切换至 "中华医学电子期刊资源库"

中华实验和临床感染病杂志(电子版) ›› 2026, Vol. 20 ›› Issue (03) : 139 -147. doi: 10.3877/cma.j.issn.1674-1358.2026.03.002

论著

结核性脑膜炎诊断预测模型构建与验证
宋雪云1,2, 张立群1,()   
  1. 1 101100 北京,首都医科大学附属北京胸科医院结核二区
    2 810000 西宁市,青海省第四人民医院呼吸四科
  • 收稿日期:2025-06-04 出版日期:2026-06-15
  • 通信作者: 张立群
  • 基金资助:
    2024年青海省卫生健康委指导性计划课题(2024wjzdx89)

Construction and verification of diagnosis and prediction model for tuberculous meningitis

Xueyun Song1,2, Liqun Zhang1,()   

  1. 1 Tuberculous Zone Ⅱ, Beijing Chest Hospital, Capital Medical University, Beijing 101100, China
    2 Respiratory Department Four, The 4th People’s Hospital of Qinghai Province, Xining 810000, China
  • Received:2025-06-04 Published:2026-06-15
  • Corresponding author: Liqun Zhang
引用本文:

宋雪云, 张立群. 结核性脑膜炎诊断预测模型构建与验证[J/OL]. 中华实验和临床感染病杂志(电子版), 2026, 20(03): 139-147.

Xueyun Song, Liqun Zhang. Construction and verification of diagnosis and prediction model for tuberculous meningitis[J/OL]. Chinese Journal of Experimental and Clinical Infectious Diseases(Electronic Edition), 2026, 20(03): 139-147.

目的

基于青海地区结核性脑膜炎(TBM)特征,探索TBM诊断的独立预测因素,构建并验证其诊断预测模型。

方法

回顾性收集2019年1月至2024年1月青海省第四人民医院诊治的931例感染性脑膜炎患者的临床资料,其中TBM患者739例(TBM组)和非结核感染性脑膜炎患者192例(非TBM组),按7∶3分为训练集(651例,其中TBM患者522例、非TBM患者129例)和验证集(280例,其中TBM患者217例、非TBM患者63例)。通过Lasso回归和多因素Logistic回归筛选TBM诊断的影响因素,构建列线图预测模型,采用受试者操作特征(ROC)曲线评价5种机器学习模型(决策树、逻辑回归、随机森林、支持向量机和极限梯度提升)的诊断效能。

结果

训练集和验证集患者一般资料和生化指标差异均无统计学意义(P均>0.05);TBM患者和非TBM患者组内训练集与验证集基线资料差异亦无统计学意义(P均>0.05)。Lasso降维筛选预测变量和多因素回归分析共筛选出6个TBM的独立预测因素:意识障碍(OR=2.988、95%CI:1.242~7.555、P=0.017)、γ-干扰素释放试验(IGRA)(OR=4.279、95%CI:2.074~9.120、P<0.001)、脑脊液单核细胞比例≥50%(OR=4.061、95%CI:1.881~9.174、P<0.001)、磁共振成像(MRI)提示结核瘤(OR=7.080、95%CI:2.019~34.77、P=0.006)、MRI提示颅底脑膜强化(OR=3.022、95%CI:1.429~6.519、P=0.004)、计算机体层成像(CT)显示肺部病变(血行播散型结核:OR=46.140、95%CI:8.765~856.9、P<0.001)。Logistic回归模型在训练集和验证集的AUC均为0.948(95%CI:0.927~0.969、0.920~0.975),灵敏度分别为87.16%和85.71%,特异度分别为89.92%和90.48%。极限梯度提升模型在诊断TBM中性能最优(AUC=0.954)。

结论

基于临床和影像特征的Logistic回归模型在TBM早期诊断中表现优异。

Objective

To investigate the independent predictive factors for diagnosis of tuberculous meningitis (TBM) based on regional characteristics in Qinghai, China, and to construct and validate a diagnostic prediction model.

Methods

The clinical data of 931 patients with infectious meningitis, including 739 cases with TBM and 192 cases with non-tuberculous infectious meningitis (non-TBM) treated at Qinghai Provincial Fourth People’s Hospital from January 2019 to January 2024 were collected, retrospectively. They were divided into a training set (651 cases, including 522 cases of TBM and 129 cases of non-TBM) and a validation set (280 cases, including 217 cases of TBM and 63 cases of non-TBM) with a ratio of 7∶3. The influencing factors of TBM diagnosis were screened by Lasso regression and multivariate Logistic regression, and a nomogram prediction model was constructed. The diagnostic efficiency of five machine learning models (decision tree, Logistic regression, random forest, support vector machine and extreme gradient boosting) were evaluated and compared by the receiver operating characteristic (ROC) curve.

Results

The general demographic data and biochemical parameters between patients in training set and validation set were without significant differences (all P>0.05); the baseline characteristics between patients in TBM group and non-TBM group across both the training set and validation set were also without significant differences (all P>0.05). By Lasso dimensionality reduction screening of predictive variables and multivariate regression analysis, a total of 6 independent predictors for TBM were identified: consciousness disorder (OR=2.988, 95%CI: 1.242-7.555, P=0.017), gamma interferon release assay (IGRA) (OR=4.279, 95%CI: 2.074-9.120, P<0.001), CSF monocyte ratio≥50% (OR=4.061, 95%CI: 1.88-9.174, P<0.001), magnetic resonance imaging (MRI) suggesting tuberculoma (OR=7.080, 95%CI: 2.019-34.77, P=0.006), MRI suggesting cranial base meningeal enhancement (OR=3.022, 95%CI: 1.429-6.519, P=0.004) and computerized tomography (CT) showing pulmonary lesions (blood disseminated tuberculosis: OR=46.140, 95%CI: 8.765-856.9, P<0.001). AUC of the Logistic regression model in training set and validation set were both 0.948 (95%CI: 0.927-0.969, 0.920-0.975), with sensitivities of 87.16% and 85.71%, and specificities of 89.92% and 90.48%. The Extreme Gradient Boosting (XGB) model performed the best in diagnosing TBM (AUC=0.954).

Conclusions

The Logistic regression model based on clinical and imaging features exhibits excellent diagnostic accuracy for TBM and is suitable for clinical application.

表1 训练集和验证集患者的临床资料
临床资料 验证集(280例) 训练集(651例) 统计量 P
年龄 [MP25P75),岁] 31(20,45) 31(20,47) Z=0.494 0.621
性别 [例(%)] χ2=0.012 0.912
147(52.50) 346(53.15)
133(47.50) 305(46.85)
民族 [例(%)] χ2=0.098 0.754
汉族 43(15.36) 107(16.44)
其他 237(84.64) 544(83.56)
海拔 [例(%)] χ2=0.021 0.884
<2 300米 63(22.50) 142(21.81)
≥2 300米 217(77.50) 509(78.19)
BMI [例(%)] χ2=0.088 0.957
<18.5 15(5.36) 36(5.53)
18.5~23.9 195(69.64) 447(68.66)
>23.9 70(25.00) 168(25.81)
TBM [例(%)] 217(77.50) 522(80.18) χ2=0.706 0.401
肢体无力 [例(%)] 44(15.71) 86(13.21) χ2=0.824 0.364
精神症状 [例(%)] 40(14.29) 111(17.05) χ2=0.907 0.341
听/视力下降 [例(%)] 11(3.93) 21(3.23) χ2=0.291 0.589
癫痫 [例(%)] 29(10.36) 60(9.22) χ2=0.177 0.674
二便障碍 [例(%)] 30(10.71) 68(10.45) χ2=0.001 0.995
意识障碍 [例(%)] 70(25.00) 144(22.12) χ2=0.762 0.383
红细胞沉降率[MP25P75),mm/h] 13.00(6.00,28.25) 14.00(6.00,30.00) Z=0.721 0.471
WBC [MP25P75),×109/L] 6.16(4.85,7.90) 6.20(4.87,8.09) Z=0.549 0.583
贫血 [例(%)] 15(5.36) 51(7.83) χ2=1.467 0.226
NLR [MP25P75)] 4.18(2.55,8.16) 4.39(2.62,8.39) Z=0.729 0.466
血钠[MP25P75),mmol/L] 136.50(133.28,139.20) 136.40(133.00,139.15) Z=0.459 0.646
低蛋白血症 [例(%)] 112(40.00) 281(43.16) χ2=0.804 0.370
CRP [MP25P75),mg/L] 6.86(1.87,28.42) 9.60(1.95,30.99) Z=1.326 0.185
IGRA阳性 [例(%)] 150(53.57) 376(57.76) χ2=1.231 0.267
痰/BALF抗酸染色阳性 [例(%)] 57(20.36) 156(23.96) χ2=1.246 0.264
CSF外观 [例(%)] χ2=0.449 0.503
透明 240(85.71) 545(83.72)
非透明 40(14.29) 106(16.28)
CSF有核细胞数[MP25P75),× 106/L] 19.50(7.00,91.25) 20.00(8.00,85.50) Z=0.522 0.602
CSF单核细胞比例 [例(%)] χ2=0.096 0.757
<50% 193(68.93) 457(70.20)
≥50% 87(31.07) 194(29.80)
CSF氯[MP25P75),mmol/L] 119.95(113.57,124.00) 120.20(113.70,123.60) Z=0.084 0.933
CSF葡萄糖[MP25P75),mmol/L] 2.93(1.82,3.47) 2.98(1.62,3.46) Z=0.160 0.873
CSF LDH [MP25P75),U/L] 23.50(15.00,54.00) 23.00(14.00,50.00) Z=0.202 0.840
CSF ADA [MP25P75),U/L] 1.39(0.36,6.54) 1.15(0.37,5.83) Z=0.407 0.684
CSF蛋白[MP25P75),g/L] 0.26(0.06,1.10) 0.22(0.06,0.96) Z=0.257 0.797
CSF结核分枝杆菌阳性 [例(%)] 16(5.71) 41(6.30) χ2=0.037 0.848
MRI颅脑检测 [例(%)]
结核瘤 77(27.50) 194(29.80) χ2=0.397 0.529
增强前颅底高密度/高信号 218(77.86) 502(77.11) χ2=0.027 0.870
颅底脑膜强化 221(78.93) 533(81.87) χ2=0.920 0.337
肺部CT [例(%)] χ2=0.396 0.530
无病变 87(31.07) 216(33.18)
血行播散型结核 75(26.79) 128(19.66)
浸润或空洞结核 102(36.43) 281(43.16)
神经系统外的结核病变 16(5.71) 26(3.99)
表2 TBM组和非TBM组内训练集与验证集基线资料
基线资料 TBM组 统计量 P 非TBM组 统计量 P
训练集(522例) 验证集(217例) 训练集(129例) 验证集(63例)
年龄 [MP25P75),岁] 31(20,47) 31(20,45) Z=0.238 0.812 31(20,46) 30(19,44) Z=0.407 0.684
男性 [例(%)] 280(53.64) 115(53.00) χ2=0.006 0.937 66(51.16) 32(50.79) χ2=0.000 1.000
海拔≥2 300米 [例(%)] 420(80.46) 175(80.65) χ2=0.000 1.000 89(68.99) 42(66.67) χ2=0.026 0.873
BMI [例(%)] χ2=0.025 0.988 χ2=0.511 0.774
<18.5 30(5.75) 12(5.53) 6(4.65) 3(4.76)
18.5~23.9 350(67.05) 145(66.82) 97(75.19) 50(79.37)
>23.9 142(27.20) 60(27.65) 26(20.16) 10(15.87)
意识障碍 [例(%)] 128(24.52) 62(28.57) χ2=1.113 0.291 16(12.40) 8(12.70) χ2=0.000 1.000
IGRA阳性 [例(%)] 350(67.05) 137(63.13) χ2=0.879 0.348 26(20.16) 13(20.63) χ2=0.000 1.000
CSF单核细胞比例≥50% [例(%)] 180(34.48) 82(37.79) χ2=0.594 0.441 14(10.85) 5(7.94) χ2=0.143 0.705
MRI [例(%)]
结核瘤 190(36.40) 75(34.56) χ2=0.152 0.697 4(3.10) 2(3.17) χ2=0.000 1.000
颅底脑膜强化 490(93.87) 205(94.47) χ2=0.021 0.886 43(33.33) 16(25.40) χ2=0.907 0.341
胸部CT [例(%)] χ2=7.577 0.056 χ2=1.033 0.793
无病变 133(25.48) 49(22.58) 83(64.34) 38(60.32)
血行播散型结核 117(22.41) 68(31.34) 11(8.53) 7(11.11)
浸润或空洞性结核 256(49.04) 91(41.94) 25(19.38) 11(17.46)
其他病变 16(3.07) 9(4.15) 10(7.75) 7(11.11)
图1 训练集TBM诊断模型系数路径图和10倍交叉验证图 注:A:不同λ值下各个变量系数的变化;B:10倍交叉验证中不同λ值下的二项偏差
表3 训练集TBM诊断影响因素的多因素Logistic回归分析
图2 TBM诊断预测模型列线图
图3 TBM预测模型训练集和验证集的ROC曲线 注:A为训练集,B为验证集
图4 五种机器学习模型预测TBM的ROC曲线
表4 五种TBM预测模型的诊断效能
[1]
Wilkinson RJ, Rohlwink U, Misra UK, et al. Tuberculous meningitis[J]. Nat Rev Neurol,2017,13(10):581-598.
[2]
Cresswell VF, Tugume L, Bahr NC, et al. Xpert MTB/RIF Ultra for the diagnosis of HIV-associated tuberculous meningitis: a prospective validation study[J]. Lancet Infect Dis,2020,20(3):308-317.
[3]
Thwaites GE, Chau TTH, Stepniewska K, et al. Diagnosis of adult tuberculous meningitis by use of clinical and laboratory features[J]. Lancet,2002,360(9342):1287-1292.
[4]
Marais S, Thwaites G, Schoeman JF, et al. Tuberculous meningitis: a uniform case definition for use in clinical research[J]. Lancet Infect Dis,2010,10(11):803-812.
[5]
Goenka A, Jeena PM, Mlisana K, et al. Rapid accurate identification of tuberculous meningitis among South African children using a novel clinical decision tool[J]. Pediatr Infect Dis J,2018,37(3):229-234.
[6]
Dendane T, Madani N, Zekraoui A, et al. A simple diagnostic aid for tuberculous meningitis in adults in Morocco by use of clinical and laboratory features[J]. Int J Infect Dis,2013,17(6):e461-e465.
[7]
Chen X, Chen F, Liang C, et al. MRI advances in the imaging diagnosis of tuberculous meningitis: opportunities and innovations[J]. Front Microbiol,2023,14:1308149.
[8]
中华医学会结核病学分会结核性脑膜炎专业委员会. 2019中国中枢神经系统结核病诊疗指南[J]. 中华传染病杂志,2020,38(7):400-408.
[9]
国家卫生和计划生育委员会. 肺结核诊断: WS 288-2017[S/OL]. 北京. 2017[2026-04-29].
[10]
World Health Organization. WHO guidelines on meningitis diagnosis, treatment and care[EB/OL]. Geneva: World Health Organization,2025[2026-04-29].
[11]
辛明明, 马序竹. 结核性脑膜炎临床诊治与研究进展[J/OL]. 中华实验和临床感染病杂志(电子版),2025,19(5):257-264.
[12]
Niu M, Bai Z, Dong L, et al. A novel diagnostic prediction model for distinguishing between tuberculous and cryptococcal meningitis[J]. Clin Med Res,2024,22(4):197-205.
[13]
Donovan J, Figaji A, Imran D, et al. The neurocritical care of tuberculous meningitis[J]. Lancet Neurol,2019,18(8):771-783.
[14]
Chalasani R, Goonathilake MR, Waqaret S, et al. The outcome of surgical intervention (ventriculoperitoneal shunt and endoscopic third ventriculostomy) in patients with hydrocephalus secondary to tuberculous meningitis: a systematic review[J]. Cureus,2022,14(5):e25317.
[15]
Wen A, Leng EL, Liu SM, et al. Diagnostic accuracy of interferon-gamma release assays for tuberculous meningitis: a systematic review and meta-analysis[J]. Front Cell Infect Microbiol,2022,12:788692.
[16]
边赛男, 刘晓清. γ-干扰素释放试验在免疫功能抑制人群中诊断结核分枝杆菌感染的应用[J/OL]. 中华实验和临床感染病杂志(电子版),2017,11(2):117-120.
[17]
陈艳, 侍效春, 刘晓清. 结核性脑膜炎并发矛盾反应一例[J/OL]. 中华实验和临床感染病杂志(电子版),2024,18(1):49-53.
[18]
Koesbandono, Muljadi R, Sutanto R, et al. Intracranial tuberculomas: review of MRI findings and clinical features[J]. Clin Radiol,2024,79(5):354-362.
[19]
Dahal P, Parajuli S. Magnetic resonance imaging findings in central nervous system tuberculosis: a pictorial review[J]. Heliyon,2024,10(8):e29779.
[20]
Bhalla A S, Goyal A, Guleria R, et al. Chest tuberculosis: radiological review and imaging recommendations[J]. Indian J Radiol Imaging,2015,25(3):213-225.
[21]
牛佩璇, 阿尔泰, 杜彩云, 等. 722例结核性脑膜炎患者流行病学及临床特征分析[J]. 临床肺科杂志,2024,29(10):1555-1559.
[22]
Moons KGM, Altman DG, Reitsma JB, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): explanation and elaboration[J]. Ann Intern Med,2015,162(1):W1-W73.
[1] 曾维波, 何燕妮, 熊燃, 王珍, 许智莉, 刘红梅. 两广地区颈动脉狭窄及广东地区肾动脉狭窄与血管超声危急值管理现状调研[J/OL]. 中华医学超声杂志(电子版), 2026, 23(05): 355-359.
[2] 陈曼, 郑雯, 常才, 俞清, 肖沪生, 赵宝珍, 施仲伟, 陈悦, 李丽蟾, 王文平. 上海市超声诊断设备配置现状调查与质量控制管理策略优化[J/OL]. 中华医学超声杂志(电子版), 2026, 23(05): 369-373.
[3] 王子怡, 谢苗苗, 薄舟洋, 孙园园, 武莉娜, 王睿丽. 产后子宫坏死的超声造影表现及诊断价值[J/OL]. 中华医学超声杂志(电子版), 2026, 23(05): 400-407.
[4] 朱雨捷, 史蒙蒙, 贾春澍, 曹也, 蔡光伟. 基因组测序技术在产前诊断中的应用[J/OL]. 中华妇幼临床医学杂志(电子版), 2026, 22(04): 283-289.
[5] 芦铭, 刘波, 钱真, 王瑶, 张艾捷, 王充, 李斌, 栗鹏程, 曹建华, 吕柏蓉, 李孟然, 单祎宁, 金彤雯, 王广志. 基于机器学习与计算机视觉技术的皮瓣移植术后血运监测模型的构建及验证[J/OL]. 中华损伤与修复杂志(电子版), 2026, 21(04): 259-265.
[6] 鲁峰刚, 刘宇, 张毅, 王秦玉. 新辅助放化疗直肠癌根治术患者术后吻合口漏影响因素分析及风险预测模型构建[J/OL]. 中华普外科手术学杂志(电子版), 2026, 20(04): 347-351.
[7] 袁强, 范慈勃, 韩丽丽, 陈光, 陈纲, 赵锁. 融合病理图像与报告的多模态模型用于乳腺癌预后预测[J/OL]. 中华普外科手术学杂志(电子版), 2026, 20(04): 374-378.
[8] 王琪璠, 王宝, 冯梅晶, 边峰. 多层螺旋计算机断层扫描和彩色高频超声对开放式腹股沟疝无张力修补术后患者早期并发症的诊断价值[J/OL]. 中华疝和腹壁外科杂志(电子版), 2026, 20(04): 423-427.
[9] 中华人民共和国国家卫生健康委员会医政司. 原发性肝癌诊疗指南(2026年版)[J/OL]. 中华肝脏外科手术学电子杂志, 2026, 15(04): 443-507.
[10] 牛海鹏, 王石. 脾脏肿瘤诊疗策略[J/OL]. 中华肝脏外科手术学电子杂志, 2026, 15(04): 529-536.
[11] 张宇, 郑卫华, 张俊晶. 门静脉高压症的诊治及并发症研究现状与展望[J/OL]. 中华肝脏外科手术学电子杂志, 2026, 15(04): 537-544.
[12] 孙慧, 雷蕊艳, 张丽, 周婷. 腹部超声联合血清激活素A、髓系细胞触发受体1检测对新生儿坏死性小肠结肠炎的诊断价值[J/OL]. 中华消化病与影像杂志(电子版), 2026, 16(04): 331-336.
[13] 桂红, 马博, 王煜, 刘晓晖. Alvarado评分、超声特征联合血清CXCR3、RANTES水平在儿童阑尾炎诊断及分型中的临床价值[J/OL]. 中华消化病与影像杂志(电子版), 2026, 16(04): 341-346.
[14] 王景明, 王磊, 许小多, 陈欣, 黄伟敏. 肌少-骨质疏松症的临床诊疗进展[J/OL]. 中华临床医师杂志(电子版), 2026, 20(05): 386-392.
[15] 齐洪武, 徐泽雨. 脑脓肿的临床诊治进展[J/OL]. 中华临床医师杂志(电子版), 2026, 20(04): 327-331.
阅读次数
全文


摘要


AI


AI小编
你好!我是《中华医学电子期刊资源库》AI小编,有什么可以帮您的吗?