Scholarlink

<p>Urosepsis is a severe infectious complication of retrograde intrarenal surgery (RIRS) that can progress to septic shock and become life-threatening. Early detection and prompt intervention are crucial for enhancing the prognosis of patients at risk for urosepsis. This study seeks to develop and verify a prediction model for post-RIRS urosepsis utilizing machine learning (ML) techniques. Electronic health record (EHR) data from 124 urosepsis patients and 1,786 non-urosepsis patients, all of whom had upper urinary tract stones and underwent RIRS, were analyzed. Participants were randomly allocated to a training cohort (80%) and a validation cohort (20%). Eight ML algorithms were employed to construct the models, including Gaussian Naive Bayes (GNB), Complement Naive Bayes, multilayer perceptron, AdaBoost, Gradient Boosting Decision Tree, logistic regression, LightGBM, and Random Forest. The predictive capability of each algorithm was evaluated by means of receiver operating characteristic (ROC) curves. Of the 78 clinical parameters collected, 41 were found to be statistically significant. Using LASSO regression, 7 variables were selected, including urine leukocytes (U-LEU), urine nitrite (U-NIT), urine pH, white blood cells (WBC), albumin (ALB), globulin (GLB), and alkaline phosphatase (ALP). Correlation analysis indicated no strong relationships among the parameters. GNB exhibited optimal performance in the validation cohort, achieving an area under the ROC curve (AUC) of 0.73 (95% CI: 0.58–0.89), with a sensitivity of 0.67 (95% CI: 0.58–0.76) and a specificity of 0.69 (95% CI: 0.64–0.73). The top five most important factors included U-NIT, GLB, ALP, WBC, and U-LEU. A ML model based on 7 clinical parameters was developed using EHR data, which aids in identifying high-risk urosepsis patients before RIRS. This approach provides a potentially valuable tool for preoperative urosepsis risk stratification of RIRS patients, facilitating the development of personalized clinical intervention strategies.</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

Explainable machine learning algorithms for preoperative risk stratification of urosepsis after retrograde intrarenal surgery

  • 期刊论文

  • 发表日期：

    2026年4月21日

查看全文
  • Zuheng Wang,
  • Yun Li,
  • Qianshi Jiang,
  • Xiao Li,
  • Chunmeng Wei,
  • Rongbin Zhou,
  • Wenhao Lu,
  • Fanchang Zeng,
  • Fubo Wang

摘要

Urosepsis is a severe infectious complication of retrograde intrarenal surgery (RIRS) that can progress to septic shock and become life-threatening. Early detection and prompt intervention are crucial for enhancing the prognosis of patients at risk for urosepsis. This study seeks to develop and verify a prediction model for post-RIRS urosepsis utilizing machine learning (ML) techniques. Electronic health record (EHR) data from 124 urosepsis patients and 1,786 non-urosepsis patients, all of whom had upper urinary tract stones and underwent RIRS, were analyzed. Participants were randomly allocated to a training cohort (80%) and a validation cohort (20%). Eight ML algorithms were employed to construct the models, including Gaussian Naive Bayes (GNB), Complement Naive Bayes, multilayer perceptron, AdaBoost, Gradient Boosting Decision Tree, logistic regression, LightGBM, and Random Forest. The predictive capability of each algorithm was evaluated by means of receiver operating characteristic (ROC) curves. Of the 78 clinical parameters collected, 41 were found to be statistically significant. Using LASSO regression, 7 variables were selected, including urine leukocytes (U-LEU), urine nitrite (U-NIT), urine pH, white blood cells (WBC), albumin (ALB), globulin (GLB), and alkaline phosphatase (ALP). Correlation analysis indicated no strong relationships among the parameters. GNB exhibited optimal performance in the validation cohort, achieving an area under the ROC curve (AUC) of 0.73 (95% CI: 0.58–0.89), with a sensitivity of 0.67 (95% CI: 0.58–0.76) and a specificity of 0.69 (95% CI: 0.64–0.73). The top five most important factors included U-NIT, GLB, ALP, WBC, and U-LEU. A ML model based on 7 clinical parameters was developed using EHR data, which aids in identifying high-risk urosepsis patients before RIRS. This approach provides a potentially valuable tool for preoperative urosepsis risk stratification of RIRS patients, facilitating the development of personalized clinical intervention strategies.