Scholarlink

<p>Urosepsis is one of the most severe complications after percutaneous nephrolithotomy (PNL). This study aimed to develop and validate a preoperative machine learning (ML) model for predicting post-PNL urosepsis using electronic medical record (EMR) data. In this retrospective study, we included 360 patients with urosepsis and 2,636 without urosepsis. Multimodal clinical parameters were collected, including demographic data, admission vital signs, imaging reports, and laboratory results. The cohort was randomly split into a training set (80%) and a validation set (20%). Eight ML algorithms were used to build prediction models, with performance evaluated via receiver operating characteristic (ROC) curves. In the training set, the LightGBM, RF, and XGBoost models showed the highest accuracy, each achieving an Area Under the ROC Curve (AUC) of 1.00. In contrast, the LR model performed best in the validation set (AUC = 0.77, 95% CI: 0.64–0.91). The key predictors for the LR model included heart rate (HR), serum creatinine (Scr), venous thromboembolism (VTE), urine leukocytes (U-LEU), and body mass index (BMI), which were part of 19 total predictive parameters. In conclusion, we developed an ML model using 19 preoperative EMR-derived parameters to predict post-PNL urosepsis risk. This model may help clinicians identify high-risk patients preoperatively, enabling timely interventions to reduce PNL-associated mortality.</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

Preoperative machine learning algorithm for predicting urosepsis after percutaneous nephrolithotomy using EMR data

  • 期刊论文

  • 发表日期：

    2025年12月22日

查看全文
  • Bin Liang,
  • Baofei Tan,
  • Zuheng Wang,
  • Kuan Pang,
  • Guan Chen,
  • Chunhua Liao,
  • Beiyuan Huang,
  • Zhenqiang Zhang,
  • Xingze Liu,
  • Qiang Ling,
  • Wenhao Lu,
  • Xiao Li,
  • Fubo Wang,
  • Guijian Pang

摘要

Urosepsis is one of the most severe complications after percutaneous nephrolithotomy (PNL). This study aimed to develop and validate a preoperative machine learning (ML) model for predicting post-PNL urosepsis using electronic medical record (EMR) data. In this retrospective study, we included 360 patients with urosepsis and 2,636 without urosepsis. Multimodal clinical parameters were collected, including demographic data, admission vital signs, imaging reports, and laboratory results. The cohort was randomly split into a training set (80%) and a validation set (20%). Eight ML algorithms were used to build prediction models, with performance evaluated via receiver operating characteristic (ROC) curves. In the training set, the LightGBM, RF, and XGBoost models showed the highest accuracy, each achieving an Area Under the ROC Curve (AUC) of 1.00. In contrast, the LR model performed best in the validation set (AUC = 0.77, 95% CI: 0.64–0.91). The key predictors for the LR model included heart rate (HR), serum creatinine (Scr), venous thromboembolism (VTE), urine leukocytes (U-LEU), and body mass index (BMI), which were part of 19 total predictive parameters. In conclusion, we developed an ML model using 19 preoperative EMR-derived parameters to predict post-PNL urosepsis risk. This model may help clinicians identify high-risk patients preoperatively, enabling timely interventions to reduce PNL-associated mortality.