AUTHOR=Li Jing , Zhou Meng-yao , Li Yang , Wu Xue , Li Xin , Xie Xiao-li , Xiong Li-jing 

TITLE=Clinical prediction model by machine learning to determine the results of maternal dietary avoidance in food protein-induced allergic proctocolitis infants

JOURNAL=Frontiers in Pediatrics

VOLUME=Volume 13 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/pediatrics/articles/10.3389/fped.2025.1612076

DOI=10.3389/fped.2025.1612076

ISSN=2296-2360

ABSTRACT=ObjectiveThe objective of this study was to investigate the factors associated with the results of maternal dietary avoidance in infants diagnosed with Food Protein-Induced Allergic Proctocolitis (FPIAP). Additionally, we aimed to develop a predictive model using machine learning techniques to forecast the results of maternal dietary restrictions.MethodsThe clinical data of FPIAP infants were retrospectively analyzed. The FPIAP infants were divided into two groups based on the results of maternal dietary restriction, and an analysis was conducted to identify the influencing factors. Variable was selected by Lasso regression model. Classification models were built utilizing various machine learning algorithms including XGB Classifier, Logistic Regression, Random Forest Classifier, Ada Boost Classifier, KNeighbors Classifier, LGBM Classifier, Decision Tree Classifier, Gradient Boosting Classifier, Support Vector Classifier. The optimal algorithm was selected to construct the final prediction model.ResultsIn a retrospective cohort study of 693 children diagnosed with FPIAP, the remission rate associated with maternal dietary avoidance was 47.38%. The overall efficacy of hypoallergenic formula was 88.48%. Multivariate analysis identified several factors influencing the outcome of maternal dietary restriction, including age, disease duration, regurgitation, eczema, and neonatal history of hematochezia. Variables were selected and incorporated into multiple machine learning models. Among them, the logistic regression model demonstrated relatively high stability and was ultimately selected for modeling. The final model achieved an AUC of 0.743 in the test set and an accuracy of 0.699. The validation set's AUC was within 10% of the test set's value, indicating acceptable generalizability. The Hosmer-Lemeshow goodness-of-fit test confirmed that the logistic regression model fit the data well (P = 0.691 > 0.05). Finally, a nomogram was used to visualize the model's performance, and the Brier Score in the calibration curve was 0.210.ConclusionThis study provided a predictive model for formulating individualized diagnostic strategies of suspected FIPAP infants. However, due to the limitations of the lack of prospective dataset validation, future studies should further validate the clinical application potential of the predictive model to improve the diagnostic efficiency and quality of life of FPIAP.