood of discharge within 24 hours. Logistic regression and XGBoost achieved high performance (AUC 0.81 for length-of-stay prediction; 0.88 for discharge disposition). Integration of these predictions into multidisciplinary rounds enabled early discharge planning, improving patient flow and resource allocation. Findings suggest that combining AI-generated summaries with ML-based discharge predictions can reduce discharge turnaround time, optimize hospital bed utilization, and enhance patient outcomes, offering a scalable strategy to mitigate the operational and clinical consequences of delayed discharges

Ähnliche Arbeiten