Koshu Ryota, Noda Masao, Nakamoto Haruna, Fukuhara Takahiro, Ito Makoto
Department of Otolaryngology and Head and Neck Surgery, Jichi Medical University, Shimotsuke, Japan.
Eur Arch Otorhinolaryngol. 2025 Jun 3. doi: 10.1007/s00405-025-09505-7.
Paediatric cervical abscesses necessitate careful assessment to determine appropriate treatment strategies. Some patients require surgical intervention, although conservative management is effective. However, the criteria for the surgical indications remain unclear. Machine learning models have demonstrated promise in improving diagnostic accuracy across different medical fields.
This study aimed to assess the use of machine learning models in predicting the requirement for surgical intervention in paediatric cervical abscesses and compare their performance with that of traditional logistic regression.
A retrospective analysis was conducted on 55 paediatric patients diagnosed with cervical abscesses between 2010 and 2024. The patient demographics, clinical findings, laboratory data, and imaging characteristics were examined. Six predictive models were developed: logistic regression, Random Forest, Lasso regression, Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine. Model performance was evaluated using the area under the curve (AUC), accuracy, precision, recall, and F1-score. Feature importance was examined to identify the main predictive factors.
Among all the factors, abscess size was the most significant predictor of surgical intervention. Machine-learning models, especially XGBoost, outperformed logistic regression, achieving the highest AUC, accuracy, and recall. Inflammatory markers, including neutrophil-to-lymphocyte ratio and neutrophil count, also substantially contributed to the prediction accuracy.
Machine learning models, particularly XGBoost, provide superior predictive performance compared with logistic regression, providing a valuable tool for optimising treatment decisions in paediatric cervical abscesses. These models improve clinical decision-making by integrating multiple factors, decreasing unnecessary surgeries, and enhancing patient outcomes.
小儿颈部脓肿需要仔细评估以确定合适的治疗策略。尽管保守治疗有效，但有些患者仍需要手术干预。然而，手术指征的标准仍不明确。机器学习模型已在不同医学领域提高诊断准确性方面展现出前景。
本研究旨在评估机器学习模型在预测小儿颈部脓肿手术干预需求方面的应用，并将其性能与传统逻辑回归进行比较。
对2010年至2024年间诊断为颈部脓肿的55例儿科患者进行回顾性分析。检查了患者的人口统计学、临床表现、实验室数据和影像学特征。开发了六个预测模型：逻辑回归、随机森林、套索回归、支持向量机（SVM）、极端梯度提升（XGBoost）和轻量级梯度提升机。使用曲线下面积（AUC）、准确性、精确性、召回率和F1分数评估模型性能。检查特征重要性以确定主要预测因素。
在所有因素中，脓肿大小是手术干预最显著的预测因素。机器学习模型，尤其是XGBoost，优于逻辑回归，实现了最高的AUC、准确性和召回率。炎症标志物，包括中性粒细胞与淋巴细胞比率和中性粒细胞计数，也对预测准确性有很大贡献。
与逻辑回归相比，机器学习模型，特别是XGBoost，具有卓越的预测性能，为优化小儿颈部脓肿的治疗决策提供了有价值的工具。这些模型通过整合多种因素改善临床决策，减少不必要的手术，并提高患者预后。
