Developing a Blood Biomarker Model for Predicting Multiple Infection Episodes Following Blunt Trauma
*Amy Tsurumi1, *Yok-Ai Que2, Colleen M. Ryan3, *Patrick J. Flaherty4, Ronald G. Tompkins5, *Marianna Almpani6, *Yashoda V. Dhole5, *Laura F. Goodfield5, *Laurence G. Rahme1
1Massachusetts General Hospital/Harvard Medical School/Shriners Hospitals for Children-Boston, Boston, MA; 2Bern University Hospital, Bern, Switzerland; 3Massachusetts General Hospital/Shriners Hospitals for Children-Boston, Boston, MA; 4University of Massachusetts, Amherst, Amherst, MA; 5Massachusetts General Hospital, Boston, MA; 6Massachusetts General Hospital/Harvard Medical School, Boston, MA
Objective: To employ a machine learning approach for developing blood biomarkers predictive of multiple independent infection episodes (MIIE) following severe blunt trauma.
Design: Secondary retrospective analysis of the Host Response to Injury Study (“Glue Grant”) prospective cohort study.
Setting: Four level-one trauma centers in the US.
Patients: 140 adult (16 years or older) blunt trauma patients (excluding penetrating), with early leukocyte samples (within 48 hrs post-injury), who developed a first infection at least two days after sample collection, and who remained in the study (did not die or were discharged) for at least 10 days.
Main Outcome Measures: Patients who developed MIIE during the course of recovery (39 cases) versus those who did not (91 controls)
Results: We identified a panel of 6 transcriptomic probe sets that were highly predictive of MIIE, with Area Under Receiver Operating Characteristic Curve (AUROC) [95% CI] of 0.89 [0.83-0.95]. This logistic model significantly outperformed various models based on clinical severity scores, including Acute Physiologic Assessment and Chronic Health Evaluation (APACHE) II, with AUROC [95% CI] of 0.62 [0.521-0.72], Injury Severity Score (ISS), with 0.61 [0.51-0.71], and New Injury Severity Score (NISS), with 0.60 [0.50-0.70]. Gene Ontology analyses of up- and down-regulated genes comparing control and hypersusceptible patients showed early alterations in various immune-related pathways, as expected.
Conclusions: Early blood biomarkers may be an effective tool for early triage of blunt trauma patients and serve as immunomodulation targets in the future. Given that clinical injury severity scores lacked the ability to sufficiently predict infections outcomes in this cohort, developing tools based on genomics may lead to the development of novel preventative and therapeutic approaches against infections.