Application of artificial intelligence in conjunction with clinical laboratory indicators to aid decision-making for surgical or conservative treatment of pediatric intestinal obstruction.

Abstract

Background: Management of pediatric intestinal obstruction remains clinically challenging, particularly regarding the selection between surgical and conservative approaches. This study aimed to develop artificial intelligence (AI) models to support treatment decision-making.

Methods: A retrospective analysis was conducted on clinical data from pediatric intestinal obstruction patients. The dataset was split via stratified sampling (70% training/ 30% test), preserving outcome distribution. Predictive models incorporating clinical indicators were developed using machine learning, with evaluation metrics including accuracy, F1-score, Kappa value, positive predictive value (PPV), negative predictive value (NPV), precision-recall curves, calibration plots and decision curve analysis (DCA).

Results: Among 765 pediatric patients, 425 responded to conservative treatment while 340 required surgery. The Random Forest model demonstrated optimal performance in the test cohort (area under the curve: 0.953; sensitivity: 0.879; specificity: 0.901; accuracy: 0.892; F1-score: 0.878; Kappa value: 0.780; PPV: 0.878; NPV: 0.905). Calibration, precision-recall, and DCAs indicated favorable clinical applicability.

Conclusion: Machine learning integration with clinical indicators shows potential as a decision-support tool for selecting surgical or conservative treatment in pediatric intestinal obstruction.