Scholarlink

Background <p>Accurate risk stratification of biliary complications (BCs) after liver transplantation (LT) remains challenging. This study aimed to develop and validate a machine learning (ML) and nomogram framework comprising a ML-based web calculator and a clinically interpretable nomogram for post-LT BCs.</p> Methods <p>This retrospective study analyzed 133 LT patients (2011–2025), randomly split into training (<i>n</i> = 94) and validation (<i>n</i> = 39) sets. Predictors were identified using Least Absolute Shrinkage and Selection Operator (LASSO) regression. Eight ML algorithms were trained with 5-fold cross-validation. Performance was evaluated using the area under the receiver operating characteristic curve (AUC), Brier score, and decision curve analysis (DCA). Additionally, SHapley Additive exPlanations (SHAP) were employed to visualize feature importance.</p> Results <p>The cumulative incidence of BCs was 38.3%. Four independent predictors were identified: hepatocellular carcinoma (HCC) emerged as a protective factor, while intraoperative crystalloid infusion, preoperative partial hepatectomy, and antiviral therapy were risk factors. Light Gradient Boosting Machine (LightGBM) yielded the highest discrimination in the validation set (AUC 0.753), outperforming standard logistic regression (AUC 0.701). The LightGBM model exhibited satisfactory calibration (Brier score = 0.212) and clinical net benefit. The nomogram provided a static scoring tool, whereas the LightGBM-based web calculator offered precise individual risk estimation.</p> Conclusions <p>As an exploratory, proof-of-concept framework and an internal validation effort, the LightGBM-based web calculator offers exploratory but promising predictive accuracy for post-LT BCs, while the nomogram facilitates bedside decision-making through visual interpretability. This combined framework supports a transition from reactive management to proactive, risk-stratified surveillance.</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

Predicting biliary complications after liver transplantation: a machine learning and nomogram-based exploratory study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月12日

查看全文
  • Han Zhang,
  • Jinxi Chen,
  • Minghang Zhang,
  • Kezhen Zong,
  • Shanshan Li,
  • Zuotian Huang,
  • Zhongjun Wu

摘要

Background

Accurate risk stratification of biliary complications (BCs) after liver transplantation (LT) remains challenging. This study aimed to develop and validate a machine learning (ML) and nomogram framework comprising a ML-based web calculator and a clinically interpretable nomogram for post-LT BCs.

Methods

This retrospective study analyzed 133 LT patients (2011–2025), randomly split into training (n = 94) and validation (n = 39) sets. Predictors were identified using Least Absolute Shrinkage and Selection Operator (LASSO) regression. Eight ML algorithms were trained with 5-fold cross-validation. Performance was evaluated using the area under the receiver operating characteristic curve (AUC), Brier score, and decision curve analysis (DCA). Additionally, SHapley Additive exPlanations (SHAP) were employed to visualize feature importance.

Results

The cumulative incidence of BCs was 38.3%. Four independent predictors were identified: hepatocellular carcinoma (HCC) emerged as a protective factor, while intraoperative crystalloid infusion, preoperative partial hepatectomy, and antiviral therapy were risk factors. Light Gradient Boosting Machine (LightGBM) yielded the highest discrimination in the validation set (AUC 0.753), outperforming standard logistic regression (AUC 0.701). The LightGBM model exhibited satisfactory calibration (Brier score = 0.212) and clinical net benefit. The nomogram provided a static scoring tool, whereas the LightGBM-based web calculator offered precise individual risk estimation.

Conclusions

As an exploratory, proof-of-concept framework and an internal validation effort, the LightGBM-based web calculator offers exploratory but promising predictive accuracy for post-LT BCs, while the nomogram facilitates bedside decision-making through visual interpretability. This combined framework supports a transition from reactive management to proactive, risk-stratified surveillance.