Scholarlink

Objective <p>Bloodstream infection (BSI) is a significant cause of mortality in patients with hematologic malignancies(HMs), particularly amid rising antibiotic resistance. This study aimed to analyze pathogen distribution, drug-resistance patterns and develop a novel predictive model for 30-day mortality in HM patients with BSIs.</p> Methods <p>A retrospective analysis of 231 HM patients with positive blood cultures was conducted. Logistic regression identified risk factors for 30-day mortality. Th1/Th2 cytokines were collected at BSI onset, with LASSO regression and restricted cubic spline analysis used to refine predictors. Seven machine learning(ML) algorithm (XGBoost, Logistic Regression, LightGBM, RandomForest, AdaBoost, GBDT and GNB) were trained using 10-fold cross-validation and model performance was evaluated with the ROC, calibration plots, decision and learning curves and the Shapley Additive Explanations (SHAP) analysis. The predictive model was developed by integrating Th1/Th2 cytokines with clinical features, aiming to enhance the accuracy of 30-day mortality prediction.</p> Results <p>Among the cohort, acute myeloid leukemia (38%) was the most common HM, while gram negative bacteria (64%) were the predominant pathogens causing BSI. Age, polymicrobial BSI, IL-4, IL-6 and AST levels were significant predictors of 30-day mortality. The Logistic Regression model achieved AUCs of 0.802, 0.792, and 0.822 in training, validation, and test cohorts, respectively, with strong calibration and clinical benefit shown in decision curves. SHAP analysis highlighted IL-4 and IL-6 as key predictors.</p> Conclusions <p>This study introduces a novel ML-based model integrating Th1/Th2 cytokines and clinical features to predict 30-day mortality in HM patients with BSIs, demonstrating strong performance and clinical applicability.</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

Machine learning-based prognostic model for bloodstream infections in hematological malignancies using Th1/Th2 cytokines

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年3月26日

查看全文
  • Qin Li,
  • Nan Lin,
  • Zuheng Wang,
  • Yuexi Chen,
  • Yuli Xie,
  • Xuemei Wang,
  • Jirui Tang,
  • Yuling Xu,
  • Min Xu,
  • Na Lu,
  • Yiqian Huang,
  • Jiamin Luo,
  • Zhenfang Liu,
  • Li Jing

摘要

Objective

Bloodstream infection (BSI) is a significant cause of mortality in patients with hematologic malignancies(HMs), particularly amid rising antibiotic resistance. This study aimed to analyze pathogen distribution, drug-resistance patterns and develop a novel predictive model for 30-day mortality in HM patients with BSIs.

Methods

A retrospective analysis of 231 HM patients with positive blood cultures was conducted. Logistic regression identified risk factors for 30-day mortality. Th1/Th2 cytokines were collected at BSI onset, with LASSO regression and restricted cubic spline analysis used to refine predictors. Seven machine learning(ML) algorithm (XGBoost, Logistic Regression, LightGBM, RandomForest, AdaBoost, GBDT and GNB) were trained using 10-fold cross-validation and model performance was evaluated with the ROC, calibration plots, decision and learning curves and the Shapley Additive Explanations (SHAP) analysis. The predictive model was developed by integrating Th1/Th2 cytokines with clinical features, aiming to enhance the accuracy of 30-day mortality prediction.

Results

Among the cohort, acute myeloid leukemia (38%) was the most common HM, while gram negative bacteria (64%) were the predominant pathogens causing BSI. Age, polymicrobial BSI, IL-4, IL-6 and AST levels were significant predictors of 30-day mortality. The Logistic Regression model achieved AUCs of 0.802, 0.792, and 0.822 in training, validation, and test cohorts, respectively, with strong calibration and clinical benefit shown in decision curves. SHAP analysis highlighted IL-4 and IL-6 as key predictors.

Conclusions

This study introduces a novel ML-based model integrating Th1/Th2 cytokines and clinical features to predict 30-day mortality in HM patients with BSIs, demonstrating strong performance and clinical applicability.