Development of an Infection Risk Index for Microbiome Targeted Intervention in Children at High-Risk of Multidrug-Resistant Bloodstream Infections
2019
Background Bloodstream infections (BSI) cause significant morbidity and mortality following pediatric hematopoietic stem cell transplantation (HSCT) and are increasingly from multidrug-resistant pathogens. Most of these BSI pathogens arise from the intestines. Increased evidence demonstrates commensal gut anaerobes maintain host resistance from enteric infection, while loss of commensals increases host susceptibility. Antibiotic use reduces gut anaerobes and allows a compensatory increased pathogen burden that precedes BSI; BSI risk is related to colonization burden. Recent reports suggest fecal microbiota transplantation (FMT) can displace an enteric pathogens and broad-spectrum antibiotic resistance genes during restoration of a healthy microbiome. However, the impact of intestinal microbiota on colonization, transmission, and invasive MDR infections is poorly understood in children undergoing HSCT. Methods Using metagenomic shotgun sequencing (MSS) data of longitudinal stool samples from high-risk children, including 63 undergoing HSCT with 166 samples preceding BSI onset (n = 23), we developed a metagenomic-based infection risk index (IRI) from key microbiome features (scaled Brady-Curtis distance relative to healthy controls and relative anaerobe and pathogen abundance). Traditional cart data analysis was compared to advanced machine learning algorithms that consider microbiome distribution, longitudinal sampling, and feature selection. Results A metagenomic-derived IRI from our cohort identified children having up to an 8-fold increased BSI risk. Evaluation of these key microbiome features as a diagnostic screening test by traditional cart analysis accurately identified patients at-risk for BSI with 91% sensitivity using its negative likelihood ratio. However, its specificity of 51% prohibits targeting interventions to those at greatest risk (Figure 1). Therefore, advanced machine learning algorithms were used for metagenomic-derived BSI prediction. While training results suggest overfitting, its performance promises clinical utility with a sensitivity of 92% and specificity of 84% warranting optimization and validation in HSCT patients (Figure 2). Analysis during heterologous FMT interventions in 5 symptomatic high-risk children (including 1 HSCT patient) revealed a concomitant pathogen-dominated gut microbiome harboring a high burden of broad-spectrum antibiotic resistance genes prior to each FMT. Following FMT, we confirmed restoration of a healthy gut microbiome with abundant anaerobes had a low metagenomic-derived IRI in all cases (Figure 3). Conclusions The application of advanced machine learning to shotgun metagenomic data provides the resolution required for accurate BSI prediction to identify HSCT patients at greatest risk preceding an illness providing new opportunity for early FMT intervention to prevent BSI.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
0
References
0
Citations
NaN
KQI