| Popis: |
Despite significant advances in pediatric burn care, bloodstream infections (BSIs) remain a compelling challenge during recovery. A personalized medicine approach for accurate prediction of BSIs before they occur would contribute to prevention efforts and improve patient outcomes.We analyzed the blood transcriptome of severely burned (total burn surface area (TBSA) ≥20%) patients in the multi-center Inflammation and Host Response to Injury ("Glue Grant") cohort. Our study included 82 pediatric (age16) patients, with blood samples at least three days before the observed BSI episode. We applied the least absolute shrinkage and selection operator (LASSO) machine learning algorithm to select a panel of biomarkers predictive of BSI outcome.We developed a panel of ten probe sets corresponding to six annotated genes (ARG2, CPT1A, FYB, ITCH, MACF1, and SSH2), two uncharacterized (LOC101928635, LOC101929599), and two unannotated regions. Our multi-biomarker panel model yielded highly accurate prediction (AUROC [95%CI]: 0.938 [0.881-0.981]) compared to models with TBSA (0.708 [0.588-0.824]) or TBSA and inhalation injury status (0.792 [0.676-0.892]). A model combining the multi-biomarker panel with TBSA and inhalation injury status further improved prediction (0.978 [0.941-1.000]).The multi-biomarker panel model yielded a highly accurate prediction of BSIs before their onset. Knowing patients' risk profile early will guide clinicians to take rapid preventative measures for limiting infections, promote antibiotic stewardship that may aid in alleviating the current antibiotic resistance crisis, shorten hospital length of stay, and burden on healthcare resources, reduce healthcare costs and significantly improve patients' outcomes. Additionally, the biomarkers' identity and molecular functions may contribute to developing novel preventative interventions. |
