Can Artificial Intelligence and Machine Learning Techniques improve the ability to detect Sepsis and Septic Shock. A retrospective study of 218,562 adult patients in a university hospital

The use of Artificial Intelligence (AI) and Machine Learning (ML) techniques has improved a sepsis (SE) and septic shock (SS) early detection compared with traditional rules according to recent retrospective, prospective and meta-analysis (1). Develop predictive models using algorithms based on AI-M...

Full description

Saved in:

Bibliographic Details
Published in:	Journal of critical care Vol. 81; p. 154684
Main Authors:	González, Veronica Lisseth Hernandez, Alaminos, Maria Ortega, Mir, Antonia Socias, Borges, Marcio, Aranda, Maria, Castillo, Alberto, Pruenza, Cristina, Serrano, Ana, Martin, Elisa, Estrada, Victor
Format:	Journal Article
Language:	English
Published:	Philadelphia Elsevier Inc 01-06-2024 Elsevier Limited
Subjects:	Algorithms Artificial intelligence Artificial Intelligence (AI) Electronic health records Intensive care Machine learning Machine Learning (ML) Sepsis Sepsis (SE) Septic Shock (SS) Sepsis (SE) Artificial Intelligence (AI) Machine Learning (ML) Septic Shock (SS)
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	The use of Artificial Intelligence (AI) and Machine Learning (ML) techniques has improved a sepsis (SE) and septic shock (SS) early detection compared with traditional rules according to recent retrospective, prospective and meta-analysis (1). Develop predictive models using algorithms based on AI-ML techniques and compare with fixed rules for SE/SS detection, assessing whether these new models improve predictive capability. We carried out an observational, retrospective non interventional study developed in our University General Hospital. The period assessed was from January 2014 to October 2018. The diagnosis and validation of each SE or SS case were made prospectively by the clinical experts of the Multidisciplinary Sepsis Unit (MSU). We used a Sepsis 2 definition. We developed AI-ML techniques from historical data from the Electronic Health Record (EHR). The structured variables were obtained from different data sources and from non-structured text from the Triage and Emergency Department (ED). The Mann-Whitney-Wilcoxon test was used to identify statistically significant clinical and analytical variables, as well as wrapper techniques, with a significance level of 0.01 and to obtain relevant unstructured data using a Natural Language Processing (NLP) techniques. A total of 815,170 records of the EHR have been assessed. We included 218.562 adult patients from all hospital departments. We divided into 2 groups: 1) with SE/SS were 9301 (4.6%); and 2) 209,261 (95.4%) who did NOT have sepsis (NSE). A total of 3927 variables have been extracted from the different data sources. By relevance and after being validated by the UMS team, 244 (6.2%) both structured and unstructured variables were associated with the detection of SE/SS. Within the structured variables, we identified many that are not blackened by the classic scorings of SE/SS, such as hemoglobin or eosinopenia. We developed about 30 different predictive models for SE/SS detection, using fixed rules individually, using only AI-ML based algorithms or the combination of fixed rules with AI-ML techniques. The best model using only fixed rules was the one using the Sepsis.2 criterion, while the best model using AI-ML techniques was called BISEPRO and was a combination of SEPSIS.2 with AI-ML techniques. In this retrospective study including adult patients in all areas of a hospital the use of AI-ML based techniques was significantly superior for the detection of SE/SS. 1. Lucas M. Fleuren; Patrick Thoral, Duncan Shillan, et all. Machine learning in intensive care medicine: ready for take-off. Intensive Care Medicine. Jul 2020.46:1486–1488
ISSN:	0883-9441 1557-8615
DOI:	10.1016/j.jcrc.2024.154684