Artificial Intelligence-Guided Artificial Nutrition in Critical Illness: Integrating Indirect Calorimetry and BIVA...

Background: Critical illness is characterized by profound and rapidly evolving metabolic derangements driven by systemic inflammation, hypercatabolism, fluid shifts, and endocrine dysregulation. These dynamic changes markedly limit the accuracy of predictive equations, increasing the risk of both underfeeding and overfeeding. Indirect Calorimetry Energy represents the gold standard for measuring energy expenditure, while bioelectrical impedance vector analysis (BIVA) provides complementary insights into hydration status, cellular integrity, and body cell mass. In palliative care, AI-supported integration of indirect calorimetry and BIVA enables goal-concordant artificial nutrition by aligning energy delivery with real-time metabolic status while minimizing symptom burden. Artificial intelligence (AI) has emerged as a promising tool to integrate these heterogeneous data streams and support adaptive nutritional strategies. Methods: We conducted a structured narrative review of the literature published between 2000 and 2025 using PubMed, Scopus, Embase, and Web of Science. Artificial intelligence was not used to perform the literature search or study selection. Instead, AI was analyzed as a clinical and technological component within the included studies and explored as a future enabling strategy. Eligible publications involved adult critically ill patients and addressed indirect calorimetry, BIVA-derived parameters, or AI-based metabolic modeling applied to nutritional support. Given the heterogeneity of study designs and outcomes, findings were synthesized qualitatively. Results: Predictive equations showed substantial inaccuracy in unstable metabolic states, with errors frequently exceeding ±20-40%. Indirect calorimetry enabled individualized assessment of energy expenditure but remained limited by intermittent availability. Serial BIVA assessments consistently identified clinically relevant alterations in hydration status, body cell mass, and phase angle, the latter being strongly associated with adverse outcomes. Studies incorporating AI demonstrated improved integration of calorimetry, BIVA, and clinical variables, allowing identification of metabolic phenotypes, anticipation of metabolic shifts, and generation of adaptive nutritional recommendations. Conclusions: This narrative review highlights the complementary roles of Indirect Calorimetry and BIVA in characterizing metabolic needs in critical illness. Artificial intelligence does not replace these tools but enhances their clinical utility by integrating multidimensional data into dynamic, patient-specific nutritional strategies. The combined AI-IC-BIVA approach represents a promising framework for metabolic precision nutrition in the ICU, warranting prospective validation.