An Approach to Rapidly Assess Sepsis Using Machine Learning Approach

Sepsis is a life-threatening condition and understanding the disease pathophysiology using host immune response biomarkers is critical for patient stratification. Lack of accurate sepsis endotyping impedes clinicians to make timely decisions alongside insufficiencies in appropriate sepsis management. The objective of this work is to demonstrate the potential feasibility of a data-driven validation model for supporting clinical decision to predict sepsis host-immune response. Herein, we used machine learning approach to determine the predictive potential of identifying sepsis host immune response for patient stratification by combining multiple biomarker measurement from a single plasma sample. Results were obtained using the following cytokines and chemokines IL-6, IL-8, IL-10, IP-10, TRAIL, PCT and CRP where the test dataset was 70%. Supervised machine learning algorithm naïve Bayes and decision tree algorithm showed promising accuracies of 96.64% and 94.64% respectively. Using unsupervised clustering algorithms, we are able to achieve silhouette score of positive 0.5. These promising findings indicate the proposed AI approach could be a valuable testing resource for promoting clinical decision making.