Scholarlink

<p>Preoperative prediction of testicular viability in pediatric testicular torsion remains challenging, with current assessment largely dependent on symptom duration and intraoperative judgment. This multicenter study aimed to develop and externally validate an interpretable machine learning model using routinely available hematological parameters. Pediatric patients with surgically confirmed testicular torsion were enrolled from three tertiary hospitals. Patients from two centers (<i>n</i> = 269) constituted the pre-training cohort, randomly divided into training (<i>n</i> = 189) and internal validation (<i>n</i> = 80) sets. An independent cohort from the third center (<i>n</i> = 107) served as external validation. Demographic data, symptom duration, and admission laboratory parameters were collected. Feature selection combined Boruta and LASSO regression. Multiple machine learning models were developed and compared. Model performance was evaluated by discrimination, calibration, and decision curve analysis, with interpretability assessed using SHAP. Seven preoperative variables—age, symptom duration, white blood cell count, monocyte count, lymphocyte count, C-reactive protein, and platelet-to-lymphocyte ratio—were identified as robust predictors. In external validation, logistic regression achieved the highest discriminative performance (AUC = 0.927, 95% CI: 0.879–0.974), with balanced sensitivity (0.867) and specificity (0.839). CatBoost demonstrated comparable discrimination (AUC = 0.925). Logistic regression exhibited stable calibration and favorable net benefit across threshold probabilities. SHAP analysis identified symptom duration as the most influential contributor, followed by inflammatory markers, providing transparent global and individual-level explanations of model behavior. An interpretable machine learning model incorporating symptom duration and routine hematological parameters may support preoperative risk stratification of testicular viability in pediatric testicular torsion. The model may serve as a complementary tool for perioperative counseling and expectation management alongside standard clinical and surgical evaluation.</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

An interpretable machine learning model for preoperative risk stratification of testicular viability in pediatric testicular torsion: A multicenter study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月30日

查看全文
  • Hang Wu,
  • Xiao Pu,
  • Nannan Gu,
  • Xiaojing Zhu,
  • Ye Lu,
  • Jinlong Yang,
  • Haobo Zhu

摘要

Preoperative prediction of testicular viability in pediatric testicular torsion remains challenging, with current assessment largely dependent on symptom duration and intraoperative judgment. This multicenter study aimed to develop and externally validate an interpretable machine learning model using routinely available hematological parameters. Pediatric patients with surgically confirmed testicular torsion were enrolled from three tertiary hospitals. Patients from two centers (n = 269) constituted the pre-training cohort, randomly divided into training (n = 189) and internal validation (n = 80) sets. An independent cohort from the third center (n = 107) served as external validation. Demographic data, symptom duration, and admission laboratory parameters were collected. Feature selection combined Boruta and LASSO regression. Multiple machine learning models were developed and compared. Model performance was evaluated by discrimination, calibration, and decision curve analysis, with interpretability assessed using SHAP. Seven preoperative variables—age, symptom duration, white blood cell count, monocyte count, lymphocyte count, C-reactive protein, and platelet-to-lymphocyte ratio—were identified as robust predictors. In external validation, logistic regression achieved the highest discriminative performance (AUC = 0.927, 95% CI: 0.879–0.974), with balanced sensitivity (0.867) and specificity (0.839). CatBoost demonstrated comparable discrimination (AUC = 0.925). Logistic regression exhibited stable calibration and favorable net benefit across threshold probabilities. SHAP analysis identified symptom duration as the most influential contributor, followed by inflammatory markers, providing transparent global and individual-level explanations of model behavior. An interpretable machine learning model incorporating symptom duration and routine hematological parameters may support preoperative risk stratification of testicular viability in pediatric testicular torsion. The model may serve as a complementary tool for perioperative counseling and expectation management alongside standard clinical and surgical evaluation.