脑脊液特征指标辅助结核性脑膜炎快速筛查的效果评估

Evaluation of the effectiveness of cerebrospinal fluid characteristics indicators in assisting rapid screening of tuberculous meningitis

目的:探索利用若干脑脊液特征指标辅助提高敏感度,实现结核性脑膜炎(tuberculous meningitis,TBM)的快速筛查和早期治疗控制。方法:2011—2019年期间,依据纳入排除标准,从江苏省苏州市、四川省自贡市和贵州省贵阳市的结核病定点医疗机构收治的383例疑似TBM患者中,随机筛选100例临床诊断TBM患者,并按年龄、性别1∶1配对抽取100例非TBM患者,收集脑脊液白细胞计数、乳酸盐、蛋白总量、腺苷脱氨酶、葡萄糖等生化指标,并采用超高通量液相色谱法分离患者脑脊液代谢物,通过偏正交最小二乘判别法分析(OPLS-DA)筛选在不同类型患者中分布具有差异的生化和代谢物指标,通过机器学习构建决策树模型(boosted-CART),评估特征指标对于快速筛查各类型TBM患者的效果。结果:以P<0.05、OPLS-DA模型变量权重值>1和代谢物组间表达量差异倍数>1为标准,利用boosted-CART模型筛选脑脊液L-谷氨酰胺和葡萄糖水平作为特征指标,其cut-off值分别为607.06mmol/L(L-谷氨酰胺)和60.03mg/dl(葡萄糖)。上述两项指标联用,可将检出很可能和可能TBM患者的敏感度从66.7%(95%CI:24.1%~94.0%)提升至83.3%(95%CI:36.5%~99.1%),AUC值从0.667(95%CI:0.444~0.889)提升至0.708(95%CI:0.494~0.923),但效果不明显(Z=0.261,P>0.05)。结论:脑脊液L-谷氨酰胺和葡萄糖指标可提高TBM快速筛查的敏感度,利于TBM早诊早治和控制结核病传播。

Objective:To explore the apllication of several cerebrospinal fluid(CSF)indicators to improve the detection sensitivity and achieve rapid screening and early treatment control of tuberculous meningitis(TBM).Methods:From 2011 to 2019,according to the inclusion and exclusion criteria,100 TBM patients were randomly selected from 383 suspected TBM patients admitted to tuberculosis designated medical institutions in Suzhou of Jiangsu Province,Zigong of Sichuan Province and Guiyang of Guizhou Province.And 100 non-TBM patients were selected based on age and sex matched of 1∶1.The biochemical indicators such as white blood cell count,lactate,total protein,adenosine deaminase,and glucose in CSF were collected.Ultra high-throughput liquid chromatography was used to separate the metabolites of CSF.The metabolic indicators with statistically significant differences in distribution were screened by Orthogonal Projections to Latent Structures-Discriminant Analysis(OPLS-DA).A classification and regression tree(boosted-CART)model was constructed by machine learning.The effect of characteristic indicators on rapid screening of TBM patients with various types was evaluated.Results:Using the criteria of P<0.05,OPLS-DA variable important in projection value>1 and absolute value of Forld Change>1,the above model was combined to screen CSF L-glutamine and glucose levels as characteristic indicators.The cutoff values were 607.06 mmol/L(L-glutamine)and 60.03 mg/dl(glucose),respectively.The combination of the above two indicators improved the sensitivity of detecting probable and possible TBM from 66.7%(95%CI:24.1%-94.0%)to 83.3%(95%CI:36.5%-99.1%),AUC increased from 0.667(95%CI:0.444-0.889)to 0.708(95%CI:0.494-0.923)(Z=0.261,P>0.05).Conclusion:Biochemical and metabolic indicators such as CSF L-glutamine and glucose can improve the sensitivity of rapid TBM screening,which is conducive to the early diagnosis and treatment of TBM and to control the spread of tuberculosis.