MODELING PHYSIOLOGICAL AND CYTOKINE DATA TO PREDICT OUTCOME
Dr. Shinichiro Kurosawa
United States of America
8 slide(s) – English – 2012-06-29
INTRODUCTION: Monitoring the course of a response to septic challenge either in the clinic or an experimental setting involves accumulating vast amounts of data that report host physiology, immune and coagulant status. We asked if it was possible to mathematically transform grouped physiology data and cyto/chemokine data obtained over time in a way that would provide a map reporting status (improvement vs deterioration) and consequently predict survival. As a test system, we used data from Papio baboons challenged with ribosome-inactivating Shiga toxins-1 and -2 from enterohemorrhagic E.coli. Intestinal infection with these bacteria and subsequent toxemia can induce hemolytic uremic syndrome, and are the primary cause of acute renal failure in otherwise healthy US children. METHODS & RESULTS: Physiology (n=24) and inflammation (n=17) parameters were grouped, normalized and analyzed without knowledge of survival outcome as a function of time using principal component analysis (PCA). PCA space analysis revealed distinct regions corresponding to survival or non-survival of the subject as well as toxin type. Major results: 1) Stx1 and Stx2 exert differing impacts on physiology and while physiology parameters separated survivors and non-survivors, the separation was late with little value for prognosis; 2) Cytokine time-map paths (PC1 vs PC2) separated survivors and non-survivors by 10 hours post-challenge giving confident prognosis by day 2-3 post-challenge. Associated with non-survival were G-CSF, IL-1Ra and IL-6 (inflammation); high ALT, AST, LDH, BUN, creatinine and thrombocytopenia (physiology). CONCLUSIONS: Mathematical transformation of large data groups over time provides maps that follow changes in host status toward healing or increased disease severity and non-survival.