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
Abstract: (192 Views)
Purpose: The Apgar score is a crucial tool for assessing the general condition of a newborn in the first minutes after birth. Effective identification and management of neonates at risk of a low Apgar score can help reduce perinatal complications and mortality. Given the importance of this issue, the present study aims to predict low Apgar scores post-delivery using machine learning algorithms based on clinical and demographic characteristics.
Material and Methods: In this study, the database of the Iranian Maternal and Neonatal Network (IMAN) registry, comprising over one million records from 2020 to 2021, was utilized. Modeling for Apgar score prediction was performed using the XGBoost machine learning algorithm. Predictive models were developed in three different classification scenarios (two-class, three-class, and eleven-class) to estimate the Apgar score at the first and fifth minutes. To validate the results, the dataset was split into training and testing subsets, and the performance of the implemented methods was evaluated using the accuracy metric.
Results: Feature importance analysis revealed that fetal anatomical indices (which are estimable pre-delivery via ultrasound), including weight and head circumference, along with maternal age, were the most influential variables in the model’s decisions and Apgar score prediction. Considering the different categorization systems for the Apgar score, the predictive performance of five distinct models was compared, yielding significant results. Specifically, the prediction accuracy for the two-class variable of the fifth-minute Apgar score was 97% , and for the first minute was 96% . For the three-class variables, the accuracies were 95% for the fifth minute and 92% for the first minute, respectively. Furthermore, for the eleven-class variables, the prediction accuracy for the fifth-minute Apgar score was 69% and for the first minute was 55%.
Conclusion: These findings underscore the effectiveness of the XGBoost algorithm in accurately predicting low Apgar scores, highlighting its potential to facilitate early intervention and improve neonatal outcomes.
Type of Study:
Research |
Subject:
Midwifery Received: 2026/05/16 | Accepted: 2025/12/22 | Published: 2026/03/21