组织细胞坏死性淋巴结炎患者神经系统损害风险的分类树模型构建

Construction of a classification tree model for neurological damage risk in patients with histiocytic necrotizing lymphadenitis

摘要:

目的 探讨组织细胞坏死性淋巴结炎(HNL)患者神经系统损害的临床特征，并建立分类树风险预测模型。方法 回顾性分析2018年4月至2021年3月在该院确诊为HNL的127例患者的临床资料，依据是否出现神经损害情况将其分为损害组（54例）和无损害组（73例）。分析HNL患者的临床特征、实验室检查结果和神经系统损害的危险因素。应用分类树模型构建HNL患者神经系统损害的风险预测模型，并利用增益图、索引图、错分概率及受试者工作特征曲线(ROC)评价该模型的临床应用价值。结果 两组患者在发热、颅内压增高、淋巴结肿大、外周血（白细胞数和淋巴细胞数）、脑脊液（葡萄糖、白细胞数、蛋白质和氯化物）等比较 ，差异有统计学意义(P<0.05)。多因素Logistic回归分析结果显示，患者颅内压升高、脑脊液葡萄糖值升高、脑脊液白细胞数增多、脑脊液氯化物值升高为HNL患者神经系统损害的危险因素。建立的分类树模型共3层，9个节点，共筛选出4个解释变量，其中脑脊液中氯化物值升高是最重要的影响因素。增益图、索引图及Risk值均显示该分类树模型具有较高的预测价值。ROC曲线下面积(AUC)为0.886，临界值为0.636时预测患者神经损害的敏感度为88.27%，特异度为97.30%。结论 患者颅内压升高、脑脊液葡萄糖值升高、脑脊液白细胞数增多、脑脊液氯化物值升高为HNL患者是否出现神经系统损害的危险因素。 分类树预测模型具有较高的预测效果，可为临床降低HNL患者神经系统损害风险提供依据。

Abstract:

Objective To explore the clinical features of neurological damage in patients with histiocytic necrotizing lymphadenitis （HNL） and establish a classification tree model for risk prediction.Methods We retrospectively analyzed the clinical data of 127 patients diagnosed with HNL in our hospital from April 2018 to March 2021. According to whether neurological damage occurred， the patients were divided into damage group （n=54） and non-damage group （n=73）. The clinical characteristics and laboratory test results were analyzed for risk factors for neurological damage. A classification tree model was used to predict the risk of neurological damage in patients with HNL， and the gain chart， index chart， error probability， and receiver operating characteristic （ROC） curve were used to evaluate the value of this model in clinical application.Results The two groups showed significant differences in the percentages of patients with fever， increased intracranial pressure， and lymphadenectasis， cerebrospinal fluid parameters （white blood cell count and glucose， protein， and chloride levels）， and peripheral blood parameters （white blood cell count and lymphocyte count） （all P<0.05）. Multivariate logistic regression analysis showed that increased intracranial pressure and increased levels of glucose， chloride， and white blood cells in cerebrospinal fluid were risk factors for neurological damage in patients with HNL （P<0.05）. The established classification tree consisted of 3 layers with 9 nodes， and 4 explanatory variables were screened out， with an increased chloride level in cerebrospinal fluid being the most important influencing factor. The gain chart， index chart， and Risk value all showed that the classification tree model had high predictive value. The area under the ROC curve was 0.886， and when the cut-off value was 0.636， the sensitivity and specificity for predicting neurological damage in patients with HNL were 88.27% and 97.30%， respectively.Conclusions Increased intracranial pressure and increased levels of glucose， chloride， and white blood cells in cerebrospinal fluid are risk factors for neurological damage in patients with HNL. The classification tree prediction model has high predictive value， and can provide a basis for reducing the risk of neurological damage in patients with HNL.