Prediction of malnutrition in patients with cancer using machine learning: a review

Malnutrition substantially affects the outcomes of antitumor therapy in patients with cancer. Nutritional support is often prescribed subjectively, which may lead to errors in determining the need for artificial nutrition, route of administration, and composition of nutritional therapy. While machine learning and artificial intelligence are increasingly being incorporated into clinical practice, their application in nutritional support remains rather limited. Here, we conducted a review to highlight the current state of this issue. We analyzed publications indexed in MEDLINE, Web of Science, and the Scientific Electronic Library (eLibrary) from 2021 to 2024 addressing the use of artificial intelligence — namely, machine learning algorithms — for early identification of protein–energy deficiency and prediction of its development in patients with cancer. The findings indicate that predictive models based on artificial intelligence, as well as models for identifying protein–energy deficiency, can be integrated into clinical decision support systems. This approach enables timely diagnosis and correction of nutritional deficiencies and helps reduce the subjectivity and limitations inherent to the conventional empirical approach to prescribing nutritional support. The review also discusses common errors encountered in the implementation of nutritional support in patients with cancer, and outlines opportunities for mitigating them through machine learning. The analysis shows that despite considerable prospects, the use of machine learning and artificial intelligence for identifying nutritional deficiencies and delivering nutritional support in real clinical practice remains rather limited.