AUTHOR=Zhang Kai , Zhang Shufang , Cui Wei , Hong Yucai , Zhang Gensheng , Zhang Zhongheng 

TITLE=Development and Validation of a Sepsis Mortality Risk Score for Sepsis-3 Patients in Intensive Care Unit

JOURNAL=Frontiers in Medicine

VOLUME=Volume 7 - 2020

YEAR=2021

URL=https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2020.609769

DOI=10.3389/fmed.2020.609769

ISSN=2296-858X

ABSTRACT=Abstract
Purpose: Many severity scores for outcome prediction are widely used for critically ill patients in the intensive care unit (ICU). However, none of them was developed for patients identified by sepsis-3 criteria. This study aimed to develop and validate a risk stratification score for mortality prediction in sepsis-3 patients.
Materials and Methods: The retrospective cohort study employed the Medical Information Mart for Intensive Care III (MIMIC III) database for model development. Sepsis patients were identified by Sepsis-3 criteria on day 1 of ICU entry and randomly assigned into development and validation dataset. We performed the Least Absolute Shrinkage and Selection Operator (LASSO) technique to select predictive variables, developed a sepsis mortality prediction model and associated risk stratification score. Model discrimination and calibration was compared with other traditional severity scores.
Results: A total of 4,601 patients fulfill sepsis-3 criteria were enrolled, the 30-day mortality was 19.5%. The score had good discrimination in development and validation sets (area under curve: 0.796 and 0.791). The calibration slope was 0.945 and Brier value was 0.128 as calculated in the validation set. The score divided patients according to mortality risk of low (3.4%), moderate (13.1%), high (28.1%), and very high (64.7%) in development dataset. Corresponding mortality in validation dataset were 3.5%, 13.4%, 30.6%, and 59.7%. Decision curve analysis showed the score always had a positive net benefit. 
Conclusions: The score, termed as Sepsis Mortality Risk Score (SMRS), has good discrimination and calibration, and allows stratification of patients according to mortality risk. However, further external validations are still needed.