On Thursday, medical artificial intelligence (AI) company Vuno announced that its research on an AI-based early warning system for predicting the need for intubation in pediatric intensive care patients has been published in the international journal Heart & Lung (IF 2.6).
This collaborative study was conducted with pediatric and respiratory medicine research teams from Yangsan Busan National University Hospital.
The researchers analyzed data from pediatric intensive care unit patients to develop a deep learning model that predicts the need for mechanical ventilation, specifically invasive intubation, and compared it to existing models.
Developing AI Specifically for Pediatrics, Overcoming the Limitations of Adult-Centered Models
Acute respiratory failure is a critical factor that can lead to pediatric intensive care admission and cardiac arrest. As the risk increases with delayed patient assessment, timely invasive mechanical ventilation is crucial.
However, pediatric patients present unique challenges due to their wide age range and diverse underlying conditions, making early detection of deterioration difficult. Most existing predictive models have been developed for adults, limiting their applicability to pediatric patients.
To address this gap, the research team analyzed electronic medical records of 1,318 patients under 18 years old admitted to the pediatric intensive care unit at Yangsan Busan National University Hospital from 2012 to 2022. They developed and validated an AI-based early warning system (DeePedIMV) that predicts the need for invasive mechanical ventilation up to eight hours in advance.
Unlike existing models limited to specific disease groups, DeePedIMV can make universal predictions by accounting for acute deterioration from various causes.
Achieving an AUROC of 0.88, Confirming Over 50% Reduction in Alarm Frequency Compared to Existing Models
The study results showed that DeePedIMV demonstrated excellent predictive performance, with a prediction accuracy (AUROC) of approximately 0.88. This performance was consistent across all age groups and disease types, with the highest accuracy observed in patients under one year old.
Notably, DeePedIMV recorded an AUPRC of about 0.47, a key indicator for precisely identifying actual risk situations, proving to be more than three times superior to existing models. Additionally, it reduced the number of alarms by more than half while maintaining the same sensitivity, effectively improving the working environment for healthcare professionals.
Vuno’s Chief Technology Officer (CTO), Joo-Seong Hoon, stated that the deep learning algorithm developed through this research has proven its ability to reduce unnecessary alarms while proactively predicting high-risk pediatric patients who require invasive mechanical ventilation. It will continue to strive to ensure that Vuno’s technology contributes to patient safety in both pediatric and adult care.