Predictive Patient Acuity Modeling for the Optimization of Decentralized Public Health Delivery

Abstract

Decentralized public health delivery systems face persistent challenges in resource allocation, particularly when

managing heterogeneous patient populations with fluctuating acuity levels. This study develops and validates a

predictive patient acuity model designed to optimize service distribution across decentralized health units. The

problem addressed is the frequent mismatch between patient needs and available resources, leading to inefficiencies,

delayed care, and increased costs. The purpose of this research is to construct a machine learning-based acuity

prediction framework that integrates real-time clinical and demographic data to forecast short-term patient

deterioration or stability. Using a retrospective cohort design, electronic health records from four decentralized public

health clinics serving a combined population of approximately 120,000 patients were analyzed over 24 months. A

gradient boosting model achieved an accuracy of 89.4% (AUC = 0.92) in predicting high-acuity events within 72

hours. Key findings indicate that predictive acuity modeling reduces resource idle time by 23% and improves patient

to-provider matching efficiency. The conclusion underscores that embedding such models into decentralized health

delivery systems can significantly enhance operational resilience and patient outcomes, particularly in low-resource

settings.