Scholarlink

Purpose <p>Postoperative ileus (POI) represents a frequent complication after posterior thoracolumbar fracture surgery. This study aimed to identify POI risk factors and construct predictive models enabling early identification and targeted intervention of vulnerable individuals.</p> Methods <p>A literature review were conducted to quantify POI incidence and establish evidence-based predictors for variable selection. Subsequently, a retrospective cohort from the Second Affiliated Hospital of Wenzhou Medical University was used for model development and internal validation. Feature selection incorporated the least absolute shrinkage and selection operator (LASSO) regression with multivariate logistic regression, followed by predictive modeling using five distinct algorithms: logistic regression (LR), random forest (RFC), categorical boosting (CatBoost), extreme gradient boosting (XGB), and light gradient boosting machine (LGBM). Model interpretability was augmented through SHapley Additive exPlanations (SHAP) analysis.</p> Results <p>The literature review encompassed 20 eligible studies, determining a pooled POI incidence of 8.9% (95% CI: 6.5–11.3). The training and testing cohorts comprised 493 and 210 patients, respectively. Among all models, CatBoost achieved peak accuracy (0.867) and specificity (0.960), with AUROC values of 0.906 (95% CI: 0.868–0.941) in the training set and 0.772 (95% CI: 0.665–0.860) in the testing set. SHAP analysis identified surgery duration, postoperative 24&#xa0;h NRS score ≥ 3, and the number of levels involved in surgery as the top three predictors of POI.</p> Conclusion <p>By integrating evidence synthesis with machine learning, this study establishes a clinically applicable POI prediction framework. The CatBoost model showed strong predictive performance and, when combined with a risk web calculator, offers a practical tool for early identification of high-risk patients, ultimately supporting precision medicine initiatives in spinal trauma care.</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

Development and comparison of machine learning models for predicting postoperative ileus after posterior thoracolumbar fracture surgery

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Yantong Zhang,
  • Renji Zheng,
  • Huiqing Gao,
  • Yingfeng Zhou,
  • Jun Li,
  • Liqiong Chen

摘要

Purpose

Postoperative ileus (POI) represents a frequent complication after posterior thoracolumbar fracture surgery. This study aimed to identify POI risk factors and construct predictive models enabling early identification and targeted intervention of vulnerable individuals.

Methods

A literature review were conducted to quantify POI incidence and establish evidence-based predictors for variable selection. Subsequently, a retrospective cohort from the Second Affiliated Hospital of Wenzhou Medical University was used for model development and internal validation. Feature selection incorporated the least absolute shrinkage and selection operator (LASSO) regression with multivariate logistic regression, followed by predictive modeling using five distinct algorithms: logistic regression (LR), random forest (RFC), categorical boosting (CatBoost), extreme gradient boosting (XGB), and light gradient boosting machine (LGBM). Model interpretability was augmented through SHapley Additive exPlanations (SHAP) analysis.

Results

The literature review encompassed 20 eligible studies, determining a pooled POI incidence of 8.9% (95% CI: 6.5–11.3). The training and testing cohorts comprised 493 and 210 patients, respectively. Among all models, CatBoost achieved peak accuracy (0.867) and specificity (0.960), with AUROC values of 0.906 (95% CI: 0.868–0.941) in the training set and 0.772 (95% CI: 0.665–0.860) in the testing set. SHAP analysis identified surgery duration, postoperative 24 h NRS score ≥ 3, and the number of levels involved in surgery as the top three predictors of POI.

Conclusion

By integrating evidence synthesis with machine learning, this study establishes a clinically applicable POI prediction framework. The CatBoost model showed strong predictive performance and, when combined with a risk web calculator, offers a practical tool for early identification of high-risk patients, ultimately supporting precision medicine initiatives in spinal trauma care.