Development and validation of a machine learning-based diagnostic system for 22 pediatric respiratory pathogens: a large-scale multicenter study

Respiratory tract infections (RTIs) are a significant cause of morbidity in children, caused by a wide range of pathogens. As treatment strategies depend on the causative pathogen, early and accurate diagnosis is crucial. We developed and validated an interpretable Pathogen Diagnostic System for Pediatric Respiratory Infections (Pathog-PDx), conducting a multicenter study involving 134,500 hospitalized children across three clinical centers and two databases. The model integrated 42 clinical and laboratory features from electronic health records (EHR) to enable early pathogen identification. Prospective validation was carried out on an independent cohort of 1338 children to assess the real-world applicability of the model. Pathog-PDx accurately distinguished 22 pathogen subtypes and outperformed conventional models in identifying both single and mixed infections. The model achieved high classification performance for key pathogens, such as influenza virus (AUC = 0.95; Sn: 0.88; Sp: 0.86), with mean AUCs of 0.88 for various pathogens of RTIs. The model has been deployed as a web-based decision support system, which is freely accessible at https://pathogpdx.zzu.edu.cn. Altogether, Pathog-PDx represents a potential tool for the early identification of respiratory tract pathogens in pediatric patients, which can provide actionable predictions ahead of conventional test results to guide timely and targeted therapy.