Machine Learning Model Predicts Pediatric Late Recurrent Intussusception Across Two Centers

A machine learning model developed with the LightGBM algorithm can accurately predict the risk of late recurrent intussusception in pediatric patients who have undergone successful non-surgical reduction, according to a two-center retrospective study published in BMC Medical Informatics and Decision Making.

Late recurrent intussusception in children presents a notable clinical challenge because its etiology remains unclear and the recurrence rate is high. Prior to this research, systematic studies focusing on predicting recurrence were lacking, creating an unmet need for objective tools that can assist clinicians in identifying high-risk patients early.

To build the predictive framework, researchers analyzed retrospective data from 1,671 children treated across two medical centers. All included children had presented with an initial episode of intussusception that was successfully managed using pneumatic reduction. The investigators collected 24 clinical and laboratory variables for each patient to identify potential predictive markers.

Feature selection was conducted using the Boruta algorithm together with recursive feature elimination, narrowing the initial candidate pool down to nine key predictive features. The internal dataset was randomly partitioned into a training set and an internal validation set at an 8:2 ratio, where ten different machine learning algorithms were trained to identify the optimal architecture.

Among all evaluated algorithms, LightGBM demonstrated the highest performance and was selected as the final predictive model. In the internal validation cohort, the model achieved a receiver operating characteristic area under the curve of 0.972 (95% CI, 0.947–0.996) and a Brier score of 0.045 (95% CI, 0.025–0.066). When evaluated in an independent external validation set, the model yielded an identical area under the curve of 0.972 (95% CI, 0.961–0.983) and a Brier score of 0.047 (95% CI, 0.035–0.060).

SHapley Additive exPlanations analysis clarified the contributions of the nine key features and the directions of their associations with late recurrence risk, while decision curve analysis showed potential clinical net benefit across a range of threshold probabilities. To support bedside decision-making, the investigators deployed an open web-based risk calculator on the Shiny platform that enables rapid estimation of recurrence probability for individual children.

Subject of Research: Machine learning prediction and risk calculation for pediatric late recurrent intussusception

Article Title: Development of a machine learning–based prediction model and an online risk calculator for late recurrent intussusception in children: a two-center retrospective study

Article References: Yang, M., Rao, X., Wang, L., Huang, A., Zhang, P., Chen, J., Rao, X., & Duan, S. (2026). Development of a machine learning–based prediction model and an online risk calculator for late recurrent intussusception in children: a two-center retrospective study. BMC Medical Informatics and Decision Making. https://doi.org/10.1186/s12911-026-03886-3

Image Credits: AI Generated

DOI: 10.1186/s12911-026-03886-3

Keywords: intussusception, pediatric intussusception, late recurrent intussusception, machine learning, LightGBM, pneumatic reduction, clinical decision support, risk calculator, Shiny platform, ROC-AUC, Brier score, decision curve analysis, SHAP analysis, Boruta algorithm, recursive feature elimination, two-center study, retrospective cohort