基于随机森林模型算法预测继发性肺结核的模型构建与验证

doi:10.19982/j.issn.1000-6621.20250166

摘要/Abstract

摘要：

目的: 探讨随机森林模型算法对继发性肺结核的筛查价值,以期为临床早期识别继发性肺结核提供依据。方法: 收集2021年3—9月期间在黑龙江省医院健康管理中心接受体检的健康人群共1208名;以及在黑龙江省传染病院首次确诊并接受治疗的继发性肺结核患者共876例,依据纳入和排除标准,367例继发性肺结核患者纳入观察组,376名体检人群纳入对照组。收集研究对象的基本情况,按照2∶1的比例,将数据分为训练集(495例)和测试集(248例)。模型共纳入38个预测变量,使用训练集数据进行随机森林模型构建,利用测试集数据进行模型验证和评价。对模型中重要性排名前十位的变量进行两组之间的比较分析。结果: 本研究随机森林模型的最优节点数为5,决策树数目为300。模型的准确率为99.60%、精确率为99.92%、敏感度为99.20%、特异度为99.87%,其受试者工作特征(receiver operating characteristic, ROC)曲线下面积(area under curve, AUC)为0.986(95%CI:0.978~0.995)。通过基尼指数(Gini 值)平均降低量筛选出排名前十位的变量,包括血小板分布宽度(30.02)、白球比(20.70)、间接胆红素(19.32)、白蛋白(17.97)、平均血红蛋白浓度(12.24)、尿比重(11.26)、总胆红素(10.09)、总胆汁酸(7.43)、淋巴细胞百分比(6.92)、谷草转氨酶/谷丙转氨酶比值(6.50)。观察组的血小板分布宽度[15.90(15.50,16.20)fl]、谷草转氨酶/谷丙转氨酶比值[1.41(1.02,1.79)]、尿比重分布[正常27例(7.36%)、异常340例(92.64%)]均高于对照组[分别为12.00(11.10,13.30)fl,0.95(0.77,1.15),正常86例(22.87%)、异常290例(77.13%)],差异均有统计学意义(Z=-16.907,P<0.001;Z=-11.951,P<0.001;χ²=34.670,P<0.001)。观察组的白蛋白[(38.42±6.47)g/L]、总胆红素[12.80(9.10,19.00)μmol/L]、间接胆红素[7.60(5.05,11.80)μmol/L]、白球比[1.17(0.97,1.40)]、总胆汁酸[2.80(1.79,5.12)μmol/L]、淋巴细胞百分比[22.90(15.50,32.55)%]、平均血红蛋白浓度[322.00(317.00,328.00)g/L]均低于对照组[分别为(45.14±2.13)g/L,22.25(18.40,26.90)μmol/L,17.40(14.30,19.10)μmol/L,1.60(1.50,1.70),4.60(3.70,5.50)μmol/L,34.60(29.07,39.90)%,336.00(331.00,343.00)g/L],差异均有统计学意义(t=-18.891,P<0.001;Z=-14.313,P<0.001;Z=-16.994,P<0.001;Z=-17.030,P<0.001;Z=-9.675,P<0.001;Z=-12.684,P<0.001;Z=-16.843,P<0.001)。结论: 本研究构建的继发性肺结核患者的随机森林预测模型性能较好,建议进行早期筛查肺结核患者时关注一些营养指标(如间接胆红素、总胆红素、白蛋白等)变化以优化筛查策略,同时在治疗继发性肺结核患者时,注重营养治疗。

关键词: 结核,肺, 预测, 模型, 统计学

Abstract:

Objective: To investigate the screening value of Random Forest model algorithm for secondary pulmonary tuberculosis, aiming to provide a basis for early clinical identification of secondary pulmonary tuberculosis. Methods: A total of 1208 healthy individuals who underwent physical examination at the Health Management Center of Heilongjiang Provincial Hospital from March to September 2021, and 876 patients with secondary pulmonary tuberculosis who were initially diagnosed and received treatment at Heilongjiang Provincial Infectious Disease Hospital were identified. Based on inclusion and exclusion criteria, 367 secondary pulmonary tuberculosis patients were assigned to the observation group and 376 healthy individuals to the control group. Basic demographic data was collected. The dataset was divided into a training set (495 cases) and a testing set (248 cases) with a 2∶1 ratio. The model incorporated 38 predictive variables. A Random Forest model was constructed using the training set data, and its performance was validated and evaluated using the testing set data. A comparative analysis of the top ten important variables identified by the model was performed for the two groups. Results: In this study, the optimal parameters for the Random Forest model were determined to be 5 nodes and 300 trees. The model demonstrated high performance with an accuracy of 99.60%, precision of 99.92%, sensitivity of 99.20%, specificity of 99.87%, and an area under the receiver operating characteristic curve (AUC) of 0.986 (95%CI: 0.978-0.995). The top ten variables ranked by mean decrease in Gini index were identified as: platelet distribution width (30.02), albumin-to-globulin ratio (20.70), indirect bilirubin (19.32), albumin (17.97), mean corpuscular hemoglobin concentration (12.24), urine specific gravity (11.26), total bilirubin (10.09), total bile acids (7.43), lymphocyte percentage (6.92), and aspartate aminotransferase/alanine aminotransferase ratio (6.50).Compared with the control group, the observation group exhibited significantly higher platelet distribution width (15.90 (15.50, 16.20) fl), AST/ALT ratio (1.41 (1.02, 1.79)), and a higher proportion of abnormal urine specific gravity (340 cases (92.64%) vs. normal: 27 cases (7.36%)) versus the control group (12.00 (11.10, 13.30) fl; 0.95 (0.77, 1.15); abnormal: 290 cases (77.13%) vs. normal: 86 cases (22.87%)). These differences were statistically significant (Z=-16.907, P<0.001; Z=-11.951, P<0.001; χ²=34.670, P<0.001, respectively). Conversely, the observation group had significantly lower levels of albumin ((38.42±6.47) g/L), total bilirubin (12.80 (9.10, 19.00) μmol/L), indirect bilirubin (7.60 (5.05, 11.80) μmol/L), albumin-to-globulin ratio (1.17 (0.97, 1.40)), total bile acids (2.80 (1.79, 5.12) μmol/L), lymphocyte percentage (22.90 (15.50, 32.55) %), and mean corpuscular hemoglobin concentration (322.00 (317.00, 328.00) g/L) compared with the control group ((45.14±2.13) g/L; 22.25 (18.40, 26.90) μmol/L; 17.40 (14.30, 19.10) μmol/L; 1.60 (1.50, 1.70); 4.60 (3.70, 5.50) μmol/L; 34.60 (29.07, 39.90) %; 336.00 (331.00, 343.00) g/L). These differences were also statistically significant (t=-18.891, P<0.001; Z=-14.313, P<0.001; Z=-16.994, P<0.001; Z=-17.030, P<0.001; Z=-9.675, P<0.001; Z=-12.684, P<0.001; Z=-16.843, P<0.001, respectively). Conclusion: The Random Forest prediction model constructed in this study for secondary pulmonary tuberculosis patients exhibited excellent performance. It is recommended that early screening for tuberculosis patients should pay attention to changes in certain nutritional indicators (such as indirect bilirubin, total bilirubin, albumin, etc.) to optimize screening strategy. Furthermore, nutritional therapy should be emphasized during the treatment of secondary pulmonary tuberculosis patients.

Key words: Tuberculosis,pulmonary, Forecasting, Models, statistical

中图分类号:

褚光炎, 李婷, 于嘉宁, 何迪, 张堃, 侯绍英, 闫世春. 基于随机森林模型算法预测继发性肺结核的模型构建与验证[J]. 中国防痨杂志, 2025, 47(10): 1311-1317. doi: 10.19982/j.issn.1000-6621.20250166

Chu Guangyan, Li Ting, Yu Jianing, He Di, Zhang Kun, Hou Shaoying, Yan Shichun. Building and evaluating a predictive model for secondary pulmonary tuberculosis based on Random Forest model[J]. Chinese Journal of Antituberculosis, 2025, 47(10): 1311-1317. doi: 10.19982/j.issn.1000-6621.20250166

图/表 4

表1

测试集与训练集研究对象的基线信息

指标	训练集[495例(名)]				测试集[248例(名)]
指标	观察组 (244例)	对照组 (251名)	统计检验值	P值	观察组 (123例)	对照组 (125名)	统计检验值	P值
性别[例(构成比,%)]			χ²=126.044	<0.001			χ²=3.843	0.950
男性	78(31.97)	203(80.88)			53(43.19)	98(78.40)
女性	166(68.03)	48(19.12)			70(56.91)	27(21.60)
年龄[岁,M(Q₁, Q₃)]	54.00 (44.00,65.00)	52.00 (49.00,57.00)	Z=-1.027	0.304	51.00 (39.00,66.00)	52.00 (50.00,58.00)	Z=-1.264	0.208
谷草转氨酶(IU/L, $x ˉ$ ±s)	29.27±44.78	19.90±3.82	t=3.297	0.001	25.68±24.84	20.25±3.76	t=2.341	0.021
谷丙转氨酶(IU/L, $x ˉ$ ±s)	26.23±72.47	22.28±7.80	t=0.858	0.392	24.16±40.62	21.53±7.27	t=0.693	0.493
直接胆红素(μmol/L, $x ˉ$ ±s)	6.19±7.27	5.84±2.57	t=0.743	0.458	5.95±6.50	5.78±2.23	t=0.255	0.801
总蛋白(g/L, $x ˉ$ ±s)	72.38±8.26	73.98±3.44	t=-2.855	0.005	71.99±8.89	74.46±3.58	t=-2.774	0.006
白蛋白(g/L, $x ˉ$ ±s)	38.54±6.54	45.15±2.21	t=-15.201	<0.001	38.18±6.34	45.10±1.92	t=-11.243	<0.001
球蛋白(g/L, $x ˉ$ ±s)	33.72±6.07	28.83±2.74	t=11.671	<0.001	33.98±8.63	29.36±2.91	t=5.456	<0.001
谷氨酰转肽酶(IU/L, $x ˉ$ ±s)	60.11±84.74	23.62±17.17	t=6.682	<0.001	62.88±115.68	20.76±11.76	t=3.921	<0.001
碱性磷酸酶(IU/L, $x ˉ$ ±s)	100.70±55.26	67.87±18.46	t=8.942	<0.001	96.84±45.28	66.92±18.41	t=6.574	<0.001
甘油三酯(mmol/L, $x ˉ$ ±s)	1.19±0.59	1.21±0.57	t=-0.485	0.628	1.20±0.67	1.18±0.53	t=0.235	0.820
高密度脂蛋白(mmol/L, $x ˉ$ ±s)	1.15±0.39	1.37±0.29	t=-7.038	<0.001	1.10±0.33	1.35±0.27	t=-6.165	<0.001
总胆固醇(mmol/L, $x ˉ$ ±s)	4.96±4.40	5.26±0.83	t=-1.098	0.273	4.51±1.21	5.21±0.81	t=-5.174	<0.001
低密度脂蛋白(mmol/L, $x ˉ$ ±s)	2.42±0.78	3.18±0.69	t=-11.813	<0.001	2.30±0.76	3.17±0.69	t=-9.001	<0.001
空腹血糖(mmol/L, $x ˉ$ ±s)	5.31±1.15	5.49±0.52	t=-2.164	0.031	5.28±0.91	5.47±0.52	t=-1.874	0.063
同型半胱氨酸(μmol/L, $x ˉ$ ±s)	14.09±10.18	10.60±3.97	t=5.078	<0.001	13.58±8.32	10.53±3.29	t=3.663	<0.001
尿素氮(mmol/L, $x ˉ$ ±s)	5.02±2.29	4.81±1.15	t=1.308	0.192	4.84±2.42	4.84±1.14	t=-0.012	0.996
肌酐(μmol/L, $x ˉ$ ±s)	55.85±25.64	63.14±12.87	t=-4.049	<0.001	55.31±25.59	63.12±13.26	t=-2.897	0.004
尿酸(μmol/L, $x ˉ$ ±s)	314.34±133.31	295.09±67.39	t=2.053	0.041	303.06±142.48	292.30±70.74	t=0.719	0.470
二氧化碳(mmol/L, $x ˉ$ ±s)	24.93±2.79	25.43±1.87	t=-2.394	0.017	25.01±2.89	25.08±1.89	t=-0.207	0.830
淋巴细胞计数(×10⁹/L, $x ˉ$ ±s)	1.73±0.78	1.96±0.52	t=-3.950	<0.001	1.82±0.86	2.01±0.61	t=-1.923	0.057
平均血红蛋白量(pg, $x ˉ$ ±s)	29.36±2.92	30.73±1.68	t=-6.501	<0.001	29.14±2.52	30.60±1.93	t=-4.895	<0.001
红细胞平均体积(fl, $x ˉ$ ±s)	91.21±7.07	91.48±4.36	t=-0.525	0.600	90.51±6.21	90.53±5.17	t=-0.021	0.987
红细胞分布宽度(%, $x ˉ$ ±s)	45.90±4.09	42.60±2.85	t=10.620	<0.001	46.07±6.14	41.79±2.70	t=6.842	<.001
单核细胞百分比(%, $x ˉ$ ±s)	7.23±2.43	6.45±1.50	t=4.367	<0.001	6.95±2.02	6.62±1.45	t=1.394	0.165
嗜酸性粒细胞计数(×10⁹/L, $x ˉ$ ±s)	0.11±0.13	0.12±0.10	t=-0.908	0.364	0.14±0.14	0.11±0.08	t=2.295	0.022
嗜碱性粒细胞计数(×10⁹/L, $x ˉ$ ±s)	0.03±0.02	0.03±0.02	t=-0.209	0.834	0.03±0.02	0.03±0.01	t=1.194	0.234
嗜酸性粒细胞百分比(%, $x ˉ$ ±s)	1.52±1.49	2.04±1.73	t=-3.719	<0.001	1.98±2.03	1.90±1.34	t=0.355	0.730
单核细胞计数(×10⁹/L, $x ˉ$ ±s)	0.56±0.27	0.37±0.11	t=10.451	<0.001	0.57±0.27	0.38±0.10	t=7.345	<0.001
血小板压积(%, $x ˉ$ ±s)	0.29±0.10	0.26±0.05	t=4.476	<0.001	0.30±0.08	0.25±0.05	t=4.864	<0.001
红细胞压积(%, $x ˉ$ ±s)	0.41±0.06	0.41±0.03	t=-0.053	0.958	0.41±0.05	0.41±0.03	t=0.793	0.429
中性粒细胞百分比(%, $x ˉ$ ±s)	66.51±12.08	56.47±8.22	t=10.990	<0.001	65.86±12.97	56.01±7.29	t=7.082	<0.001
大型血小板比率(%, $x ˉ$ ±s)	24.12±8.52	27.61±7.08	t=-5.051	<0.001	24.09±8.48	28.23±6.78	t=-4.044	<0.001
白细胞(×10⁹/L, $x ˉ$ ±s)	7.88±3.12	5.76±1.18	t=10.080	<0.001	8.47±4.54	5.73±1.03	t=6.351	<0.001
红细胞(×10¹²/L, $x ˉ$ ±s)	4.52±0.63	4.51±0.39	t=0.147	0.884	4.56±0.56	4.50±0.37	t=1.001	0.316
血红蛋白(g/L, $x ˉ$ ±s)	132.25±21.48	138.52±12.53	t=-4.022	<0.001	132.84±18.57	137.37±11.21	t=-2.234	0.027
中性粒细胞计数(×10⁹/L, $x ˉ$ ±s)	5.45±2.98	3.28±0.96	t=10.973	<0.001	5.85±4.45	3.20±0.69	t=6.353	<0.001
红细胞分布宽度(%, $x ˉ$ ±s)	13.60±1.59	12.70±1.00	t=7.666	<0.001	13.75±1.95	12.63±1.17	t=5.274	<0.001
平均血小板体积(fl, $x ˉ$ ±s)	9.90±1.27	10.36±0.89	t=-4.783	<0.001	10.41±6.50	10.44±0.85	t=-0.059	0.952
尿液酸碱度(pH, $x ˉ$ ±s)	6.12±0.62	6.18±0.55	t=-1.241	0.215	6.02±0.80	6.26±0.51	t=-2.684	0.008
潜血( $x ˉ$ ±s)	0.19±0.55	0.64±1.06	t=-6.205	<0.001	0.29±0.75	0.52±0.96	t=-1.964	0.051
尿比重[例(构成比,%)]			χ²=23.062	<0.001			χ²=0.013	0.909
正常	19(7.79)	53(21.12)			20(16.26)	21(16.80)
异常	225(92.21)	198(78.88)			103(83.74)	104(83.20)
血小板分布宽度[fl,M(Q₁, Q₃)]	15.90 (15.50,16.10)	12.00 (11.03,13.30)	Z=-14.287	<0.001	16.00 (15.60,16.20)	11.95 (11.10,13.40)	Z=-8.949	<0.001
白球比[M(Q₁, Q₃)]	1.19 (0.96,1.39)	1.60 (1.50,1.70)	Z=-14.519	<0.001	1.17 (0.97,1.42)	1.60 (1.40,1.70)	Z=-8.867	<0.001
间接胆红素[μmol/L,M(Q₁, Q₃)]	7.75 (5.00,12.70)	17.50 (14.43,19.30)	Z=-13.438	<0.001	7.10 (5.30,10.50)	16.80 (14.12,18.60)	Z=-10.434	<0.001
平均血红蛋白浓度[g/L, M(Q₁, Q₃)]	322.00 (317.00,328.00)	336.00 (331.00,342.00)	Z=-13.527	<0.001	322.00 (317.00,327.00)	338.00 (331.25,344.00)	Z=-9.921	<0.001
总胆红素[μmol/L,M(Q₁, Q₃)]	13.10 (9.00,19.80)	22.35 (18.60,27.18)	Z=-11.130	<0.001	12.00 (9.30,16.50)	22.00 (18.15,26.10)	Z=-9.066	<0.001
总胆汁酸[μmol/L,M(Q₁, Q₃)]	2.88 (1.87,5.20)	4.60 (3.60,5.50)	Z=-7.329	<0.001	2.67 (1.60,5.05)	5.00 (4.00,5.68)	Z=-6.434	<0.001
淋巴细胞百分比[%,M(Q₁, Q₃)]	22.35 (15.30,32.88)	34.45 (28.85,39.78)	Z=-10.315	<0.001	23.10 (16.20,31.20)	35.20 (29.22,39.90)	Z=-7.381	<0.001
谷草转氨酶/谷丙转氨酶比值 [M(Q₁, Q₃)]	1.44 (1.02,1.81)	0.94 (0.77,1.13)	Z=-10.548	<0.001	1.40 (0.98,1.74)	0.95 (0.79,1.20)	Z=-5.572	<0.001

表1

表2

图1

表3

年龄及随机森林模型前十位的相对重要因子水平的组间比较

指标	观察组(367例)	对照组(376例)	统计检验值	P值
年龄[岁,M(Q₁, Q₃)]	53.00(40.50,65.00)	52.00(49.00,58.00)	Z=-0.084	0.938
白蛋白(g/L, $x ˉ$ ±s)	38.42±6.47	45.14±2.13	t=-18.891	<0.001
尿比重[例(构成比,%)]			χ²=34.670	<0.001
正常	27(7.36)	86(22.87)
异常	340(92.64)	290(77.13)
总胆红素[μmol/L,M(Q₁, Q₃)]	12.80(9.10,19.00)	22.25(18.40,26.90)	Z=-14.313	<0.001
间接胆红素[μmol/L,M(Q₁, Q₃)]	7.60(5.05,11.80)	17.40(14.30,19.10)	Z=-16.994	<0.001
白球比[M(Q₁, Q₃)]	1.17(0.97,1.40)	1.60(1.50,1.70)	Z=-17.030	<0.001
总胆汁酸[μmol/L, M(Q₁, Q₃)]	2.80(1.79,5.12)	4.60(3.70,5.50)	Z=-9.675	<0.001
谷草转氨酶/谷丙转氨酶比值[M(Q₁, Q₃)]	1.41(1.02,1.79)	0.95(0.77,1.15)	Z=-11.951	<0.001
淋巴细胞百分比[%,M(Q₁, Q₃)]	22.90(15.50,32.55)	34.60(29.08,39.90)	Z=-12.684	<0.001
血小板分布宽度[fl,M(Q₁, Q₃)]	15.90(15.50,16.20)	12.00(11.10,13.30)	Z=-16.907	<0.001
平均血红蛋白浓度[g/L,M(Q₁, Q₃)]	322.00(317.00,328.00)	336.00(331.00,343.00)	Z=-16.843	<0.001

表3

参考文献 21

[1]	Gourgas O, Lemire G, Eaton AJ, et al. Author Correction: Specific heterozygous variants in MGP lead to endoplasmic reticulum stress and cause spondyloepiphyseal dysplasia. Nat Commun, 2024, 15(1):3655. doi:10.1038/s41467-024-47898-x. pmid: 38688929
[2]	罗欢, 朱世琴, 沈玉兰, 等. 基于随机森林算法的糖尿病周围神经病变预测模型构建与验证. 中国糖尿病杂志, 2024, 32(8):591-594. doi:10.3969/j.issn.1006-6187.2024.08.006.
[3]	Belur Nagaraj S, Pena MJ, Ju W, et al. Machine-learning-based early prediction of end-stage renal disease in patients with diabetic kidney disease using clinical trials data. Diabetes Obes Metab, 2020, 22(12):2479-2486. doi:10.1111/dom.14178.
[4]	符剑, 陆峰, 王小平, 等. 基于多因素回归分析南通地区肺结核发病预测模型构建. 公共卫生与预防医学, 2023, 34(6):57-60. doi:10.3969/j.issn.1006-2483.2023.06.013.
[5]	Speiser JL. A random forest method with feature selection for developing medical prediction models with clustered and longitudinal data. Biomed Inform, 2021, 117:103763. doi:10.1016/j.jbi.2021.103763.
[6]	Şahin F, Yazar E, Yıldız P. Prominent features of platelet count, plateletcrit, mean platelet volume and platelet distribution width in pulmonary tuberculosis. Multidiscip Respir Med, 2012, 7(1):38. doi:10.1186/2049-6958-7-38.
[7]	梁建英, 杜元平, 张昌艳, 等. 肺结核患者血清中PCT、CRP、IgG抗PPD-IgG和白细胞计数的临床诊断意义. 标记免疫分析与临床, 2018, 25(7):997-1000,1004. doi:10.11748/bjmy.issn.1006-1703.2018.07.019.
[8]	柯学, 李国保, 沈生荣. 结核病的营养治疗. 中华结核和呼吸杂志, 2020, 43(1):8-10. doi:10.3760/cma.j.issn.1001-0939.2020.01.003.
[9]	Nikitina IY, Panteleev AV, Kosmiadi GA, et al. Th1, Th17, and Th1Th17 Lymphocytes during Tuberculosis: Th1 Lymphocytes Predominate and Appear as Low-Differentiated CXCR3⁺CCR6⁺ Cells in the Blood and Highly Differentiated CXCR3^+/-CCR6^- Cells in the Lungs. J Immunol, 2018, 200(6):2090-2103. doi:10.4049/jimmunol.1701424. pmid: 29440351
[10]	谢蓝田, 阮桂仁, 刘晓清, 等. 212例活动性结核病患者外周血淋巴细胞亚群特点:单中心描述性研究. 协和医学杂志, 2023, 14(1):131-138. doi:10.12290/xhyxzz.2022-0645.
[11]	Gazzin S, Vitek L, Watchko J, et al. A novel perspective on the biology of bilirubin in health and disease. Trends Mol Med, 2016, 22(9):758-768. doi:10.1016/j.molmed.2016.07.004. pmid: 27515064
[12]	王飞, 杨滨. 冠状动脉粥样硬化性心脏病患者血清总胆红素水平与外周动脉疾病的相关性研究. 中国医药, 2018, 13(12):1796-1799. doi:10.3760/j.issn.1673-4777.2018.12.008.
[13]	王苏, 张京梅, 张艳艳, 等. 吸烟与男性冠状动脉粥样硬化性心脏病患者血清总胆红素相关性研究. 中国医药, 2018, 13(11):1616-1619. doi:10.3760/j.issn.1673-4777.2018.11.005.
[14]	Duman H, Özyurt S. Low serum bilirubin levels associated with subclinical atherosclerosis in patients with obstructive sleep apnea. Interv Med Appl Sci, 2018, 10(4):179-185. doi:10.1556/1646.10.2018.39.
[15]	Guo X, Yang Y, Zhang B, et al. Nutrition and clinical manifestations of pulmonary tuberculosis: A cross-sectional study in Shandong province, China. Asia Pac J Clin Nutr, 2022, 31(1):41-48. doi:10.6133/apjcn.202203_31(1).0005. pmid: 35357102
[16]	Church RJ, Watkins PB. Serum biomarkers of drug-induced liver injury: Current status and future directions. J Dig Dis, 2019, 20(1):2-10. doi:10.1111/1751-2980.12684.
[17]	Sang D, Lin K, Yang Y, et al. Prolonged sleep deprivation induces a cytokine-storm-like syndrome in mammals. Cell, 2023, 186(25):5500-5516.e21. doi:10.1016/j.cell.2023.10.025. pmid: 38016470
[18]	赵慈余, 陆玲娜, 邱莲女, 等. 肺结核患者外周血T细胞活化亚群和细胞因子的变化. 中国卫生检验杂志, 2019, 29(6):705-708,712.
[19]	An HR, Bai XJ, Liang JQ, et al. The relationship between absolute counts of lymphocyte subsets and clinical features in patients with pulmonary tuberculosis. Clin Respir J, 2022, 16(5):369-379. doi:10.1111/crj.13490.
[20]	Chen LY, Liu C, Liang T, et al. Monocyte-to-Lymphocyte Ratio was an independent factor of the severity of Spinal Tuberculosis. Oxid Med Cell Longev, 2022, 2022: 7340330. doi:10.1155/2022/7340330.
[21]	Rajamanickam A, Munisankar S, Dolla CK, et al. Undernutrition is associated with perturbations in T cell-, B cell-, monocyte- and dendritic cell-subsets in latent Mycobacterium tuberculosis infection. PLoS One, 2019, 14(12):e0225611. doi:10.1371/journal.pone.0225611.