Machine learning-based prediction of duodenal stump leakage following laparoscopic gastrectomy for gastric cancer

Yanqi Li; Yang Su; Shengli Shao; Tao Wang; Xiaokun Liu; Jichao Qin

doi:10.1016/j.surg.2024.108999

Machine learning-based prediction of duodenal stump leakage following laparoscopic gastrectomy for gastric cancer

Surgery. 2024 Dec 27:180:108999. doi: 10.1016/j.surg.2024.108999. Online ahead of print.

Authors

Yanqi Li¹, Yang Su¹, Shengli Shao², Tao Wang³, Xiaokun Liu³, Jichao Qin⁴

Affiliations

¹ Department of Surgery, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, Hubei Province, China; Molecular Medicine Center, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, Hubei Province, China.
² The Affiliated Cancer Hospital of Zhengzhou University & Henan Cancer Hospital, Zhengzhou, China.
³ Department of Gastrointestinal Surgery, The First Affiliated Hospital, Zhejiang University School of Medicine, Hangzhou, China.
⁴ Department of Surgery, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, Hubei Province, China; Molecular Medicine Center, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, Hubei Province, China; Department of Gastrointestinal Surgery, The First Affiliated Hospital, Zhejiang University School of Medicine, Hangzhou, China. Electronic address: [email protected].

PMID: 39732029
DOI: 10.1016/j.surg.2024.108999

Abstract

Background: Duodenal stump leakage is one of the most critical complications following gastrectomy surgery, with a high mortality rate. The present study aimed to establish a predictive model based on machine learning for forecasting the occurrence of duodenal stump leakage in patients who underwent laparoscopic gastrectomy for gastric cancer.

Materials and methods: The present study included the data of 4,070 patients with gastric adenocarcinoma who received laparoscopic gastrectomy. Five algorithms, namely, k-nearest neighbors, logistic regression, random forest, support vector machine, and eXtreme Gradient Boosting, were used to establish models with the preoperative and intraoperative clinical features of patients. Performance assessment was carried out to determine the optimal model.

Results: The present study involved 4,070 patients and incorporated 11 clinicopathologic features to construct machine learning models (males, 2,688, 66.0%; females, 1,382, 34.0%; age, 58 ± 11 years). Among the 5 algorithms, the support vector machine model exhibited the optimal performance, with an area under the curve of 0.866 (95% confidence interval, 0.803-0.928), sensitivity of 0.806, accuracy of 0.821, and specificity of 0.821. The analysis using the support vector machine model revealed that tumor location and clinic tumor stage significantly contributed to duodenal stump leakage.

Conclusion: The support vector machine model independently predicted duodenal stump leakage in patients with gastric cancer and exhibited favorable discrimination and accuracy. Thus, the construction of an efficient and intuitive online predictive tool demonstrated that the support vector machine model may exhibit potential in the prevention and adjunctive treatment of duodenal stump leakage. The model indicates that besides tumor location and stage, operation time, preoperative pyloric obstruction, and patient age also are important factors that have a significant impact on the occurrence of duodenal stump leakage after surgery.