• Jessimara Ribeiro Henrique, Ramon Gonçalves Pereira, Rosaria Silva Ferreira, Heather Keller, Marian de Van der Schueren, Maria Cristina Gonzalez, Wagner Meira, and CorreiaMaria Isabel Toulson DavissonMITDDepartment of Surgery, Medical School, Federal University of Minas Gerais, Belo Horizonte, Minas Gerais, Brazil. Electronic address: isabel_correia@uol.com.br..
• Food Sciences Post Graduation Program, Pharmacy School, Federal University of Minas Gerais, Belo Horizonte, Minas Gerais, Brazil.
• Nutrition. 2020 Nov 1; 79-80: 110961.

ObjectivesThe Global Leadership Initiative on Malnutrition (GLIM) was proposed to provide a common malnutrition diagnostic framework. The aims of this study were to evaluate the applicability and validity of the GLIM and use machine-learning techniques to help provide the best malnutrition-related variables/combinations to predict complications in patients undergoing gastrointestinal (GI) surgeries.MethodThis was a prospective cohort study enrolling surgical patients with GI diseases. Malnutrition prevalence was classified by the GLIM, subjective global assessment (SGA), and various anthropometric parameters. The various combination of the phenotypic criteria generated 10 different models. Sensibility (SE) and specificity (SP) were calculated using SGA as the reference criterion. Machine-learning approaches were used to predict complications. P < 0.05 was set as statistically significant.ResultsWe evaluated 206 patients. Half of the patients were malnourished according SGA, and 16.5% had postoperative complications. The prevalence of malnutrition using GLIM varied from 10.7% to 41.3% among the whole population, 11.7% and 43.6% in the elderly, from 0 to 24% in overweight non-obese and from 0 to 19.6% in obese patients. SE and SP values varied between 61.2% and 100% and 55.3% and 98.1%, respectively, for the general population. Machine-learning models indicated that midarm circumference, one of the GLIM models, and midarm muscle area were the most relevant criteria to predict complications.ConclusionsThe various GLIM combinations provided different rates of malnutrition according to the population. Machine-learning techniques supported the use of common single variables and one GLIM model to predict postoperative complications.Copyright © 2020 Elsevier Inc. All rights reserved.

15 keywords...

hide…