Application of Logistic Regression and Artificial Neural Networks in the Differential Diagnosis of LC-MPE
-
摘要:
目的 应用Logistic回归分析方法和人工神经网络(artificial neural network,ANN)技术,评估血清(serum,S-)和胸腔积液(pleural effusion,P-)中的癌胚抗原(CEA)、铁蛋白(FRT)、神经元特异性稀醇化酶(NSE)、鳞状细胞癌相关抗原(SCC)、糖类抗原50(CA50)、糖类抗原125(CA125)和细胞角蛋白19片段(CY21-1)在鉴别肺癌性胸腔积液(malignant pleural effusion of lung cancer,LC-MPE)与良性胸腔积液(benign pleural effusion,BPE)中的应用价值,建立通过肿瘤标志物诊断LC-MPE的诊断模型。 方法 对临床初诊的LC-MPE和BPE患者的血清和胸腔积液肿瘤标志物结果进行分析,应用Logistic回归分析和ANN技术分别建立诊断LC-MPE的诊断模型。 结果 S-NSE、S-CY21-1、P-CEA和P-NSE4项指标被筛选出并用于建模,研究建立的诊断LC-MPE的Logistic回归模型的灵敏度为93.23%,特异度为97.46%,ROC曲线下面积为0.992。建立的ANN模型的灵敏度为95.35%,特异度为97.22%,ROC曲线下面积为0.990 (P < 0.05)。 结论 在通过肿瘤标志物诊断LC-MPE方面,建立的Logistic回归模型和ANN模型均有较好的诊断性能,上述2个模型均可辅助临床医生提高诊断准确率。 Abstract:Objectives To evaluate the application value of carcinoembryonic antigen (CEA), ferritin (FRT), neuron-specific enolase (NSE), squamous cell carcinoma-related antigen (SCC), carbohydrate antigen 50 (CA50), carbohydrate antigen 125 (CA125), and cytokeratin 19 fragment (CY21-1) in serum (S-) and pleural effusion (P-) for differentiating malignant pleural effusion of lung cancer (LC-MPE) from benign pleural effusion (BPE). We aim to establish a diagnostic model for LC-MPE using tumor markers and analyze the data using logistic regression and artificial neural network (ANN) techniques. Methods The serum and pleural effusion tumor marker results of patients with newly diagnosed LC-MPE and BPE were analyzed, and diagnostic models for LC-MPE were established using Logistic regression analysis and ANN technology. Results The indicators S-NSE, S-CY21-1, P-CEA, and P-NSE were selected and used for modeling. The Logistic regression model for diagnosing LC-MPE established in this study had a sensitivity of 93.23% and a specificity of 97.46%, with an area under the ROC curve of 0.992. The established ANN model had a sensitivity of 95.35%, a specificity of 97.22%, and an area under the ROC curve of 0.990 (P < 0.05). Conclusions In diagnosing LC-MPE through tumor markers, both the Logistic regression model and the ANN model established in this study showed good diagnostic efficacy. These two models can assist clinicians in improving diagnostic accuracy. -
Key words:
- Pleural effusion /
- Artificial neural network /
- Tumor markers /
- Lung cancer /
- Diagnostic model
-
表 1 2组胸腔积液患者基本资料[n(%)]
Table 1. Basic characteristics of patients with pleural effusion in 2 groups [n(%)]
指标 LC-MPE组
(n=133)BPE组
(n=118)χ2/t P 性别 3.098 0.078 男 77(57.90) 81(68.64) 女 56(42.10) 37(31.36) 年龄 1.029 0.305 (岁) 37~88 35~89 ($\bar x \pm s $) 64.06±10.84 62.51±12.81 表 2 各单项肿瘤标志物在两组中的结果比较[M(P25,P75)]
Table 2. Comparison of results for individual tumor markers between the two groups [M(P25,P75)]
指标 LC-MPE组(n=133) BPE组(n=118) Z P S-CEA (ng/mL) 9.70 (3.64,49.01) 1.54 (1.03,2.33) 10.155 <0.001* S-FRT (ng/mL) 179.00 (109.30,307.25) 299.85 (152.93,517.55) 4.443 <0.001* S-NSE (ng/mL) 17.84 (12.86,25.85) 13.79 (10.51,19.21) 4.450 <0.001* S-SCC (ng/mL) 0.48 (0.34,0.65) 0.46 (0.36,0.56) 1.227 0.220 S-CA50 (U/mL) 8.16 (5.16,16.45) 5.44 (3.65,7.99) 4.841 <0.001* S-CA125 (U/mL) 99.89 (47.13,185.30) 76.25 (39.10,127.20) 2.682 0.007* S-CY21-1 (ng/mL) 5.64 (3.21,12.19) 1.83 (1.30,2.62) 10.946 <0.001* P-CEA (ng/mL) 125.90 (13.55,623.90) 1.21 (0.68,1.78) 11.486 <0.001* P-FRT (ng/mL) 1410.00 (883.65,2000.00 )1138.00 (597.53,2000.00 )2.414 0.016* P-NSE (ng/mL) 10.67 (5.51,30.69) 5.73 (2.99,10.74) 5.136 <0.001* P-SCC (ng/mL) 0.82 (0.57,1.81) 1.00 (0.65,2.03) 1.428 0.153 P-CA50 (U/mL) 8.39 (3.58,42.95) 3.62 (2.11,5.93) 6.638 <0.001* P-CA125 (U/mL) 965.40 (451.85, 1697.50 )397.25 (177.90,804.38) 6.056 <0.001* P-CY21-1 (ng/mL) 87.28 (30.16,227.45) 10.23 (5.63,22.34) 10.126 <0.001* *P < 0.05。 表 3 各单项肿瘤标志物的诊断性能分析
Table 3. Analysis of the diagnostic performance of individual tumor markers
指标 灵敏度(%) 特异度 (%) 阳性
预测值(%)阴性
预测值(%)准确率(%) 约登指数 AUC P 95%CI S-CEA 75.19 94.92 94.34 77.24 84.46 0.701 0.871 <0.001* 0.826~0.917 S-FRT 81.95 42.37 61.58 67.57 63.35 0.243 0.663 <0.001* 0.595~0.730 S-NSE 57.89 66.10 65.81 58.21 61.75 0.240 0.663 <0.001* 0.596~0.729 S-CA50 51.88 76.27 71.13 58.44 63.35 0.282 0.677 <0.001* 0.612~0.743 S-CA125 54.89 58.47 59.84 53.49 56.57 0.134 0.598 0.007* 0.528~0.668 S-CY21-1 72.93 86.44 85.84 73.91 79.28 0.594 0.900 <0.001* 0.864~0.937 P-CEA 84.96 98.31 98.26 85.29 91.24 0.833 0.920 <0.001* 0.880~0.961 P-FRT 66.17 47.46 58.67 55.45 57.37 0.136 0.587 0.017* 0.516~0.659 P-NSE 47.37 77.97 70.79 56.79 61.75 0.253 0.688 <0.001* 0.623~0.753 P-CA50 56.39 84.75 80.65 63.29 69.72 0.411 0.743 <0.001* 0.682~0.803 P-CA125 58.65 68.64 67.83 59.56 63.35 0.273 0.722 <0.001* 0.660~0.783 P-CY21-1 63.91 89.83 87.63 68.83 76.10 0.537 0.870 <0.001* 0.828~0.913 *P < 0.05。 表 4 肿瘤标志物的二元Logistic多因素分析
Table 4. Binary Logistic Multivariate Analysis of Tumor Markers
指标 β SE Wald P OR 95%CI S-NSE 0.099 0.042 5.627 0.018* 1.104 (1.017~1.197) S-CY211 0.606 0.214 8.048 0.005* 1.833 (1.206~2.787) P-CEA 0.829 0.225 13.577 <0.001* 2.291 (1.474~3.562) P-NSE 0.037 0.017 4.738 0.030* 1.038 (1.004~1.073) 常数 −7.900 1.503 27.645 <0.001* 0.000 − *P < 0.05。“−”表示无数据。 表 5 2种诊断模型的诊断性能比较
Table 5. Comparison of diagnostic performance of two diagnostic models
诊断模型 灵敏度(%) 特异度(%) 阳性预测值(%) 阴性预测值(%) 约登指数 AUC P 95%CI Logistic 93.23 97.46 97.64 92.74 0.907 0.992 <0.001* 0.986~0.999 ANN-测试集 90.63 98.37 98.31 90.98 0.890 0.992 <0.001* 0.985~0.999 ANN-验证集 95.35 97.22 97.62 94.59 0.926 0.990 <0.001* 0.982~0.998 *P < 0.05。 -
[1] Kim N Y,Jang B,Gu K M,et al. Differential diagnosis of pleural effusion using machine learning[J]. Ann Am Thorac Soc,2024,21(2):211-217. doi: 10.1513/AnnalsATS.202305-410OC [2] Gonnelli F,Hassan W,Bonifazi M,et al. Malignant pleural effusion: current understanding and therapeutic approach[J]. Respir Res,2024,25(1):47. doi: 10.1186/s12931-024-02684-7 [3] Han Y Q,Yan L,Li P,et al. A study investigating markers in pleural effusion (SIMPLE): A prospective and double-blind diagnostic study[J]. BMJ Open,2019,9(8):e27287. [4] Brun C,Gay P,Cottier M,et al. Comment from the authors: the tests combination in patients with lung cancer and malignant pleural effusion[J]. J Thorac Dis,2019,11(5):E74-E75. doi: 10.21037/jtd.2019.05.27 [5] Liu Q,Yu Y X,Wang X J,et al. Diagnostic accuracy of interleukin-27 between tuberculous pleural effusion and malignant pleural effusion: A meta-analysis[J]. Respiration,2018,95(6):469-477. doi: 10.1159/000486963 [6] Da C L V,Ribeiro-Alves M,Da S C R,et al. Predominance of Th1 immune response in pleural effusion of patients with tuberculosis among other exudative etiologies[J]. J Clin Microbiol,2019,58(1):e919-e927. [7] 中华医学会呼吸病学分会胸膜与纵隔疾病学组. 胸腔积液诊断的中国专家共识[J]. 中华结核和呼吸杂志,2022,45(11):1080-1096. doi: 10.3760/cma.j.cn112147-20220511-00403 [8] Gonnelli F,Hassan W,Bonifazi M,et al. Malignant pleural effusion: Current understanding and therapeutic approach[J]. Respir Res,2024,25(1):1-11. doi: 10.1186/s12931-023-02626-9 [9] Wu Q,Li M,Zhang S,et al. Clinical diagnostic utility of CA 15-3 for the diagnosis of malignant pleural effusion: A meta-analysis[J]. Exp Ther Med,2015,9(1):232-238. doi: 10.3892/etm.2014.2039 [10] Nguyen A H,Miller E J,Wichman C S,et al. Diagnostic value of tumor antigens in malignant pleural effusion: A meta-analysis[J]. Transl Res,2015,166(5):432-439. doi: 10.1016/j.trsl.2015.04.006 [11] Charakorn C,Thadanipon K,Chaijindaratana S,et al. The association between serum squamous cell carcinoma antigen and recurrence and survival of patients with cervical squamous cell carcinoma: A systematic review and meta-analysis[J]. Gynecol Oncol,2018,150(1):190-200. doi: 10.1016/j.ygyno.2018.03.056 [12] Ma Q,Liu W,Jia R,et al. Inflammation-based prognostic system predicts postoperative survival of esophageal carcinoma patients with normal preoperative serum carcinoembryonic antigen and squamous cell carcinoma antigen levels[J]. World J Surg Oncol,2016,141(14):1-6. [13] 李锐成,郜赵伟,董轲,等. 胸腔积液与血清中的癌胚抗原及其比值对结核性与肺癌性胸腔积液的诊断价值[J]. 南方医科大学学报,2019,39(2):175-180. [14] Son S M,Han H S,An J Y,et al. Diagnostic performance of CD66c in lung adenocarcinoma-associated malignant pleural effusion: Comparison with CEA,CA 19-9,and CYFRA 21-1[J]. Pathology,2015,47(2):123-129. doi: 10.1097/PAT.0000000000000215 [15] Shin Y M,Yun J,Lee O J,et al. Diagnostic value of circulating extracellular miR-134,miR-185,and miR-22 Levels in lung adenocarcinoma-associated malignant pleural effusion[J]. Cancer Res Treat,2014,46(2):178-185. doi: 10.4143/crt.2014.46.2.178 [16] Abbas M,Kassim S A,Habib M,et al. Clinical evaluation of serum tumor markers in patients with advanced-stage non-small cell lung cancer treated with palliative chemotherapy in China[J]. Front Oncol,2020,10(6):1-12. [17] Heijnen B J,Bohringer S,Speyer R. Prediction of aspiration in dysphagia using logistic regression: oral intake and self-evaluation[J]. Eur Arch Otorhinolaryngol,2020,277(1):197-205. doi: 10.1007/s00405-019-05687-z [18] Chiou S H,Betensky R A,Balasubramanian R. The missing indicator approach for censored covariates subject to limit of detection in logistic regression models[J]. Ann Epidemiol,2019,38(10):57-64. [19] Teglia C M,Guinez M,Goicoechea H C,et al. Enhancement of multianalyte mass spectrometry detection through response surface optimization by least squares and artificial neural network modelling[J]. J Chromatogr A,2020,1611(40):460613. [20] Yogeswari M K,Dharmalingam K,Mullai P. Implementation of artificial neural network model for continuous hydrogen production using confectionery wastewater[J]. J Environ Manage,2019,252(20):109684. [21] Chen Y C,Chang Y C,Ke W C,et al. Cancer adjuvant chemotherapy strategic classification by artificial neural network with gene expression data: An example for non-small cell lung cancer[J]. J Biomed Inform,2015,56(4):1-7. [22] Ligor T,Pater L,Buszewski B. Application of an artificial neural network model for selection of potential lung cancer biomarkers[J]. J Breath Res,2015,9(2):27106. doi: 10.1088/1752-7155/9/2/027106