中国实用外科杂志

中国实用外科杂志 ›› 2025, Vol. 45 ›› Issue (08): 923-930.DOI: 10.19538/j.cjps.issn1005-2208.2025.08.14

• 论著 • 上一篇    下一篇

基于术前血清学标记物的早期肝细胞癌长期生存预测模型：一项多中心回顾性研究

曾建兴1a，吴琼兰1b，曾金华1a，黄    尧1a，周伟平2，刘景丰1a，曾永毅1a   

  1. 1福建医科大学孟超肝胆医院  a.肝胆外科  b.病理科，福建福州 350025；2海军军医大学附属第三医院肝外三科，上海 200025
  • 出版日期:2025-08-01 发布日期:2025-09-02

  • Online:2025-08-01 Published:2025-09-02

摘要： 目的    基于客观、简便、临床常规监测的术前血清学标记物，建立早期肝细胞癌病人长期生存预测模型。方法  回顾性分析2012年1月至2018年12月多中心数据库——福建医科大学孟超肝胆医院原发性肝癌大数据库3100例病人的资料，随机数字表法分为训练队列（2100例）和验证队列（1000例）；应用LASSO回归对性别、年龄、病因、12个血清学标记物等15个变量筛选出与5年生存相关的因素，进一步纳入多因素Cox回归构建长期生存预测模型。模型分辨力由C指数、K指数和时间依赖的受试者工作特征曲线下面积（tdAUC）来测量。模型校准能力通过校准曲线进行测量。模型净收益通过决策曲线进行测量。结果    经过10折交叉验证，应用LASSO回归最终筛选出白蛋白（ALB）、碱性磷酸酶（ALP）及甲胎蛋白（AFP）3个血清学标记物，并构建早期肝细胞癌长期生存预测模型，命名为APP模型，计算肝细胞癌病人5年内死亡风险值= 0.390×ln（ALP）+0.063×ln（AFP）-0.033×ALB。在训练队列和验证队列，APP模型的C指数分别为0.619（标准误=0.010）和0.613（标准误=0.015）。APP模型的K指数分别为0.595（标准误=0.009）和0.574（标准误=0.013）。APP模型的1年tdAUC分别是0.685（标准误=0.021）和0.670（标准误=0.033），3年tdAUC分别是0.660（标准误=0.014）和0.659（标准误=0.021），5年tdAUC分别是0.610（标准误=0.016）和0.581（标准误=0.023）。以上指标均高于其他5个模型。根据模型计算得分将病人分为3个风险组（低危组、中危组、高危组），高危组病人总生存期低于中危及低危组，差异有统计学意义（P＜0.05）。结论    APP模型是基于客观、简便、临床常规监测的术前血清学标记物构建的早期肝细胞癌长期生存预测模型，其预测效能优于其他预测模型，能够指导个体化随访及监测。

关键词: 肝细胞癌, 血清学标记物, 长期生存, 预后模型, 个体化预测

Abstract: To develop a simple and low-cost prognostic score for patients with early hepatocellular carcinoma (HCC) based on serum biomarkers in routine clinical practice. Methods    3100 patients from a multicenter database (primary liver cancer big data of Mengchao Hepatobiliary Hospital, Fujian Medical University) between January 2012 and December 2018 were retrospective analyzed. The patients were randomly divided into training cohorts (2100 cases) and validation cohorts (1000 cases) using the random number table method. Least absolute shrinkage and selector operation (LASSO) was used to select the significant factors of 5-year survival from 15 variables including gender, age, etiology, and 12 serum biomarkers. The prognostic model was built based on the results of multivariate Cox regression. Model discrimination was measured by C-index, K-index, and time-dependent areas under the receiver operating characteristic curve (tdAUC). Model calibration was measured by the calibration curve. Net benefit of the model was measured by the decision curve. Results    Albumin (ALB), alkaline phosphatase (ALP), and alpha-fetoprotein (AFP) were selected by the LASSO algorithm with 10-fold cross-validation. The three variables were incorporated into multivariate Cox regression to construct the risk model (APP model). The risk score of 5-year death = 0.390×ln(ALP) + 0.063×ln(AFP) - 0.033×ALB. In training and validation cohorts, the C-indexes of APP model were 0.619 ［standard errors (SE)=0.010］ and 0.613 (SE=0.015), the K-indexes were 0.595 (SE=0.009) and 0.574 (SE=0.013), the 1-year tdAUC were 0.685 (SE=0.021) and 0.670 (SE=0.033), the 3-year tdAUC were 0.660 (SE=0.014) and 0.659 (SE=0.021), the 5-year tdAUC were 0.610 (SE=0.016) and 0.581 (SE=0.023), respectively. The above indicators showed that the APP model was greater than 5 other models. Based on the scores calculated by this model, the patients were divided into 3 risk groups (low-risk group, medium-risk group, and high-risk group). The median overall survival of patients in the high-risk group was shorter than that of the medium-risk group and the low-risk group, and the difference was statistically significant (P<0.05). Conclusion    The APP model was developed on objective and convenient preoperative serum biomarkers in routine clinical practice to estimate long-term survival probability of early HCC, outperforming other prediction models and helping guide individualized follow-up and monitoring.

Key words: hepatocellular carcinoma, serum biomarker, long-term survival, prognostic model, individualized prediction