Construction of a prediction model for postoperative acute respiratory failure risk in patients with esophageal cancer based on SMOTE algorithm

doi:10.3760/cma.j.cn371439-20250626-00035

Abstract

Abstract:

Objective To construct a prediction model for postoperative acute respiratory failure （ARF） risk in patients with esophageal cancer based on the synthetic minority oversampling technique （SMOTE） algorithm. Methods A retrospective analysis was conducted on the clinical data of 450 patients who underwent surgical treatment for esophageal cancer in Jiangsu Province Hospital from March 2023 to March 2025. Based on the occurrence of ARF after surgery， the patients were divided into an ARF group （45 cases） and a non-ARF group （405 cases）. The clinical data were compared between patients in the two groups， and multivariate logistic regression analysis was used to identify influencing factors for ARF in patients with esophageal cancer after surgery. A logistic regression prediction model was constructed. A prediction model for the occurrence of ARF in patients with esophageal cancer after surgery was constructed using the SMOTE algorithm. Model calibration was assessed using Cox-Snell R². The predictive efficacy was evaluated using the receiver operator characteristic （ROC） curve. Results There were statistically significant differences in smoking history， surgery duration， anastomotic leakage， postoperative sputum obstruction， postoperative pneumonia， and postoperative mechanical ventilation duration between the ARF group and the non-ARF group （all P<0.05）. Multivariate analysis showed that smoking history （OR=3.57， 95%CI： 1.60-7.97， P=0.002）， surgery duration >4 h （OR=2.89， 95%CI： 1.49-5.98， P=0.002）， anastomotic leakage （OR=3.09， 95%CI： 1.04-9.17， P=0.042）， postoperative sputum obstruction （OR=2.69， 95%CI： 1.34-5.41， P=0.005）， postoperative pneumonia （OR=2.61， 95%CI： 1.24-5.50， P=0.011）， and postoperative mechanical ventilation duration >48 h （OR=4.26， 95%CI： 1.68-10.80， P=0.002） were independent risk factors for ARF in patients with esophageal cancer after surgery. The prediction model based on the SMOTE algorithm was logit（P）=-4.74+2.90×smoking history+2.52×surgery duration+1.69×anastomotic leakage+1.51×postoperative sputum obstruction+1.49×postoperative pneumonia+1.88×postoperative mechanical ventilation duration. The prediction model based on the SMOTE algorithm demonstrated good calibration （Cox-Snell R²=0.537； χ²=118.34， df=6， P<0.001）. The ROC curve analysis showed that the area under the curve of the prediction model based on the SMOTE algorithm was 0.835 （95%CI： 0.803-0.866）， which was higher than that of the logistic regression prediction model （the area under the curve was 0.783， 95%CI： 0.712-0.854； Z=2.35， P=0.019）. Conclusions The risk prediction model for ARF after esophageal cancer surgery based on the SMOTE algorithm exhibits excellent predictive efficacy， supporting its potential utility in identifying patients with high risk of ARF after esophageal cancer surgery.

Key words: Esophageal neoplasms, Respiratory insufficiency, Proportional hazards models, SMOTE algorithm

Xie Wenjuan, Zhu Yuan, Xu Jing. Construction of a prediction model for postoperative acute respiratory failure risk in patients with esophageal cancer based on SMOTE algorithm[J]. Journal of International Oncology, 2026, 53(4): 213-218.

Figures/Tables 4

References 21

[1]	郑渊杰, 滕熠, 何思怡, 等. 2022年全球及中国食管癌流行病学特征[J]. 中国肿瘤, 2025, 34(3): 165-170. DOI: 10.11735/j.issn.1004-0242.2025.03.A001.
[2]	Dong J, Wang GD, Wang HZ. Clinical analysis of patients with respiratory failure after esophageal cancer operation[J]. Transl Cancer Res, 2021, 10(12): 5238-5245. DOI: 10.21037/tcr-21-1505. pmid: 35116373
[3]	Saillard C, Lambert J, Tramier M, et al. High-flow nasal cannula failure in critically ill cancer patients with acute respiratory failure: moving from avoiding intubation to avoiding delayed intubation[J]. PLoS One, 2022, 17(6): e0270138. DOI: 10.1371/journal.pone.0270138.
[4]	Heybati K, Deng JW, Bhandarkar A, et al. Outcomes of acute respiratory failure in patients with cancer in the United States[J]. Mayo Clin Proc, 2024, 99(4): 578-592. DOI: 10.1016/j.mayocp.2023.07.024. pmid: 38456872
[5]	Liu LH, Bi B, Cao L, et al. Predictive model and risk analysis for peripheral vascular disease in type 2 diabetes mellitus patients using machine learning and shapley additive explanation[J]. Front Endocrinol (Lausanne), 2024, 15(1): 1320335. DOI: 10.3389/fendo.2024.1320335.
[6]	Zhang W, Zeng WM, Chen HY, et al. STANet: a novel spatio-temporal aggregation network for depression classification with small and unbalanced FMRI data[J]. Tomography, 2024, 10(12): 1895-1914. DOI: 10.3390/tomography10120138. pmid: 39728900
[7]	王祎, 郑晨, 李启慧, 等. 《NCCN食管和食管胃结合部癌临床实践指南(2023年第3版)》更新解读[J]. 中国胸心血管外科临床杂志, 2024, 31(1): 11-16. DOI: 10.7507/1007-4848.202310024.
[8]	Wang HJ, Chen W, Wang HZ, et al. Expert consensus on acute respiratory failure in critically Ⅲ cancer patients (2023)[J]. Chin Med Sci J, 2023, 38(3): 163-177. DOI: 10.24920/004203.
[9]	Blagus R, Lusa L. Improved shrunken centroid classifiers for high-dimensional class-imbalanced data[J]. BMC Bioinformatics, 2013, 14(1): 64. DOI: 10.1186/1471-2105-14-64.
[10]	Karamti H, Alharthi R, Anizi AA, et al. Improving prediction of cervical cancer using KNN imputed SMOTE features and multi-model ensemble learning approach[J]. Cancers (Basel), 2023, 15(17): 4412. DOI: 10.3390/cancers15174412.
[11]	Hirano Y, Konishi T, Kaneko H, et al. Impact of prophylactic corticosteroid use on in-hospital mortality and respiratory failure after esophagectomy for esophageal cancer: nationwide inpatient data study in Japan[J]. Ann Surg, 2023, 277(6): e1247-e1253. DOI: 10.1097/SLA.0000000000005502.
[12]	Donlon NE, Moran B, Kamilli A, et al. CROSS versus FLOT regimens in esophageal and esophagogastric junction adenocarcinoma: a propensity-matched comparison[J]. Ann Surg, 2022, 276(5): 792-798. DOI: 10.1097/SLA.0000000000005617.
[13]	刘会鹏, 李莉, 张彦平. 食管癌患者术后发生急性呼吸衰竭的影响因素研究[J]. 实用癌症杂志, 2023, 38(9): 1536-1539. DOI: 10.3969/j.issn.1001-5930.2023.09.037.
[14]	张湛, 王晓兵, 凌萌, 等. 食管癌术后呼吸衰竭危险因素分析及预测模型的构建[J]. 河北医学, 2019, 25(10): 1593-1598. DOI: 10.3969/j.issn.1006-6233.2019.10.003.
[15]	马欢, 任志慧, 杜辉, 等. 食管癌根治术后急性呼吸衰竭危险因素分析及列线图预测模型构建[J]. 肿瘤学杂志, 2024, 30(6): 492-497. DOI: 10.11735/j.issn.1671-170X.2024.06.B007.
[16]	Yu BY, Liu ZC, Zhang L, et al. Pre- and intra-operative risk factors predict postoperative respiratory failure after minimally invasive oesophagectomy[J]. Eur J Cardiothorac Surg, 2024, 65(4): ezae107. DOI: 10.1093/ejcts/ezae107.
[17]	梁绮红, 余晓琳, 安胜利. 定量资料非平衡设计下各组样本量对检验效能的影响[J]. 南方医科大学学报, 2020, 40(5): 713-717. DOI: 10.12122/j.issn.1673-4254.2020.05.16.
[18]	Fu JQ, Cai WF, Zeng BW, et al. Development and validation of a predictive model for peripherally inserted central catheter-related thrombosis in breast cancer patients based on artificial neural network: a prospective cohort study[J]. Int J Nurs Stud, 2022, 135(1): 104341. DOI: 10.1016/j.ijnurstu.2022.104341.
[19]	Rehman MU, Naseem S, Butt AUR, et al. Predicting coronary heart disease with advanced machine learning classifiers for improved cardiovascular risk assessment[J]. Sci Rep, 2025, 15(1): 13361. DOI: 10.1038/s41598-025-96437-1.
[20]	Truong B, Zheng JY, Hornsby L, et al. Development and validation of machine learning algorithms to predict 1-year ischemic stroke and bleeding events in patients with atrial fibrillation and cancer[J]. Cardiovasc Toxicol, 2024, 24(4): 365-374. DOI: 10.1007/s12012-024-09843-8.
[21]	Hassanzadeh R, Farhadian M, Rafieemehr H. Hospital mortality prediction in traumatic injuries patients: comparing different SMOTE-based machine learning algorithms[J]. BMC Med Res Methodol, 2023, 23(1): 101. DOI: 10.1186/s12874-023-01920-w. pmid: 37087425

一般资料	ARF组（n=45）	无ARF组（n=405）	t/χ²值	P值
年龄（岁）^a	59.22±6.42	58.43±6.37	0.79	0.432
性别^b
男	31	299	0.51	0.477
女	14	106	0.51	0.477
BMI（kg/m²）^a	22.43±3.12	22.56±2.48	0.33	0.738
吸烟史^b
有	35	219	9.26	0.002
无	10	186	9.26	0.002
饮酒史^b
有	23	194	0.17	0.683
无	22	211	0.17	0.683
高血压^b
是	19	124	2.52	0.113
否	26	281	2.52	0.113
糖尿病^b
有	23	152	3.14	0.076
无	22	253	3.14	0.076
肿瘤分期^b
Ⅰ~Ⅱ期	18	220	3.33	0.068
Ⅲ~Ⅳ期	27	185	3.33	0.068
吻合部位^b
颈部	21	201	0.14	0.706
胸部	24	204	0.14	0.706
术中出血量（ml）^b
>150	14	115	0.15	0.702
≤150	31	290	0.15	0.702
手术时间（h）^b
>4	28	157	9.20	0.002
≤4	17	248	9.20	0.002
吻合口渗漏^b
是	8	21	8.67	0.003
否	37	384	8.67	0.003
术后痰液堵塞^b
是	19	87	9.68	0.002
否	26	318	9.68	0.002
术后并发肺炎^b
是	15	62	9.28	0.002
否	30	343	9.28	0.002
术后机械通气时间（h）^b
>48	9	26	10.41	0.001
≤48	36	379	10.41	0.001

自变量	β值	SE值	Wald χ²值	OR值	95%CI	P值
吸烟史（有/无）	1.27	0.41	9.65	3.57	1.60~7.97	0.002
糖尿病（有/无）	0.64	0.35	3.36	1.89	0.96~3.75	0.067
肿瘤分期（Ⅰ~Ⅱ期/Ⅲ~Ⅳ期）	0.64	0.35	3.38	1.90	0.96~3.77	0.066
手术时间（>4 h/≤4 h）	1.09	0.36	9.46	2.98	1.49~5.98	0.002
吻合口渗漏（是/否）	1.13	0.56	4.14	3.09	1.04~9.17	0.042
术后痰液堵塞（是/否）	0.99	0.36	7.77	2.69	1.34~5.41	0.005
术后并发肺炎（是/否）	0.96	0.38	6.42	2.61	1.24~5.50	0.011
术后机械通气时间（>48 h/≤48 h）	1.45	0.48	9.31	4.26	1.68~10.80	0.002
常量	-5.07	0.59	74.12	0.01		<0.001

自变量	β值	SE值	Wald χ²值	OR值	95%CI	P值
吸烟史（有/无）	2.90	0.30	9.76	18.25	10.14~32.86	<0.001
糖尿病（有/无）	0.42	0.27	2.41	1.52	0.90~2.57	0.121
肿瘤分期（Ⅰ~Ⅱ期/Ⅲ~Ⅳ期）	0.38	0.25	2.31	1.46	0.90~2.40	0.129
手术时间（>4 h/≤4 h）	2.52	0.25	10.61	12.48	7.64~20.38	<0.001
吻合口渗漏（是/否）	1.69	0.42	16.45	5.44	2.40~12.32	<0.001
术后痰液堵塞（是/否）	1.51	0.26	34.74	4.55	2.75~7.52	<0.001
术后并发肺炎（是/否）	1.49	0.29	27.50	4.45	2.55~7.77	<0.001
术后机械通气时间（>48 h/≤48 h）	1.88	0.41	20.97	6.55	2.93~14.65	<0.001
常量	-4.74	0.40	138.01	0.01		<0.001