Integrating intestinal microbiome and urinary metabolome data to predict secondary infection in critically ill patients.

backgroundSecondary infection (SI), including ventilator-associated pneumonia (VAP) and bloodstream infection (BSI), represents a major complication in critically ill patients. Current clinical risk stratification approaches prove inadequate for timely and precise identification of at-risk patients. This study identifies intestinal microbiome and urinary metabolome characteristics (“multi-omics data”) associated with SI occurrence, investigates convergence of the respiratory microbiome with the intestinal microbiome, and determines whether multi-omics integration enhances prognostic discrimination for patients at risk of developing SI.

methodsWe analyzed data from mechanically ventilated patients from two cohorts: University Hospital Cologne (UHC), Germany, and Columbia University Medical Center (CUMC), New York, United States. The core dataset (n = 88; 64 UHC and 24 CUMC) assessed multi-omics integration for SI prediction, with an UHC subset (n = 55) providing more comprehensive clinical and microbiome characterization. Baseline intestinal and respiratory microbiome, as well as urinary metabolome data were collected within 48 h of intensive care unit admission or intubation using 16 S ribosomal ribonucleic acid (rRNA) sequencing and nuclear magnetic resonance (NMR) spectroscopy. SI was defined as new-onset BSI or VAP occurring ≥ 48 h after enrollment. Regression and classification models compared clinical-only approaches with integrated multi-omics models using model selection criteria, area under the curve (AUC), and Matthews correlation coefficients.

resultsSI occurred in 28% of patients, with prior antibiotic exposure associated with SI (84% vs. 41%, q < 0.01; odds ratio 2.57, p = 0.17). SI patients exhibited significantly lower baseline intestinal microbial diversity (Shannon diversity, 1.96 vs. 3.47, p < 0.01) and greater Enterococcus abundance (46% vs. 11%, q = 0.02), with similar patterns observed in the respiratory microbiome. Urinary NMR analysis identified metabolites mapping to features at 0.935 ppm (2-oxoisocaproate, isoleucine) in the core dataset, and at 8.025 ppm (quinolinate) in the UHC subset as elevated in SI patients. Multi-omics models demonstrated modest but consistent improvement over clinical-only models (AUC: 0.75 vs. 0.64).

conclusionsSI susceptibility in critically ill patients associates with underlying clinical severity, prior antibiotic exposure, and microbiota disruption. Multi-omics integration yielded consistent predictive improvement, supporting prospective validation as a proof-of-concept approach for early SI risk stratification.