Exploring the use of the GLIM criteria to diagnose malnutrition in cancer inpatients

•The observed prevalence of malnutrition in the study was 26.1%, according to the subjective global assessment, whereas considering the various Global Leadership Initiative on Malnutrition combinations, the prevalence of malnutrition ranged from 3.9% to 30.0%. None of the Global Leadership Initiative on Malnutrition combinations reached adequate concurrent validity, with the subjective global assessment as the gold standard.•Combinations that included C-reactive protein indicated a higher prevalence of malnutrition and higher sensitivity values compared with combinations with serum albumin as an etiologic criterion.•The presence of malnutrition, according to the Global Leadership Initiative on Malnutrition criteria, was an independent predictor of complications in surgical patients. The Global Leadership Initiative on Malnutrition (GLIM) criteria establish a diagnosis of malnutrition based on the presence of at least one phenotypic and one etiologic criterion. This study aimed to assess the concurrent and predictive validity of the GLIM criteria in hospitalized cancer patients. This is an observational retrospective study, including 885 cancer patients, ages >18 y, admitted to a medical oncology inpatient unit between 2019 and 2020. All patients at risk for malnutrition according to the Nutritional Risk Screening 2002 score were assessed by the subjective global assessment (SGA) and 14 different combinations of the GLIM criteria. The SGA was considered the gold standard for assessing the concurrent validity of the GLIM combinations. For a subsample of patients with data available on inflammatory markers (n = 198), the serum albumin and C-reactive protein were included in the combinations as etiologic criteria. The predictive validity of the different combinations was tested using the occurrence of surgical complications as the clinical outcome. The sensitivity and specificity values were calculated to assess the concurrent validity, univariate and multivariate logistic regression models were used to test predictive validity. Adequate concurrent and predictive validity were determined as sensitivity and specificity values >80% and odds ratio values ≥2.0, respectively. The median age of the patients was 61.0 y (interquartile range = 51.0–70.0). Head and neck cancer was the prevailing diagnosis and 375 patients were at nutritional risk. According to the SGA, 173 (26.1%) patients were malnourished (SGA categories B or C) and the prevalence of malnutrition rang