PD238 - QUANTIFYING GLOBAL (DIS)AGREEMENT IN ICU NUTRITION GUIDELINES: AN UMBRELLA REVIEW USING NATURAL LANGUAGE PROCESSING

PD238

QUANTIFYING GLOBAL (DIS)AGREEMENT IN ICU NUTRITION GUIDELINES: AN UMBRELLA REVIEW USING NATURAL LANGUAGE PROCESSING

O. Leemans^1,2,*, Z. Rosseel^1,2,3, L. Leemans^1,2,4, E. De Waele^1,2

¹Vitality Research Group, Vrije Universiteit Brussel (VUB), ²Clinical Nutrition, ³Pharmacy, Universitair Ziekenhuis Brussel (UZ Brussel), ⁴Rehabilitation Research Group, Vrije Universiteit Brussel (VUB), Brussels, Belgium

Rationale: ICU nutrition guidelines frequently deliver diverging recommendations, hindering unified bedside practice. This umbrella review aims to (i) evaluate the topics addressed in international guidelines and (ii) quantify recommendation convergence and reliance on expert consensus using automated semantic analysis.

Methods: A systematic search (PubMed, Web of Science, snowballing) identified ICU nutrition guidelines and consensus statements published after 2000, using Rayyan®. Recommendations were categorized into 14 topics. Using Natural Language Processing (all-MiniLM-L6-v2 transformer model, pairwise cosine similarity) semantic convergence within topics and consensus reliance was quantified. Clinical disagreements were defined as similarity <0.30 and conflicting numerical targets (e.g., timing, dosing).

Results: 648 Recommendations across 14 guidance documents were analyzed, showing significant heterogeneity. Highest convergence occurred in Screening (n=26, mean similarity 0.573, for 73% reliant on expert consensus). Lowest convergence was found in Energy (n=98, 0.420, 54%) and Intake Route (n=122, 0.405, 45%). While Specific Patient Populations (n=196) and Intake Route (n=122) were most prevalent, Staff Protocols (n=25), Physical Activity (n=3), and Post-ICU follow-up (n=1) were underrepresented. Overall, 77% (498/648) of all recommendations were based on expert consensus or low-quality evidence.

Image:

Conclusion: ICU nutrition recommendations predominantly focus on subpopulations, intake route, and energy targets, yet most are based on expert opinion rather than high-grade evidence. Agreement across guidelines is strongest for screening-related recommendations. The near-absence of post-ICU guidance and high divergence in recommendations highlight the urgency for trial-based research and international harmonization. 

Disclosure of Interest: None declared