Principal component analysis of a swine injury model identifies multiple phenotypes in trauma

BACKGROUND: Trauma is the third leading cause of death in the United States and the primary cause of death for people between the ages of 1 year and 44 years. In addition to tissue damage, trauma may also activate an inflammatory state known as trauma-induced coagulopathy (TIC) that is associated with clotting malfunctions, acidemia, and end-organ dysfunction. Prior work has also demonstrated benefit to acknowledging the type and severity of endothelial injury, coagulation derangements, and systemic inflammation in the management of trauma patients. This study builds upon prior work by combining laboratory, metabolic, and clinical metrics into an analysis of trauma phenotypes, evolution of phenotypes over time after trauma, and significance of trauma phenotype on mortality. METHODS: Seventy 3-month-old female Yorkshire crossbred swine were randomized to injury and resuscitation groups. Principal component analysis (PCA) of longitudinal swine TEG data (Reaction time, Alpha-Angle, Maximum Amplitude, and Clot Lysis at 30 minutes), pH, lactate, and MAP was completed in R at baseline, 1 hour postinjury, 3 hours postinjury, 6 hours postinjury, and 12 hours postinjury. Subjects were compared by principal component factor scores to assess differences in survival, injury severity, and treatment group. RESULTS: Among injured animals, three phenotypes were observed at each time point. Five phenotypes were associated with differences in survival, and of these, four were associated with differences in injury severity. Phenotype alignment was not significantly different by treatment group. CONCLUSION: This application of PCA to a set of coagulation, hemodynamic, and organ perfusion variables has identified multiple evolving phenotypes after trauma. Some of these phenotypes may correlate with injury severity and may have implications for survival. Next steps include validating these findings over greater numbers of subjects and exploring other machine-learning techniques for phenotype identification. )Copyright (c) 2023Wolters Kluwer Health, Inc. All rights reserved.)