Article In Press
Back to the articles list |
Back to browse issues page
1- Nursing and Midwifery Care Research Center, Health Management Research Institute, Iran University of Medical Sciences, Tehran, Iran.
2- Department of Health Information Management and Medical Informatics, School of Allied Medical Science, Tran University of Medical Sciences, Tehran, Iran
3- Nursing and Midwifery Care Research Center, Health Management Research Institute, Iran University of Medical Sciences, Tehran, Iran
4- Nursing and Midwifery Care Research Center, Health Management Research Institute, Iran University of Medical Sciences, Tehran, Iran ,gharacheh.m@gmail.com
2- Department of Health Information Management and Medical Informatics, School of Allied Medical Science, Tran University of Medical Sciences, Tehran, Iran
3- Nursing and Midwifery Care Research Center, Health Management Research Institute, Iran University of Medical Sciences, Tehran, Iran
4- Nursing and Midwifery Care Research Center, Health Management Research Institute, Iran University of Medical Sciences, Tehran, Iran ,
Abstract: (2 Views)
Background and Objective: The Apgar score is a long-established clinical tool used to evaluate the immediate condition of newborns following delivery, assessing five physiological parameters — heart rate, respiratory effort, muscle tone, reflex response, and skin color — at both one and five minutes after birth. Despite its simplicity and usefulness in neonatal assessment, anticipating low Apgar scores before or during childbirth is difficult due to the complex interplay of biological and clinical factors influencing newborn outcomes. Many maternal, fetal, and perinatal characteristics, including maternal age, gestational history, delivery type, and infant physiology, contribute to neonatal conditions.
Early identification of infants at risk for low Apgar scores is crucial for timely intervention and reducing perinatal morbidity and mortality. Traditional statistical models have been applied to study relevant risk factors; however, the exponential growth in healthcare data has enabled the application of advanced machine learning methods that can capture nonlinear relationships and complex factor interactions. Among these, ensemble techniques such as Extreme Gradient Boosting (XGBoost) have shown outstanding performance across medical prediction tasks.
This study aimed to develop a machine learning–based predictive model using XGBoost to assess the risk of low Apgar scores in neonates, based on a massive dataset collected from the Iranian Neonatal and Maternal Network (IMAN)—a national electronic registry system that records comprehensive maternal and neonatal information. The main goal was not only to achieve accurate prediction but also to identify the most critical features driving low Apgar outcomes. The model’s predictive power and interpretability have the potential to support pre‑delivery risk estimation and improve neonatal management nationwide.
Early identification of infants at risk for low Apgar scores is crucial for timely intervention and reducing perinatal morbidity and mortality. Traditional statistical models have been applied to study relevant risk factors; however, the exponential growth in healthcare data has enabled the application of advanced machine learning methods that can capture nonlinear relationships and complex factor interactions. Among these, ensemble techniques such as Extreme Gradient Boosting (XGBoost) have shown outstanding performance across medical prediction tasks.
This study aimed to develop a machine learning–based predictive model using XGBoost to assess the risk of low Apgar scores in neonates, based on a massive dataset collected from the Iranian Neonatal and Maternal Network (IMAN)—a national electronic registry system that records comprehensive maternal and neonatal information. The main goal was not only to achieve accurate prediction but also to identify the most critical features driving low Apgar outcomes. The model’s predictive power and interpretability have the potential to support pre‑delivery risk estimation and improve neonatal management nationwide.
Methods This study utilized a large dataset extracted from the Iranian Neonatal and Maternal Network (IMAN), an electronic health information system that systematically records maternal and neonatal data across healthcare facilities. The dataset included more than one million registered birth records, providing a comprehensive representation of maternal, fetal, and delivery-related characteristics. Such a large dataset offers a valuable opportunity for applying machine learning methods to uncover patterns associated with neonatal outcomes.
Data preprocessing was conducted to ensure the quality and consistency of the dataset. Records containing incomplete or inconsistent information were carefully reviewed, and appropriate data cleaning procedures were implemented. Relevant features related to maternal characteristics, pregnancy conditions, and neonatal attributes were selected as potential predictors. These variables included demographic and physiological factors such as maternal age, infant birth weight, and neonatal head circumference, among other clinical indicators routinely recorded during childbirth.
To build the predictive model, the Extreme Gradient Boosting (XGBoost) algorithm was employed. XGBoost is an advanced ensemble learning technique based on gradient boosting decision trees and is known for its efficiency, scalability, and high predictive performance in structured data analysis. The algorithm iteratively constructs decision trees and combines them to minimize prediction error while controlling model complexity through regularization techniques. This approach enables the model to capture complex nonlinear relationships among predictors and improve generalization performance.
The predictive modeling process involved training the XGBoost model on a large portion of the dataset while reserving a subset of data for validation and testing. Multiple classification schemes were considered to evaluate the model’s ability to predict Apgar score categories with varying levels of granularity. Specifically, the study examined two-class, three-class, and eleven-class classification scenarios for Apgar scores measured at both the first and fifth minutes after birth. Model performance was evaluated using standard classification metrics, including prediction accuracy, allowing for comparison between different classification structures and time points.
In addition to prediction accuracy, feature importance analysis was performed to determine which variables contributed most significantly to the model’s predictions. This step provides valuable clinical insights by identifying key maternal and neonatal factors that are strongly associated with neonatal health status immediately after birth.
Data preprocessing was conducted to ensure the quality and consistency of the dataset. Records containing incomplete or inconsistent information were carefully reviewed, and appropriate data cleaning procedures were implemented. Relevant features related to maternal characteristics, pregnancy conditions, and neonatal attributes were selected as potential predictors. These variables included demographic and physiological factors such as maternal age, infant birth weight, and neonatal head circumference, among other clinical indicators routinely recorded during childbirth.
To build the predictive model, the Extreme Gradient Boosting (XGBoost) algorithm was employed. XGBoost is an advanced ensemble learning technique based on gradient boosting decision trees and is known for its efficiency, scalability, and high predictive performance in structured data analysis. The algorithm iteratively constructs decision trees and combines them to minimize prediction error while controlling model complexity through regularization techniques. This approach enables the model to capture complex nonlinear relationships among predictors and improve generalization performance.
The predictive modeling process involved training the XGBoost model on a large portion of the dataset while reserving a subset of data for validation and testing. Multiple classification schemes were considered to evaluate the model’s ability to predict Apgar score categories with varying levels of granularity. Specifically, the study examined two-class, three-class, and eleven-class classification scenarios for Apgar scores measured at both the first and fifth minutes after birth. Model performance was evaluated using standard classification metrics, including prediction accuracy, allowing for comparison between different classification structures and time points.
In addition to prediction accuracy, feature importance analysis was performed to determine which variables contributed most significantly to the model’s predictions. This step provides valuable clinical insights by identifying key maternal and neonatal factors that are strongly associated with neonatal health status immediately after birth.
Results
The proposed XGBoost model achieved very high predictive accuracy across all classification levels. In the two-class model, accuracy reached 99.11% for the first-minute Apgar prediction and 99.32% for the fifth-minute prediction. The three-class experiments yielded 97.66% and 98.95% accuracy for the first and fifth minutes, respectively. Even in the more complex eleven-class task, the model maintained excellent performance with 94.42% accuracy for the first minute and 94.90% for the fifth minute.
These results highlight XGBoost’s strong capability for modeling nonlinear relationships within large perinatal datasets. The model consistently outperformed traditional approaches reported in similar studies, demonstrating the potential of machine learning to revolutionize neonatal risk prediction.
Feature importance analysis identified several key predictors. The most influential variables were infant birth weight, maternal age, and neonatal head circumference, all of which align with clinical knowledge regarding neonatal health risks. These features exhibited the strongest contributions to model decisions and are biologically plausible indicators of perinatal outcomes. Secondary but relevant factors included gestational age, maternal parity, and type of delivery, suggesting that both maternal and fetal conditions collectively shape neonatal Apgar scores.
By interpreting the relative importance of these factors, the study offers valuable guidance for clinicians seeking early identification of at-risk neonates. The capacity to anticipate low Apgar scores before delivery may improve obstetric decision-making and allow better preparedness in neonatal assistance, especially in high-risk pregnancies and under-resourced hospital settings.
The proposed XGBoost model achieved very high predictive accuracy across all classification levels. In the two-class model, accuracy reached 99.11% for the first-minute Apgar prediction and 99.32% for the fifth-minute prediction. The three-class experiments yielded 97.66% and 98.95% accuracy for the first and fifth minutes, respectively. Even in the more complex eleven-class task, the model maintained excellent performance with 94.42% accuracy for the first minute and 94.90% for the fifth minute.
These results highlight XGBoost’s strong capability for modeling nonlinear relationships within large perinatal datasets. The model consistently outperformed traditional approaches reported in similar studies, demonstrating the potential of machine learning to revolutionize neonatal risk prediction.
Feature importance analysis identified several key predictors. The most influential variables were infant birth weight, maternal age, and neonatal head circumference, all of which align with clinical knowledge regarding neonatal health risks. These features exhibited the strongest contributions to model decisions and are biologically plausible indicators of perinatal outcomes. Secondary but relevant factors included gestational age, maternal parity, and type of delivery, suggesting that both maternal and fetal conditions collectively shape neonatal Apgar scores.
By interpreting the relative importance of these factors, the study offers valuable guidance for clinicians seeking early identification of at-risk neonates. The capacity to anticipate low Apgar scores before delivery may improve obstetric decision-making and allow better preparedness in neonatal assistance, especially in high-risk pregnancies and under-resourced hospital settings.
Conclusion
This investigation demonstrated the effectiveness of machine learning, specifically the XGBoost algorithm, for predicting neonatal Apgar scores using a large and diverse clinical dataset. The model achieved remarkably high predictive accuracy at both one- and five-minute intervals under multiple class structures, confirming its reliability for clinical application. By identifying birth weight, maternal age, and head circumference as leading predictive features, the study provides evidence that machine learning models can uncover clinically meaningful associations from routine medical data.
From a practical perspective, the findings underline the feasibility of integrating data-driven predictive systems into perinatal care workflows. Embedding such models within national hospital information systems could enable real-time neonatal risk monitoring, assisting healthcare professionals in allocating resources and planning interventions effectively. The interpretability and scalability of XGBoost make it suitable for deployment in clinical settings, either as a supplementary decision-support module or as part of automated risk alert systems.
Although the study’s performance metrics are outstanding, further research is recommended to ensure external validation across different populations and healthcare infrastructures. Future work may also explore combining multiple algorithms or including additional maternal health indicators, such as biochemical or imaging features, to further enhance prediction.
In conclusion, the application of advanced machine learning methods to extensive perinatal datasets offers a transformative approach to neonatal care. This study proves that XGBoost-based predictive modeling can reliably anticipate low Apgar outcomes, support timely clinical interventions, and contribute to improved neonatal survival and wellbeing.
This investigation demonstrated the effectiveness of machine learning, specifically the XGBoost algorithm, for predicting neonatal Apgar scores using a large and diverse clinical dataset. The model achieved remarkably high predictive accuracy at both one- and five-minute intervals under multiple class structures, confirming its reliability for clinical application. By identifying birth weight, maternal age, and head circumference as leading predictive features, the study provides evidence that machine learning models can uncover clinically meaningful associations from routine medical data.
From a practical perspective, the findings underline the feasibility of integrating data-driven predictive systems into perinatal care workflows. Embedding such models within national hospital information systems could enable real-time neonatal risk monitoring, assisting healthcare professionals in allocating resources and planning interventions effectively. The interpretability and scalability of XGBoost make it suitable for deployment in clinical settings, either as a supplementary decision-support module or as part of automated risk alert systems.
Although the study’s performance metrics are outstanding, further research is recommended to ensure external validation across different populations and healthcare infrastructures. Future work may also explore combining multiple algorithms or including additional maternal health indicators, such as biochemical or imaging features, to further enhance prediction.
In conclusion, the application of advanced machine learning methods to extensive perinatal datasets offers a transformative approach to neonatal care. This study proves that XGBoost-based predictive modeling can reliably anticipate low Apgar outcomes, support timely clinical interventions, and contribute to improved neonatal survival and wellbeing.
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
Copyright © The Author(s);
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC-By-NC), which permits use, distribution, and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.
Contact Information
