A team in Italy has developed a real-time machine-learning (ML) system that can predict x-ray service times in emergency departments.
The system has the potential to provide managers with early warnings of potential delays in the radiology unit and could enable proactive interventions to improve patient management, according to the group.
“Emergency departments are increasingly challenged by overcrowding, resource shortages, and rising demand for care, which compromise operational efficiency and service quality,” noted lead author Davide Aloini, PhD, of the University of Pisa, and colleagues. The research was published March 19 in Socio-Economic Planning Sciences.
X-rays are the most frequently conducted medical exams in emergency departments (EDs) and use a remarkable proportion of the ED budget, the authors noted. Delays or issues related to x-ray exams can significantly impact length of stay for patients, overcrowding, and the overall functioning of the ED, they added.
While an increasing number of AI and ML models are being applied in EDs, there is a notable gap in studies targeting the prediction of service times for ED diagnostic procedures, the group wrote. Hence, the researchers first developed an ML-based system that forecasts expected service times for x-ray exams -- defined as the complete cycle time from prescription to report release -- when prescribed to a patient.
The model leverages a machine learning technique called "gradient boosting," which builds a strong predictive model by combining a number of weaker models, usually decision trees, and was trained on preprocessed data on patients who presented to the ED over a 27-month period from January 2016 to March 2018. The system exploits 23 predictors, based on the current ED status and patient characteristics.
In a subsequent case study, the researchers evaluated the system’s performance for predicting the service time required to complete x-ray exams prescribed to ED patients. They used a real dataset of 50,070 patients who presented at a medium-sized Italian ED over an approximately two-year period.
“This approach aimed to simulate a real-world scenario in which a new x-ray request is received, and the system, using only the information available up to that moment, provides an immediate forecast,” the group wrote.
According to the findings, the ML forecasting system delivered reasonably accurate predictions for x-ray service times, with an average error of approximately 17 minutes and an accuracy rate of nearly 70% within a 20-minute error margin.
“These results underscore the system's potential applicability and utility for forecasting diagnostic service times within the ED,” the group wrote.
From a scientific perspective, the study demonstrates the potential of developing a forecasting system that accurately predicts ED diagnostics activity service times in real-time using ML algorithms, the group wrote.
In terms of applicability, the case study illustrates how the system can provide ED staff with valuable information for managing patients and alerting them to potential overload issues related to x-ray exams, they added.
“These promising findings potentially lay the foundation for expanding the scope of ED predictive systems to include a broader range of diagnostic activities, such as lab tests and CT scans, as well as treatment processes like orthopedic or ophthalmology consultations,” the researchers concluded.
The full study is available here.
