356. LLM-assisted Meta-analysis of Seven Feed Additive Categories Reveals Differential Effects on Pork Quality Traits.

Abstract

Animal nutrition research is characterized by expanding yet fragmented evidence, and existing meta-analyses remain narrowly scoped to single feed ingredients, precluding cross-intervention synthesis. Our objective was to develop LANMA (LLM-driven Animal Nutrition Meta-Analysis workflow) and apply it to systematically quantify the effects of seven dietary additive categories on core pork quality outcomes. A Boolean search of PubMed and Web of Science (2000–2025) retrieved 19,594 records. LANMA integrated LLM-assisted screening and structured data extraction alongside dual independent human verification to ensure accuracy and traceability. After full-text assessment, 142 eligible studies comprising 2,893 independent comparisons were retained, spanning seven additive categories (protein and energy feeds, methyl donors, trace minerals and vitamins, fatty acids, plant extracts, probiotics, and fermented feed) and five primary outcomes: post-mortem pH at 24 h, drip loss, meat lightness L_24h, shear force, and marbling score. Random-effects meta-analyses (REML; Knapp-Hartung adjustment) were performed using weighted mean differences (WMD) with unit-aligned data. To account for the non-monotonic nature of pH and meat color, a probabilistic desirability scoring framework (1–10 scale) was applied to pH_24h and L_24h, in which a positive WMD indicated a shift toward the predefined optimal physiological window. Plant extracts significantly improved water-holding capacity and tenderness, reducing drip loss (WMD = −0.42% [−0.80, −0.05], P < 0.05) and shear force (WMD = −3.31 N [−5.13, −1.48], P < 0.01), whereas their effect on marbling score was not significant. Methyl donor feed additives did not significantly improve pH_24h optimal-window desirability but significantly reduced shear force (WMD = −2.76 N [−5.21, −0.30], P < 0.05). Fermented feed was the only additive category that significantly improved L_24h optimal-window desirability (WMD = 0.50 [0.13, 0.87], P < 0.05), indicating a beneficial shift toward the optimal lightness range; plant extracts showed a positive but non-significant trend for L_24h (WMD = 0.37 [−0.03, 0.77], P = 0.062). Protein and energy feeds significantly reduced drip loss (WMD = −0.22% [−0.39, −0.06], P < 0.05) and increased marbling score (WMD = 0.15 [0.02, 0.28], P < 0.05), whereas their effect on shear force was not significant. Trace minerals and vitamins, fatty acids, and probiotics showed no significant improvement across the five primary outcomes under the unified WMD interpretation. No additive category significantly improved pH_24h optimal-window desirability. These findings demonstrate distinct, category-specific effects of dietary interventions on pork quality: plant extracts improved water-holding capacity and tenderness, protein and energy feeds improved drip loss and marbling score, methyl donor feed additives improved tenderness, and fermented feed improved L*_24h optimal-window desirability. LANMA provides an auditable, human-verified, LLM-assisted framework that enables scalable cross-category evidence synthesis, offering a reproducible template for precision swine nutrition research.

For image description, please refer to the figure legend and surrounding text.