TY - JOUR
T1 - Predictive modelling in times of public health emergencies: patients’ non-transport decisions during the COVID-19 pandemic
AU - Farhat, Hassan
AU - Abid, Cyrine
AU - Alinier, Guillaume
AU - Khadhraoui, Moncef
AU - Gargouri, Imed
AU - Shaikh, Loua Al
AU - Laughton, James
N1 - © 2025 The Author(s). This is an open access article distributed under the Creative Commons Attribution License, to view a copy of the license, see: https://creativecommons.org/licenses/by/4.0/
PY - 2025/9/11
Y1 - 2025/9/11
N2 - BACKGROUND: During the COVID-19 pandemic, there was a surge in pre-hospital emergency calls due to the increased prevalence of flu-like symptoms and panic related to the pandemic. However, some patients declined transportation to hospital due to their fear of accessing healthcare facilities. This posed a significant risk to their health outcomes. This study aimed to utilise an extensive dataset, which included the period of the COVID-19 pandemic, in a modern Middle Eastern Emergency Medical Service to comprehend and predict the behaviour of non-transport decisions, a major multi-variable factor in pre-hospital emergency medicine.METHODS: Using Python
® programming language, this study employed various supervised machine-learning algorithms, including parametric probabilistic models, such as logistic regression, and non-parametric models, including decision trees, random forest (RF), extra trees, AdaBoost, and k-nearest neighbours (KNN), using a dataset of non-transported patients (refused transport and did not receive treatment versus those who refused transport and received treatment) between 2018 and 2022. Model performance was comprehensively evaluated using Accuracy, F1 score, Matthews correlation coefficient (MCC), receiver operating characteristic area under the curve (ROC AUC), kappa, and R-squared metrics to ensure robust model selection.
RESULTS: From June 2018 to July 2022, 334,392 non-transport cases were recorded. The random forest model demonstrated the best optimised predictive performance, with accuracy = 74.78%, F1 = 0.74, MCC = 0.35, ROC AUC = 0.81, kappa = 0.34, and R-squared = 0.81.CONCLUSION: This study indicated that predictive modelling could accurately help identify patients who refuse transport to hospital and may not require treatment on the scene. This enables them to be redirected from the call-taking phase to alternative primary healthcare facilities. This reduces the strain on emergency healthcare resources. The findings suggest that machine learning has the potential to revolutionise pre-hospital care, especially during pandemics, by improving resource allocation and patient outcomes, while highlighting the need for ongoing research to refine these models.SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12873-025-01340-7.
AB - BACKGROUND: During the COVID-19 pandemic, there was a surge in pre-hospital emergency calls due to the increased prevalence of flu-like symptoms and panic related to the pandemic. However, some patients declined transportation to hospital due to their fear of accessing healthcare facilities. This posed a significant risk to their health outcomes. This study aimed to utilise an extensive dataset, which included the period of the COVID-19 pandemic, in a modern Middle Eastern Emergency Medical Service to comprehend and predict the behaviour of non-transport decisions, a major multi-variable factor in pre-hospital emergency medicine.METHODS: Using Python
® programming language, this study employed various supervised machine-learning algorithms, including parametric probabilistic models, such as logistic regression, and non-parametric models, including decision trees, random forest (RF), extra trees, AdaBoost, and k-nearest neighbours (KNN), using a dataset of non-transported patients (refused transport and did not receive treatment versus those who refused transport and received treatment) between 2018 and 2022. Model performance was comprehensively evaluated using Accuracy, F1 score, Matthews correlation coefficient (MCC), receiver operating characteristic area under the curve (ROC AUC), kappa, and R-squared metrics to ensure robust model selection.
RESULTS: From June 2018 to July 2022, 334,392 non-transport cases were recorded. The random forest model demonstrated the best optimised predictive performance, with accuracy = 74.78%, F1 = 0.74, MCC = 0.35, ROC AUC = 0.81, kappa = 0.34, and R-squared = 0.81.CONCLUSION: This study indicated that predictive modelling could accurately help identify patients who refuse transport to hospital and may not require treatment on the scene. This enables them to be redirected from the call-taking phase to alternative primary healthcare facilities. This reduces the strain on emergency healthcare resources. The findings suggest that machine learning has the potential to revolutionise pre-hospital care, especially during pandemics, by improving resource allocation and patient outcomes, while highlighting the need for ongoing research to refine these models.SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12873-025-01340-7.
KW - Non-transport
KW - Machine learning
KW - EMS
KW - Prediction
KW - Pre-hospital
UR - https://www.scopus.com/pages/publications/105015358504
U2 - 10.1186/s12873-025-01340-7
DO - 10.1186/s12873-025-01340-7
M3 - Article
C2 - 40936081
SN - 1471-227X
VL - 25
SP - 181
JO - BMC Emergency Medicine
JF - BMC Emergency Medicine
IS - 1
M1 - 181
ER -