EFETIVIDADE DE ALGORITMOS DE INTELIGÊNCIA ARTIFICIAL PARA PREDIÇÃO DE SEPSE EM ADULTOS DE UNIDADES DE TERAPIA INTENSIVA

REVISÃO DE ESCOPO

Abstract

The early clinical recognition of sepsis is one of the main health challenges today, especially those admitted to the Intensive Care Unit (ICU). With the advancement of machine learning through artificial intelligence, several models have emerged with the purpose of predicting sepsis in a timely manner, having seen this, the present review aims to identify and synthesize scientific evidence on the effectiveness of intelligence algorithms artificial to predict sepsis in ICU patients. A scoping review was performed on the Cumulative Index to Nursing and Allied Health Literature (CINAHL) databases; National Library of Medicine (MEDLINE/PubMed); Latin American and Caribbean Center for Health Information and Sciences (LILACS), Embase, Cochrane Library and Web of Science with blind matching, where 3,864 studies were identified and 17 of them met the study questions. Studies targeting sepsis or septic shock in the ICU were eligible for inclusion. Models with a reported area under the receiver operating characteristic metric (AUROC) curve were analyzed to identify the strongest contributors to the model and its performance. Substantial heterogeneity was observed across studies and across models and their settings, test index, and outcome. This review pointed out that, in retrospective data, machine learning and models can accurately predict the onset of sepsis in advance. Although they present alternatives to traditional technologies, the heterogeneity between the studies limits the evaluation of the combined results.