Machine learning-based prediction of inborn errors of immunity in patients with low IgE levels.

BACKGROUND

Low total serum IgE has emerged as a potential marker of inborn errors of immunity (IEI), but no predictive tool exists to stratify risk in pediatric patients.

METHODS

In this retrospective study, 677 children with IgE <2.5 IU/mL were analyzed. We handled missing data with mean-value imputation, applied SMOTE to address class imbalance, and conducted feature selection via ANOVA F-test and SelectKBest. Ten machine-learning models were trained and tuned using nested five-fold cross-validation (5 × 5 repeats). Primary evaluation metrics included sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC).

RESULTS

The Random Forest classifier achieved the highest performance (AUROC 0.86; sensitivity 0.81; specificity 0.75). Key predictors included lymphocyte, platelet, and neutrophil counts, mean platelet volume, and C-reactive protein.

CONCLUSION

Our data-driven framework accurately identifies children at risk for IEI using routine laboratory parameters. Prospective external validation and integration into clinical workflows are warranted to facilitate early diagnosis.