Scholarlink

Background <p>Burn injuries are a major cause of death, and infections with antibiotic‑resistant Staphylococcus aureus further increase mortality in these patients. There is a need for accurate and interpretable tools to predict the risk of death in this high‑risk group.</p> Materials and methods <p>This retrospective study included 222 burn patients with confirmed antibiotic‑resistant Staphylococcus aureus infections admitted to Velayat center between 2021 and 2023. Clinical and demographic data (age, total body surface area burned, length of hospital stay, sex, ventilation, inhalation injury, underlying diseases, and outcome) were collected. Seven machine learning models (XGBoost, Random Forest, Naïve Bayes, Logistic Regression, SVM, KNN, and Decision Tree) were developed and tuned using 5‑fold cross‑validation, then evaluated in an independent validation set using AUC, accuracy, precision, recall, and F1‑score, with additional 10‑fold cross‑validation. Model interpretability was assessed using SHapley Additive exPlanations (SHAP).</p> Results <p>The mean age of patients was 39.91 ± 14.39 years. Non‑survivors had larger burned areas and shorter hospital stays than survivors. Among the seven models, XGBoost achieved the best performance, with an AUC of 0.8704 and an accuracy of 0.8889 in the validation set. SHAP analysis identified total body surface area, inhalation injury, age, hospital stay, underlying disease, and ventilation as the main predictors of in‑hospital mortality and provided individualized explanations of predictions.</p> Conclusion <p>This study introduces an interpretable XGBoost‑based model for predicting in‑hospital mortality in burn patients with antibiotic‑resistant Staphylococcus aureus infections in Northern Iran. The model’s strong performance and explainability suggest that it can complement traditional scoring systems and help clinicians make transparent, patient‑specific decisions in similar settings.</p> Graphical Abstract <p></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

Prediction of death in burn patients infected with antibiotic-resistant Staphylococcus aureus using machine learning based techniques

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月12日

查看全文
  • Erfan Naseri,
  • Mojtaba Hedayati Ch,
  • Mohmmadreza Mobayen,
  • Hamid Sedighian,
  • Abbas Ali Imani Fooladi

摘要

Background

Burn injuries are a major cause of death, and infections with antibiotic‑resistant Staphylococcus aureus further increase mortality in these patients. There is a need for accurate and interpretable tools to predict the risk of death in this high‑risk group.

Materials and methods

This retrospective study included 222 burn patients with confirmed antibiotic‑resistant Staphylococcus aureus infections admitted to Velayat center between 2021 and 2023. Clinical and demographic data (age, total body surface area burned, length of hospital stay, sex, ventilation, inhalation injury, underlying diseases, and outcome) were collected. Seven machine learning models (XGBoost, Random Forest, Naïve Bayes, Logistic Regression, SVM, KNN, and Decision Tree) were developed and tuned using 5‑fold cross‑validation, then evaluated in an independent validation set using AUC, accuracy, precision, recall, and F1‑score, with additional 10‑fold cross‑validation. Model interpretability was assessed using SHapley Additive exPlanations (SHAP).

Results

The mean age of patients was 39.91 ± 14.39 years. Non‑survivors had larger burned areas and shorter hospital stays than survivors. Among the seven models, XGBoost achieved the best performance, with an AUC of 0.8704 and an accuracy of 0.8889 in the validation set. SHAP analysis identified total body surface area, inhalation injury, age, hospital stay, underlying disease, and ventilation as the main predictors of in‑hospital mortality and provided individualized explanations of predictions.

Conclusion

This study introduces an interpretable XGBoost‑based model for predicting in‑hospital mortality in burn patients with antibiotic‑resistant Staphylococcus aureus infections in Northern Iran. The model’s strong performance and explainability suggest that it can complement traditional scoring systems and help clinicians make transparent, patient‑specific decisions in similar settings.

Graphical Abstract