Oral Presentation Clinical Oncology Society of Australia Annual Scientific Meeting 2022

Machine learning-based prediction of adverse outcomes from malnutrition in people with cancer (#105)

Nicole Kiss 1 , Belinda Steer 2 , Marian de van der Schueren 3 , Jenelle Loeliger 2 , Roohallah Alizadehsani 4 , Lara Edbrooke 5 , Irene Deftereos 6 , Erin Laing 2 , Abbas Khosravi 4
  1. Institute for Physical Activity and Nutrition, Deakin University, Burwood, Victoria, Australia
  2. Nutrition and Speech Pathology Department, Peter MacCallum Cancer Centre, Melbourne, Victoria, Australia
  3. Department of Nutrition, Dietetics and Lifestyle, HAN University of Applied Sciences, Nijmegen, Netherlands
  4. Institute for Intelligent Systems Research adn Innovation, Deakin University, Geelong, Victoria, Australia
  5. Department of Health Services Research, Peter MacCallum Cancer Centre, Melbourne, Victoria, Australia
  6. Department of Surgery, Western Health and the University of Melbourne, Parkville, Victoria, Australia

Background: Equipment to assess muscle mass (MM) is not available in all health services. There is limited evidence regarding whether excluding MM from the Global Leadership Initiative on Malnutrition (GLIM) criteria affects the ability to predict adverse outcomes. This study used machine learning algorithms to determine which GLIM criteria combinations were most important for the prediction of adverse events with and without inclusion of MM.

Methods: In participants from two cancer malnutrition point prevalence studies, we applied the GLIM criteria with and without MM. Phenotypic criteria were assessed using > 5% unintentional weight loss, body mass index, and subjective assessment of muscle stores from the Patient Generated-Subjective Global Assessment. Etiologic criteria included self-reported reduced food intake and inflammation (metastatic disease). Machine learning approaches were applied to predict 30-day mortality and unplanned hospital admission using models with and without MM.

Results: Overall, 2494 participants were included (49.6% male, mean (SD) 62.3 (14.2) years). Malnutrition prevalence was 19.5% (including MM) and 17.5% (excluding MM). However, 48/485 (10%) of malnourished participants were missed if MM was excluded. In the nine GLIM combinations excluding MM, the most important combinations to predict mortality were: 1) weight loss and inflammation and 2) weight loss and reduced food intake. Performance metrics were similar in models with and without MM to predict mortality (average accuracy: 84% vs. 88%). Weight loss and reduced food intake were the most important combination to predict unplanned hospital admission. Performance metrics were almost identical in models with and without MM to predict unplanned hospital admission (average accuracy: 76.86% vs. 76.97%).

Conclusions: Results indicate predictive ability is maintained, although the ability to identify all malnourished patients is compromised, when muscle mass is excluded from the GLIM diagnosis. This has important implications in health services where equipment to assess muscle mass is not available.