Scholarlink

<p>Very early recurrence (within one year) after curative hepatectomy significantly impairs long-term survival in patients with hepatitis B virus (HBV)-related hepatocellular carcinoma (HCC). Accurately predicting this risk preoperatively is crucial for adjusting clinical decision-making. This study aimed to develop and externally validate a robust machine learning model using exclusively routinely available preoperative clinical data to predict very early recurrence. We retrospectively analyzed 943 patients with HBV-related HCC from two centers. To capture complex, non-linear clinical interactions, we systematically evaluated seven feature selection strategies in combination with seven machine learning algorithms. Sequential Forward Selection (SFS) and SHapley Additive exPlanations (SHAP) were applied to identify core predictive features and interpret variable importance. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) across training, validation, and external test sets, and compared with conventional benchmark models. X-tile was used for risk stratification, followed by Kaplan-Meier survival analyses. The cohort comprised a training-validation set (<i>n</i> = 754) and an external test set (<i>n</i> = 189). The Random Forest (RF) model demonstrated the most stable and superior predictive accuracy, achieving AUCs of 0.864 (95% CI: 0.832–0.897), 0.814 (95% CI: 0.758–0.870), and 0.843 (95% CI: 0.786–0.899) in the training, validation, and external test sets, respectively, significantly outperforming conventional statistical models. The model effectively stratified patients into high- and low-risk cohorts with significantly distinct disease-free survival (DFS) across all datasets (all <i>P</i> &lt; 0.001). A parsimonious set of eight core preoperative features—comprehensively reflecting tumor biology (tumor size, number, capsule, and AFP) and the underlying liver condition (age, total bilirubin, PALBI grade, and FIB-4 index)—was integrated into an accessible web-based calculator. By systematically integrating key characteristics of tumor aggressiveness and the liver microenvironment, this RF-based preoperative model reliably predicts very early recurrence after hepatectomy. The user-friendly web tool empowers clinicians to preoperatively identify high-risk patients, potentially preventing rushed upfront surgeries in favor of tailored neoadjuvant therapies to optimize survival outcomes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development and validation of a machine learning model to preoperatively predict very early recurrence in patients with hepatitis B virus-related hepatocellular carcinoma

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月7日

查看全文
  • Yuting Yang,
  • Haojie Yang,
  • Shuxiang Hou,
  • Kelan Zhang,
  • Zemin Xiao

摘要

Very early recurrence (within one year) after curative hepatectomy significantly impairs long-term survival in patients with hepatitis B virus (HBV)-related hepatocellular carcinoma (HCC). Accurately predicting this risk preoperatively is crucial for adjusting clinical decision-making. This study aimed to develop and externally validate a robust machine learning model using exclusively routinely available preoperative clinical data to predict very early recurrence. We retrospectively analyzed 943 patients with HBV-related HCC from two centers. To capture complex, non-linear clinical interactions, we systematically evaluated seven feature selection strategies in combination with seven machine learning algorithms. Sequential Forward Selection (SFS) and SHapley Additive exPlanations (SHAP) were applied to identify core predictive features and interpret variable importance. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) across training, validation, and external test sets, and compared with conventional benchmark models. X-tile was used for risk stratification, followed by Kaplan-Meier survival analyses. The cohort comprised a training-validation set (n = 754) and an external test set (n = 189). The Random Forest (RF) model demonstrated the most stable and superior predictive accuracy, achieving AUCs of 0.864 (95% CI: 0.832–0.897), 0.814 (95% CI: 0.758–0.870), and 0.843 (95% CI: 0.786–0.899) in the training, validation, and external test sets, respectively, significantly outperforming conventional statistical models. The model effectively stratified patients into high- and low-risk cohorts with significantly distinct disease-free survival (DFS) across all datasets (all P < 0.001). A parsimonious set of eight core preoperative features—comprehensively reflecting tumor biology (tumor size, number, capsule, and AFP) and the underlying liver condition (age, total bilirubin, PALBI grade, and FIB-4 index)—was integrated into an accessible web-based calculator. By systematically integrating key characteristics of tumor aggressiveness and the liver microenvironment, this RF-based preoperative model reliably predicts very early recurrence after hepatectomy. The user-friendly web tool empowers clinicians to preoperatively identify high-risk patients, potentially preventing rushed upfront surgeries in favor of tailored neoadjuvant therapies to optimize survival outcomes.