Samenvatting

BACKGROUND & AIMS: The Global Leadership Initiative on Malnutrition (GLIM) criteria require validation in various clinical populations. This study determined the prevalence of malnutrition in people with cancer using all possible diagnostic combinations of GLIM etiologic and phenotypic criteria and determined the combinations that best predicted mortality and unplanned hospital admission within 30 days. METHODS: The GLIM criteria were applied, in a cohort of participants from two cancer malnutrition point prevalence studies (N = 2801), using 21 combinations of the phenotypic (≥5% unintentional weight loss, body mass index [BMI], subjective assessment of muscle stores [from PG-SGA]) and etiologic (reduced food intake, inflammation [using metastatic disease as a proxy]) criteria. Machine learning approaches were applied to predict 30-day mortality and unplanned admission. RESULTS: We analysed 2492 participants after excluding those with missing data. Overall, 19% (n = 485) of participants were malnourished. The most common GLIM combinations were weight loss and reduced food intake (15%, n = 376), and low muscle mass and reduced food intake (12%, n = 298). Machine learning models demonstrated malnutrition diagnosis by weight loss and reduced muscle mass plus either reduced food intake or inflammation were the most important combinations to predict mortality at 30-days (accuracy 88%). Malnutrition diagnosis by weight loss or reduced muscle mass plus reduced food intake was most important for predicting unplanned admission within 30-days (accuracy 77%). CONCLUSIONS: Machine learning demonstrated that the phenotypic criteria of weight loss and reduced muscle mass combined with either etiologic criteria were important for predicting mortality. In contrast, the etiologic criteria of reduced food intake in combination with weight loss or reduced muscle mass was important for predicting unplanned admission. Understanding the phenotypic and etiologic criteria contributing to the GLIM diagnosis is important in clinical practice to identify people with cancer at higher risk of adverse outcomes.