Research code: --
Mohsen Ebrahimi 
, Maryam Ziaei , Lohrasb Taheri , Samaneh Abiri , Reza Akhavan , Amir Hossein Pourdavood , Ali Ravaghi , Mansour Deilami
, Maryam Ziaei , Lohrasb Taheri , Samaneh Abiri , Reza Akhavan , Amir Hossein Pourdavood , Ali Ravaghi , Mansour Deilami
Abstract: (6 Views)
Introduction: The 2020 World Society of Emergency Surgery (WSES) Liver Injury Classification is one of the latest guidelines that combines the severity of anatomical injury with hemodynamic status, and in this study its performance is compared with modern machine learning methods. To understand the variables of interest, the SHapley Additive exPlanations method was used.
Methods: This secondary study analyzed retrospective study. Eight supervised machine learning algorithms including XGBoost, LightGBM, CatBoost, Random Forest, and Logistic Regression were trained to predict each outcome, and their performance was compared with the WSES rule using AUC-ROC. Interpretation of the models was performed with SHAP values.
Results: ML models performed better for predicting the need for surgery (CatBoost AUC 0.833 vs 0.573, Δ +0.260, p < 0.001), hepatic complications (Logistic Regression AUC 0.722 vs 0.549, Δ +0.173, p = 0.012), serious complications (CatBoost AUC 0.809 vs 0.622, Δ +0.187, p < 0.001), and angioembolization/re-laparotomy (LightGBM AUC 0.833 vs 0.598, Δ +0.235, p < 0.001). SHAP analysis identified hemodynamic instability, injury severity scores, and associated abdominal injuries as the most important predictors.
Conclusion: The WSES 2020 liver injury classification predicts in-hospital mortality well but performs worse than ML models in predicting the need for surgery and complications.
Methods: This secondary study analyzed retrospective study. Eight supervised machine learning algorithms including XGBoost, LightGBM, CatBoost, Random Forest, and Logistic Regression were trained to predict each outcome, and their performance was compared with the WSES rule using AUC-ROC. Interpretation of the models was performed with SHAP values.
Results: ML models performed better for predicting the need for surgery (CatBoost AUC 0.833 vs 0.573, Δ +0.260, p < 0.001), hepatic complications (Logistic Regression AUC 0.722 vs 0.549, Δ +0.173, p = 0.012), serious complications (CatBoost AUC 0.809 vs 0.622, Δ +0.187, p < 0.001), and angioembolization/re-laparotomy (LightGBM AUC 0.833 vs 0.598, Δ +0.235, p < 0.001). SHAP analysis identified hemodynamic instability, injury severity scores, and associated abdominal injuries as the most important predictors.
Conclusion: The WSES 2020 liver injury classification predicts in-hospital mortality well but performs worse than ML models in predicting the need for surgery and complications.
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC-By-NC), which permits use, distribution, and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.
Contact Information
© 2015 All Rights Reserved | Development Strategies in Medical Education
Designed & Developed by : Yektaweb
Development Strategies in Medical Education by http://dsme.hums.ac.ir/ is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.