Comparison of Respiratory Microbiome Disruption Indices to Predict VAP and VAE risk at LTACH Admission

Background: Healthcare exposure results in significant microbiome disruption, particularly in the setting of critical illness, which may contribute to risk for healthcare-associated infections (HAIs). Patients admitted to long-term acute-care hospitals (LTACHs) have extensive prior healthcare exposure and critical illness; significant microbiome disruption has been previously documented among LTACH patients. We compared the predictive value of 3 respiratory tract microbiome disruption indices—bacterial community diversity, dominance, and absolute abundance—as they relate to risk for ventilator-associated pneumonia (VAP) and adverse ventilator-associated events (VAE), which commonly complicate LTACH care. Methods: We enrolled 83 subjects on admission to an academic LTACH for ventilator weaning and performed longitudinal sampling of endotracheal aspirates, followed by 16S rRNA gene sequencing (Illumina HiSeq), bacterial community profiling (QIIME2) for diversity, and 16S rRNA quantitative PCR (qPCR) for total bacterial abundance. Statistical analyses were performed with R and Stan software. Mixed-effects models were fit to relate the admission MDIs to subsequent clinically diagnosed VAP and VAE. Results: Of the 83 patients, 19 had been diagnosed with pneumonia during the 14 days prior to LTACH admission (ie, “recent past VAP”); 23 additional patients were receiving antibiotics consistent with empiric VAP therapy within 48 hours of admission (ie, “empiric VAP therapy”); and 41 patients had no evidence of VAP at admission (ie, “no suspected VAP”). We detected no statistically significant differences in admission Shannon diversity, maximum amplicon sequence variant (ASV)–level proportional abundance, or 16S qPCR across the variables of interest. In isolation, all 3 admission microbiome disruption indices showed poor predictive performance, though Shannon diversity performed better than maximum ASV abundance. Predictive models that combined (1) bacterial diversity or abundance with (2) recent prior VAP diagnosis and (3) concurrent antibiotic exposure best predicted 14-day VAP (type S error < 0.05) and 30-day VAP (type S error < 0.003). In this cohort, VAE risk was paradoxically associated with higher admission Shannon diversity and lower admission maximum ASV abundance. Conclusions: In isolation, respiratory tract microbiome disruption indices obtained at LTACH admission showed poor predictive performance for subsequent VAP and VAE. But diversity and abundance models incorporating recent VAP history and admission antibiotic exposure performed well predicting 14-day and 30-day VAP.Disclosures: NoneFunding: None