Scholarlink

Background <p>Cytomegalovirus (CMV) reactivation in critically ill populations is associated with adverse outcomes. This study aimed to develop and validate an interpretable machine learning model to predict CMV reactivation within 28 days of ICU admission in sepsis patients, providing a risk stratification tool to optimize clinical management.</p> Methods <p>A retrospective cohort study was conducted using data from intensive care unit (ICU) patients meeting SEPSIS-3 criteria. Key clinical variables, including acute physiology and chronic health evaluation Ⅱ(APACHE Ⅱscore), CMV IgG titer, CD4<sup>+</sup> lymphocyte count, and corticosteroid use duration, were selected through univariate logistic regression. Seven machine learning models were developed, with the gradient boosting machine (GBM) demonstrating superior performance. Model interpretability was enhanced using Shapley Additive Explanations (SHAP) to identify critical predictors and visualize feature contributions.</p> Results <p>Among 221 patients, 19.0% experienced CMV reactivation. The test set AUC values for GBM, logistic regression (LR), neural network (NN), support vector machine (SVM), random forest (RF), adaptive boosting (AdaBoost) and k-nearest neighbors (KNN) models were recorded as 0.761, 0.684, 0.673, 0.659, 0.65, 0.624, and 0.623, respectively. GBM model achieved the highest AUC in the test set, with robust calibration and net benefit of the GBM model in the test set exceeded that of the other models at threshold probabilities ranging from 10% to 60%. SHAP analysis identified the APACHE II score, corticosteroid use, CMV IgG levels, and CD4 + lymphocyte count as the top four important predictors. Kaplan-Meier curves demonstrated significant stratification between high- and low-risk groups (<i>p</i> &lt; .05).</p> Conclusion <p>This interpretable machine learning model provides accurate CMV reactivation prediction and clear insights into contributing factors. It has the potential to enhance early risk stratification and guide targeted management strategies for sepsis patients in ICU settings.</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

Risk stratification for CMV reactivation in sepsis patients: development of an interpretable machine learning model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月22日

查看全文
  • Chengcheng Shen,
  • Yang Chen,
  • Tingting Pan,
  • Hongping Qu,
  • Ling Zhang,
  • Rui Tian,
  • Tong Wu,
  • Ruoming Tan

摘要

Background

Cytomegalovirus (CMV) reactivation in critically ill populations is associated with adverse outcomes. This study aimed to develop and validate an interpretable machine learning model to predict CMV reactivation within 28 days of ICU admission in sepsis patients, providing a risk stratification tool to optimize clinical management.

Methods

A retrospective cohort study was conducted using data from intensive care unit (ICU) patients meeting SEPSIS-3 criteria. Key clinical variables, including acute physiology and chronic health evaluation Ⅱ(APACHE Ⅱscore), CMV IgG titer, CD4+ lymphocyte count, and corticosteroid use duration, were selected through univariate logistic regression. Seven machine learning models were developed, with the gradient boosting machine (GBM) demonstrating superior performance. Model interpretability was enhanced using Shapley Additive Explanations (SHAP) to identify critical predictors and visualize feature contributions.

Results

Among 221 patients, 19.0% experienced CMV reactivation. The test set AUC values for GBM, logistic regression (LR), neural network (NN), support vector machine (SVM), random forest (RF), adaptive boosting (AdaBoost) and k-nearest neighbors (KNN) models were recorded as 0.761, 0.684, 0.673, 0.659, 0.65, 0.624, and 0.623, respectively. GBM model achieved the highest AUC in the test set, with robust calibration and net benefit of the GBM model in the test set exceeded that of the other models at threshold probabilities ranging from 10% to 60%. SHAP analysis identified the APACHE II score, corticosteroid use, CMV IgG levels, and CD4 + lymphocyte count as the top four important predictors. Kaplan-Meier curves demonstrated significant stratification between high- and low-risk groups (p < .05).

Conclusion

This interpretable machine learning model provides accurate CMV reactivation prediction and clear insights into contributing factors. It has the potential to enhance early risk stratification and guide targeted management strategies for sepsis patients in ICU settings.