Early Recognition of Secondary Asthma Caused by Lower Respiratory Tract Infection in Children Based on Multi-Omics Signature: A Retrospective Cohort Study.

Abstract

Objective: To explore the types of pathogens causing lower respiratory tract infections (LTRIs) in children and construction of a predictive model for monitoring secondary asthma caused by LTRIs.

Methods: Seven hundred and seventy-five children with LTRIs treated from June 2017 to July 2024 were selected as research subjects. Bacterial isolation and culture were performed on all children, and drug sensitivity tests were conducted on the isolated pathogens; And according to whether the child developed secondary asthma during treatment, they were divided into asthma group (n = 116) and non-asthma group (n = 659); Using logistic regression model to analyze the risk factors affecting secondary asthma in children with LTRIs, and establishing machine learning (ie nomogram and decision tree) prediction models; Using ROC curve analysis machine learning algorithms to predict AUC values, sensitivity, and specificity of secondary asthma in children with LTRIs.

Results: 792 pathogenic bacteria were isolated from 775 children with LTRIs through bacterial culture, including 261 Gram positive bacteria (32.95%) and 531 Gram negative bacteria (67.05%). Logistic regression model analysis showed that Glycerophospholipids, Sphingolipids and radiomics characteristics were risk factors for secondary asthma in children with LTRIs (P < 0.05). The AUC, sensitivity, and specificity of nomogram prediction for secondary asthma in children with LTRIs were 0.817(95CI: 0.760-0.874), 82.3%, and 76.6%, respectively; The AUC of decision tree prediction for secondary asthma in children with LTRIs is 0.926(95% CI: 0.869-0.983), with a sensitivity of 96.7% and a specificity of 87.8%.

Conclusion: LTRIs in children are mainly caused by Staphylococcus aureus, Streptococcus pneumoniae, Staphylococcus epidermidis, Escherichia coli, Klebsiella pneumoniae, and Pseudomonas aeruginosa; In addition, machine learning combined with multi-omics prediction models has shown good ability in predicting LTRIs combined with asthma, providing a non-invasive and effective method for clinical decision-making.