Scholarlink

Background <p>Severe acute pancreatitis (SAP) involves significant fluid loss, necessitating precise fluid therapy. However, standardized protocols are lacking. This study aimed to develop a machine learning-based prediction model for fluid requirement in SAP patients.</p> Methods <p>We conducted a retrospective observational study of SAP patients admitted to Peking University Third Hospital (2016–2020) within 48&#xa0;h of onset. Fluid requirement was quantified by 48-hour rehydration volume. Using the Lasso algorithm, we screened predictive variables and trained five models: Gradient Boosting Decision Tree (GBDT), eXtreme Gradient Boosting (XGBoost), LightGBM, CatBoost, and multiple linear regression. Model performance was evaluated via mean absolute error (MAE), root mean square error (RMSE), R<sup>2</sup>, and fitting curves. The soft voting method was used to fuse the above five prediction models to improve the performance of the model. The optimal model was interpreted using SHapley Additive exPlanations (SHAP). To illustrate potential clinical use, ten randomly selected cases from the test set are presented.</p> Results <p>Among 308 included patients, 90% were allocated to the training set. Sixteen key variables were selected for prediction. Among the five machine learning algorithms used to build prediction models, the MAE and RMSE values of the XGBoost algorithm were the smallest and the R<sup>2</sup> value was the closest to 1, which indicated that XGBoost was the best-performing model in our study. After model fusion, the model performance was further improved. SHAP analysis of the optimal model, XGBoost, revealed the relative importance of each predictor. To illustrate the model’s practical application, we present the predictions for 10 randomly selected test-set cases. The differences between the predicted and actual fluid volumes in these individual cases ranged from 31.07 to 329.80 mL, serving as concrete examples of how the model could be used at the bedside.</p> Conclusion <p>In this study, we developed the Fluid Requirement Predicting Model for SAP (FRPM-SAP), which can predict the specific amount of fluid loss in SAP patients. The predictive performance was good, demonstrating that the model has practical application for guiding clinicians in their assessment of 48-hour rehydration volume.</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

Predictive model for early fluid requirement in severe acute pancreatitis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月22日

查看全文
  • Ayijiang Jiamaliding,
  • Gaolei Sun,
  • Junhong Wang,
  • Minjie Duan,
  • Qingbian Ma

摘要

Background

Severe acute pancreatitis (SAP) involves significant fluid loss, necessitating precise fluid therapy. However, standardized protocols are lacking. This study aimed to develop a machine learning-based prediction model for fluid requirement in SAP patients.

Methods

We conducted a retrospective observational study of SAP patients admitted to Peking University Third Hospital (2016–2020) within 48 h of onset. Fluid requirement was quantified by 48-hour rehydration volume. Using the Lasso algorithm, we screened predictive variables and trained five models: Gradient Boosting Decision Tree (GBDT), eXtreme Gradient Boosting (XGBoost), LightGBM, CatBoost, and multiple linear regression. Model performance was evaluated via mean absolute error (MAE), root mean square error (RMSE), R2, and fitting curves. The soft voting method was used to fuse the above five prediction models to improve the performance of the model. The optimal model was interpreted using SHapley Additive exPlanations (SHAP). To illustrate potential clinical use, ten randomly selected cases from the test set are presented.

Results

Among 308 included patients, 90% were allocated to the training set. Sixteen key variables were selected for prediction. Among the five machine learning algorithms used to build prediction models, the MAE and RMSE values of the XGBoost algorithm were the smallest and the R2 value was the closest to 1, which indicated that XGBoost was the best-performing model in our study. After model fusion, the model performance was further improved. SHAP analysis of the optimal model, XGBoost, revealed the relative importance of each predictor. To illustrate the model’s practical application, we present the predictions for 10 randomly selected test-set cases. The differences between the predicted and actual fluid volumes in these individual cases ranged from 31.07 to 329.80 mL, serving as concrete examples of how the model could be used at the bedside.

Conclusion

In this study, we developed the Fluid Requirement Predicting Model for SAP (FRPM-SAP), which can predict the specific amount of fluid loss in SAP patients. The predictive performance was good, demonstrating that the model has practical application for guiding clinicians in their assessment of 48-hour rehydration volume.