Real-time prediction of cardiorespiratory deterioration during paediatric critical care transport using interpretable
Milan Kapur1, Kezhi Li2, Alexander Brown3
1Department of Population, Policy and Practice, UCL Great Ormond Street Institute of Child Health, London, United Kingdom.
Insights
Machine learning models can predict critical illness in children during transport. These explainable AI tools offer early warnings for respiratory and cardiovascular events, improving patient care.
Area of Science:
- Pediatric critical care medicine
- Machine learning applications in healthcare
- Biomedical data science
Background:
- Interhospital transport of critically ill children presents significant risks, including potential for sudden respiratory and cardiovascular decline.
- Early detection of patient deterioration is crucial for timely medical intervention and preventing adverse outcomes.
Purpose of the Study:
- To develop and assess lightweight, explainable machine learning models for forecasting adverse physiological events in critically ill children during interhospital transport.
- To predict patient deterioration up to 15 minutes in advance using real-time vital signs and clinical data.
Main Methods:
- Development and evaluation of transformer-based machine learning models using time-series vital sign data and vector-embedded diagnoses.
- Models were trained and validated on data from 1,519 interhospital transports of critically ill children (2016-2021).
- Integrated Gradients were used for model interpretability, ensuring alignment with clinical reasoning.
Main Results:
- Transformer models incorporating vital sign time-series and diagnoses achieved high predictive performance: AUROC of 0.851 for respiratory and 0.792 for cardiovascular deterioration.
- The developed models are lightweight and designed for deployment on edge devices, enabling real-time predictions.
- Model interpretability confirmed that predictions align with established clinical judgment.
Conclusions:
- Real-time, explainable machine learning models can accurately predict deterioration in critically ill children during interhospital transport.
- These AI tools, utilizing routinely collected data, have the potential to enhance early clinical intervention and improve patient safety.
- The models offer a valuable solution for resource-limited transport settings, providing interpretable risk predictions.
Abstract:
Interhospital transport of critically ill children carries inherent risks, including unexpected respiratory and cardiovascular deterioration. Early warning of impending patient deterioration may allow physicians to intervene and prevent further decline. We developed and evaluated lightweight, explainable machine learning models to forecast adverse physiological events up to 15 minutes in advance using continuously streamed vital signs and clinical data. Models were trained and evaluated on 1,519 transports conducted by a specialist paediatric critical care team in London (2016-2021). Transformer-based models incorporating vital sign time-series and vector-embedded diagnoses outperformed simpler models, achieving AUROC scores of 0.851 for respiratory and 0.792 for cardiovascular deterioration. Model interpretability was provided using Integrated Gradients, revealing alignment with clinical reasoning. Designed for deployment on edge devices, these models offer real-time, interpretable risk predictions in resource-limited transport settings. These results demonstrate that real-time, explainable machine learning models can accurately predict deterioration during interhospital paediatric transport using routinely collected data, supporting their potential role in enhancing early clinical intervention.