Application of physiological network mapping in the prediction of survival in critically ill patients with acute liver failure

Abstract

Reduced functional connectivity of physiological systems is associated with poor prognosis in critically ill patients. However, physiological network analysis is not commonly used in clinical practice and awaits quantitative evidence. Acute liver failure (ALF) is associated with multiorgan failure and mortality. Prognostication in ALF is highly important for clinical management but is currently dependent on models that do not consider the interaction between organ systems. This study aims to examine whether physiological network analysis can predict survival in patients with ALF. Data from 640 adult patients admitted to the ICU for paracetamol-induced ALF were extracted from the MIMIC-III database. Parenclitic network analysis was performed on the routine biomarkers using 28-day survivors as reference population and network clusters were identified for survivors and non-survivors using k-clique percolation method. Network analysis showed that liver function biomarkers were more clustered in survivors than in non-survivors. Arterial pH was also found to cluster with serum creatinine and bicarbonate in survivors compared with non-survivors, where it clustered with respiratory nodes indicating physiologically distinctive compensatory mechanism. Deviation along the pH-bicarbonate and pH-creatinine axes significantly predicts mortality independent of current prognostic indicators. These results demonstrate that network analysis can provide pathophysiologic insight and predict survival in critically ill patients with ALF.