Developing an Early Diagnostic Signature and Deciphering the Microbial-Host Dynamics in Lower Respiratory Tract Infection (LRTI) in Paediatric Intensive Care Unit (PICU) Patients

Lower respiratory tract infection (LRTI) is a leading cause of morbidity and mortality among children admitted to paediatric intensive care units (PICUs). Diagnosis is hampered by overlapping symptoms and limited sensitivity of conventional microbiology. We aimed to identify early diagnostic biomarkers by integrating microbial and host responses in paediatric LRTI. We re-analysed a metagenomic next-generation sequencing (mNGS) dataset from 261 PICU patients with acute respiratory failure, combining microbial and host transcriptomic profiles using differential expression, network, and machine-learning approaches. Candidate biomarkers were validated in an independent prospective cohort of 100 critically ill children (RASCALS), with non-bronchoscopic bronchoalveolar lavage (mini-BAL), blood cytokine profiling, and pathogen detection. Respiratory syncytial virus (RSV) and Haemophilus influenzae were the most enriched pathogens in LRTI cases. Host transcriptomics revealed activation of cytokine and chemokine signalling pathways. A seven-gene panel (IRF7, FFAR3, GZMB, FABP4, FN1, CXCL5, BCAR1) achieved high diagnostic accuracy, comparable to a published 14-gene model. In the RASCALS cohort, mini-BAL IL-1β, IL-4, and IL-8 classified bacterial LRTI with 65% accuracy, and blood IL-6 and TRAIL achieved 82% accuracy. Integrating host and microbial markers provides a feasible route to early, accurate diagnosis of paediatric LRTI. The identified 7-gene panel and cytokine markers could be translated into PCR- or ELISA-based bedside assays to support rapid clinical decision-making and antimicrobial stewardship in PICU. LRTIs are a leading cause of morbidity and mortality in critically ill children1, yet conventional diagnostics often cannot distinguish true infection from colonisation, driving broad-spectrum antimicrobial use. Previous studies have generally examined host or microbial factors in isolation, leaving host–microbe interactions in ventilated patients poorly understood. Mick et al.10 showed that a 14-gene host signature could separate bacterial from viral infections, but did not address host–microbe dynamics. This highlighted the need for integrated diagnostic models in PICU. This study advances paediatric LRTI diagnostics by extending the analysis of the microbial and host inflammatory response using a machine learning and network analysis approach. Re-analysing 261 critically ill children from Mick et al.10 we identified respiratory syncytial virus (RSV) and Haemophilus influenzae as dominant pathogens and delineated immune pathways associated with disease progression. From this, we derived a seven-gene host biomarker panel (IRF7, FFAR3, GZMB, FABP4, FN1, CXCL5, BCAR1) that matched the performance of the 14-gene model but with greater simplicity. Network analyses revealed inflammatory pathways linked to RSV co-infections and prolonged ventilation. Importantly, we validated these findings in the independent RASCALS cohort, where IL-1β, IL-4, IL-8, IL-6 and TRAIL were associated with bacterial LRTI diagnosis and clinical outcomes (ventilator-free days). Our results support moving from pathogen-only diagnostics to integrative host–microbe models, particularly in mechanically ventilated children. RSV and H. influenzae emerge as major drivers of paediatric LRTI. The seven-gene host panel, together with cytokine markers identified through our network analysis, could be developed into rapid PCR- or ELISA-based point-of-care assays to guide antimicrobial decisions. Multi-omic diagnostics may allow earlier, more precise LRTI diagnosis and support antimicrobial stewardship. Further studies should test performance across diverse patient populations.