{"title":"机器学习模型将肝细胞癌患者在治愈切除后分为高、低复发或死亡风险组。","authors":"Jun-Jun Jia, Yu-Yang Wang, Xin-Yue Tan, Yu Nie, Shu-Sen Zheng, Hang-Jin Jiang","doi":"10.1016/j.hbpd.2025.09.001","DOIUrl":null,"url":null,"abstract":"<p><strong>Background: </strong>The high recurrence rate of hepatocellular carcinoma (HCC) following curative resection affects patient survival. The present study combined critical clinicopathological features and molecular markers to develop machine learning models to predict the risk of recurrence and mortality. We aimed to individualize risk stratification, post-surgical management strategies, and ultimately improve long-term prognosis for HCC patients with curative resections.</p><p><strong>Methods: </strong>A total of 815 HCC patients undergoing surgical resection were divided randomly into a training cohort (n = 652) and a validation cohort (n = 163). To build a high-accuracy recurrent/death classifier using clinicopathological characteristics and molecular biomarkers, four different machine learning models, including the Cox proportional risk model, generalized linear model, extreme gradient boosting (XGBoost) model, and random survival forest (RSF) model, were developed and comprehensively compared. The outcomes were recurrence-free survival (RFS) and overall survival (OS).</p><p><strong>Results: </strong>Factors including diabetes, albumin, tumor numbers, HCC diameter, portal vein tumor thrombus, blood loss, mismatch repair protein 2 (MSH2), and epithelial membrane antigen were significantly associated with RFS, while albumin, HCC diameter, MSH2, and Barcelona Clinic Liver Cancer (BCLC) stage were significantly associated with OS. The RSF model not only grouped HCC patients into high- and low-probability recurrence groups with significant differences in 5-year recurrence probability rate (training cohort: 87.3 % vs. 51.5 %, P < 0.0001; validation cohort: 75.9 % vs. 64.8 %, P < 0.0001), but also grouped HCC patients into high- and low-probability death groups with significant differences in 5-year death probability rate (training cohort: 56.0 % vs. 15.3 %, P < 0.0001; validation cohort: 50.0 % vs. 23.1 %, P < 0.0001).</p><p><strong>Conclusions: </strong>The RSF model accurately stratified HCC patient into high- and low-risk recurrence or death groups, which guides the surgeons to plan adjuvant therapy after surgery.</p>","PeriodicalId":55059,"journal":{"name":"Hepatobiliary & Pancreatic Diseases International","volume":" ","pages":""},"PeriodicalIF":4.4000,"publicationDate":"2025-09-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Machine learning model stratify hepatocellular carcinoma patient into high- and low-risk recurrence or death group post curative resection.\",\"authors\":\"Jun-Jun Jia, Yu-Yang Wang, Xin-Yue Tan, Yu Nie, Shu-Sen Zheng, Hang-Jin Jiang\",\"doi\":\"10.1016/j.hbpd.2025.09.001\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p><strong>Background: </strong>The high recurrence rate of hepatocellular carcinoma (HCC) following curative resection affects patient survival. The present study combined critical clinicopathological features and molecular markers to develop machine learning models to predict the risk of recurrence and mortality. We aimed to individualize risk stratification, post-surgical management strategies, and ultimately improve long-term prognosis for HCC patients with curative resections.</p><p><strong>Methods: </strong>A total of 815 HCC patients undergoing surgical resection were divided randomly into a training cohort (n = 652) and a validation cohort (n = 163). To build a high-accuracy recurrent/death classifier using clinicopathological characteristics and molecular biomarkers, four different machine learning models, including the Cox proportional risk model, generalized linear model, extreme gradient boosting (XGBoost) model, and random survival forest (RSF) model, were developed and comprehensively compared. The outcomes were recurrence-free survival (RFS) and overall survival (OS).</p><p><strong>Results: </strong>Factors including diabetes, albumin, tumor numbers, HCC diameter, portal vein tumor thrombus, blood loss, mismatch repair protein 2 (MSH2), and epithelial membrane antigen were significantly associated with RFS, while albumin, HCC diameter, MSH2, and Barcelona Clinic Liver Cancer (BCLC) stage were significantly associated with OS. The RSF model not only grouped HCC patients into high- and low-probability recurrence groups with significant differences in 5-year recurrence probability rate (training cohort: 87.3 % vs. 51.5 %, P < 0.0001; validation cohort: 75.9 % vs. 64.8 %, P < 0.0001), but also grouped HCC patients into high- and low-probability death groups with significant differences in 5-year death probability rate (training cohort: 56.0 % vs. 15.3 %, P < 0.0001; validation cohort: 50.0 % vs. 23.1 %, P < 0.0001).</p><p><strong>Conclusions: </strong>The RSF model accurately stratified HCC patient into high- and low-risk recurrence or death groups, which guides the surgeons to plan adjuvant therapy after surgery.</p>\",\"PeriodicalId\":55059,\"journal\":{\"name\":\"Hepatobiliary & Pancreatic Diseases International\",\"volume\":\" \",\"pages\":\"\"},\"PeriodicalIF\":4.4000,\"publicationDate\":\"2025-09-09\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Hepatobiliary & Pancreatic Diseases International\",\"FirstCategoryId\":\"3\",\"ListUrlMain\":\"https://doi.org/10.1016/j.hbpd.2025.09.001\",\"RegionNum\":3,\"RegionCategory\":\"医学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"GASTROENTEROLOGY & HEPATOLOGY\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Hepatobiliary & Pancreatic Diseases International","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.1016/j.hbpd.2025.09.001","RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"GASTROENTEROLOGY & HEPATOLOGY","Score":null,"Total":0}
Machine learning model stratify hepatocellular carcinoma patient into high- and low-risk recurrence or death group post curative resection.
Background: The high recurrence rate of hepatocellular carcinoma (HCC) following curative resection affects patient survival. The present study combined critical clinicopathological features and molecular markers to develop machine learning models to predict the risk of recurrence and mortality. We aimed to individualize risk stratification, post-surgical management strategies, and ultimately improve long-term prognosis for HCC patients with curative resections.
Methods: A total of 815 HCC patients undergoing surgical resection were divided randomly into a training cohort (n = 652) and a validation cohort (n = 163). To build a high-accuracy recurrent/death classifier using clinicopathological characteristics and molecular biomarkers, four different machine learning models, including the Cox proportional risk model, generalized linear model, extreme gradient boosting (XGBoost) model, and random survival forest (RSF) model, were developed and comprehensively compared. The outcomes were recurrence-free survival (RFS) and overall survival (OS).
Results: Factors including diabetes, albumin, tumor numbers, HCC diameter, portal vein tumor thrombus, blood loss, mismatch repair protein 2 (MSH2), and epithelial membrane antigen were significantly associated with RFS, while albumin, HCC diameter, MSH2, and Barcelona Clinic Liver Cancer (BCLC) stage were significantly associated with OS. The RSF model not only grouped HCC patients into high- and low-probability recurrence groups with significant differences in 5-year recurrence probability rate (training cohort: 87.3 % vs. 51.5 %, P < 0.0001; validation cohort: 75.9 % vs. 64.8 %, P < 0.0001), but also grouped HCC patients into high- and low-probability death groups with significant differences in 5-year death probability rate (training cohort: 56.0 % vs. 15.3 %, P < 0.0001; validation cohort: 50.0 % vs. 23.1 %, P < 0.0001).
Conclusions: The RSF model accurately stratified HCC patient into high- and low-risk recurrence or death groups, which guides the surgeons to plan adjuvant therapy after surgery.
期刊介绍:
Hepatobiliary & Pancreatic Diseases International (HBPD INT) (ISSN 1499-3872 / CN 33-1391/R) a bimonthly journal published by First Affiliated Hospital, Zhejiang University School of Medicine, China. It publishes peer-reviewed original papers, reviews and editorials concerned with clinical practice and research in the fields of hepatobiliary and pancreatic diseases. Papers cover the medical, surgical, radiological, pathological, biochemical, physiological and historical aspects of the subject areas under the headings Liver, Biliary, Pancreas, Transplantation, Research, Special Reports, Editorials, Review Articles, Brief Communications, Clinical Summary, Clinical Images and Case Reports. It also deals with the basic sciences and experimental work. The journal is abstracted and indexed in SCI-E, IM/MEDLINE, EMBASE/EM, CA, Scopus, ScienceDirect, etc.