Nguyen Rang N, Lam Hue T, Phan Hung V, Bui Nghia Q
Pediatrics, Can Tho University of Medicine and Pharmacy, Can Tho, VNM.
Pediatrics, Bac Lieu General Hospital, Bac Lieu, VNM.
Cureus. 2025 Apr 7;17(4):e81819. doi: 10.7759/cureus.81819. eCollection 2025 Apr.
Background Early identification of pediatric patients at high risk for dengue shock syndrome (DSS) is crucial to enable timely and appropriate clinical interventions. However, the application of machine learning (ML) models for predicting DSS risk remains underexplored. Objective This study aimed to develop and validate a ML-based nomogram for predicting DSS risk in pediatric patients with dengue fever, supporting clinical decision-making. Methods A prospective study was conducted on 230 children with dengue fever admitted to Can Tho Children's Hospital, Vietnam, from January 2020 to December 2022. Clinical and laboratory data were collected and analyzed using R software (version 4.4.1). Six ML algorithms were used to develop risk prediction models for hospitalized pediatric patients with dengue, and their predictive performances were compared. The best-performing model was used to construct a nomogram for DSS prediction. Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC), and the calibration of the nomogram was assessed using a calibration curve. Results Among the 230 dengue patients enrolled, 124 (53.9%) were male, with a median age of 11 years (IQR: 8-13 years). The cohort was randomly divided into a training set (n = 173) and a test set (n = 57). Five key predictors selected for the nomogram were albumin, activated partial thromboplastin time (APTT), fibrinogen, aspartate aminotransferase (AST), and platelet count. In the test set, the AUROC for the six models ranged from 0.888 to 0.945. The random forest model demonstrated the best performance, with an AUROC of 0.945 (95% CI: 0.886-1.000), an accuracy of 0.951 (95% CI: 0.865-0.989), sensitivity of 0.894, specificity of 0.976, and a Kappa score of 0.884. Conclusions ML-based models can be established and potentially help identify hospitalized pediatric patients with dengue who are at high risk of progressing to DSS. The proposed nomogram may be a valuable tool for predicting DSS in clinical practice.
早期识别登革热休克综合征（DSS）高危儿科患者对于及时进行适当的临床干预至关重要。然而，机器学习（ML）模型在预测DSS风险方面的应用仍未得到充分探索。目的：本研究旨在开发并验证一种基于ML的列线图，用于预测登革热患儿发生DSS的风险，以支持临床决策。方法：对2020年1月至2022年12月期间越南芹苴儿童医院收治的230例登革热患儿进行了一项前瞻性研究。使用R软件（版本4.4.1）收集并分析临床和实验室数据。使用六种ML算法为住院登革热患儿开发风险预测模型，并比较它们的预测性能。使用表现最佳的模型构建用于DSS预测的列线图。使用受试者工作特征曲线下面积（AUROC）评估模型性能，并使用校准曲线评估列线图的校准情况。结果：在纳入的230例登革热患者中，124例（53.9%）为男性中位数年龄为11岁（IQR：8-13岁）。该队列被随机分为训练集（n = 173）和测试集（n = 57）。为列线图选择的五个关键预测因素是白蛋白、活化部分凝血活酶时间（APTT）、纤维蛋白原、天冬氨酸氨基转移酶（AST）和血小板计数。在测试集中，六个模型的AUROC范围为0.888至0.945。随机森林模型表现最佳，AUROC为0.945（95%CI：0.886-1.000），准确率为0.951（95%CI：0.865-0.989），敏感性为0.894，特异性为0.976，Kappa评分为0.884。结论：可以建立基于ML的模型，可能有助于识别有进展为DSS高风险的住院登革热患儿。所提出的列线图可能是临床实践中预测DSS的有价值工具。
