预测模型鉴别川崎病与急性颈部淋巴结炎
本研究基于以发热和颈部淋巴结肿大首诊的儿科住院患者数据,构建决策树模型鉴别川崎病淋巴结首发与急性颈部淋巴结炎。模型纳入脓肿、C反应蛋白百分比变化和中性粒细胞计数百分比变化三个因素,在验证队列中表现出较高的敏感性和特异性,但需多中心外部验证后方可临床应用。
要点
- 决策树模型纳入颈部CT或超声定义的脓肿、CRP百分比变化、中性粒细胞计数百分比变化。
- 验证队列中模型敏感性89.5%、特异性88.9%、总体准确率89.3%。
- 模型预测能力优于已知实验室标志物,但尚需多中心外部验证。
核心结论
- 决策树模型可提高鉴别川崎病淋巴结首发与急性颈部淋巴结炎的准确性。
- 临床应用前需基于多中心数据进行外部验证。
研究设计与样本
- 数据来自以发热和颈部淋巴结肿大首诊的儿科住院患者。
- 训练队列206例,验证队列103例。
- 通过组间比较或单因素 logistic 回归分析识别鉴别因素,并采用决策树分析构建模型。
主要发现
- 决策树模型包含3个决定因素:颈部CT或超声定义的脓肿、C反应蛋白水平百分比变化、中性粒细胞计数百分比变化。
- 验证队列中模型敏感性89.5%、特异性88.9%、阳性预测值95.8%、阴性预测值75.0%、总体准确率89.3%、约登指数0.784。
- 模型预测能力优于先前已知的实验室标志物。
关键数值与推荐
- 决策树模型决定因素:颈部CT或超声定义的脓肿、C反应蛋白水平百分比变化、中性粒细胞计数百分比变化。
- 验证队列敏感性89.5%、特异性88.9%、阳性预测值95.8%、阴性预测值75.0%、总体准确率89.3%、约登指数0.784。
临床怎么用
- 对于以发热和颈部淋巴结肿大首诊的患儿,可考虑使用决策树模型辅助鉴别川崎病淋巴结首发与急性颈部淋巴结炎。
- 模型涉及颈部影像学(CT或超声)评估脓肿及动态监测CRP和中性粒细胞计数变化。
- 目前模型尚未经外部验证,临床应用需谨慎。
局限性
- 需要基于多中心数据的外部验证后方可临床应用。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
OBJECTIVES: To distinguish early-stage lymph node first presentation of Kawasaki disease from acute cervical lymphadenitis, we developed an algorithm using sequential laboratory marker levels and radiologic findings. STUDY DESIGN: Data were obtained from pediatric inpatients initially presenting with fever and cervical lymphadenopathy. Discriminative factors for the differential diagnosis of acute cervical lymphadenitis and lymph node first presentation of Kawasaki disease were identified from intergroup comparison or univariate logistic regression analysis. A model for differentiating between lymph node first presentation of Kawasaki disease and acute cervical lymphadenitis was constructed using decision-tree analysis. RESULTS: Patients were divided into 2 cohorts: training (206 patients) and validation (103 patients) cohorts. A decision-tree model developed from the data of the training cohort included 3 determinants: neck computed tomography- or ultrasonography-defined abscess, percentage change in C-reactive protein level, and percentage change in neutrophil count. The prediction power of our decision-tree model for the validation cohort was superior to that of previously known laboratory markers (sensitivity of 89.5%, specificity of 88.9%, positive predictive value of 95.8%, negative predictive value of 75.0%, overall accuracy of 89.3%, and a Youden index of 0.784). CONCLUSIONS: A decision-tree model could differentiate lymph node first presentation of Kawasaki disease from acute cervical lymphadenitis with an increased accuracy. External validation based on multicenter data is needed before clinical application.
引用
- Kim JM, Kim J. Prediction Model for the Differential Diagnosis of Kawasaki Disease and Acute Cervical Lymphadenitis in Patients Initially Presenting with Fever and Cervical Lymphadenitis. J Pediatr. 2020. PMID: 32450069. doi:10.1016/j.jpeds.2020.05.031
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。