Plasma metabolic fingerprinting identifies phenylalanine as a mortality biomarker in elderly ARDS patients

Abstract Background Currently, there remains a lack of effective prognostic biomarkers specifically for ARDS patients. Methods We conducted a retrospective study profiling plasma metabolomics (ARDS = 55, non-ARDS = 37) via high-throughput NPELDI-MS. Machine learning (LR, SVM, NN) developed age-stratified prognostic models (< 65 vs. ≥65 years). Univariate logistic regression identified key mortality-associated metabolites in elderly ARDS, followed by cross-platform validation. Results The plasma metabolic fingerprints-based model distinguished ARDS from non-ARDS with an AUC of 0.928, revealing systemic metabolic dysregulation involving amino acid metabolism, one‑carbon metabolism, cofactor biosynthesis, and glutathione‑related pathways. Furthermore, we developed age‑stratified prognostic prediction models: the AUC for distinguishing non‑survivors from survivors was 0.989 in < 65 years patients and 0.937 in ≥ 65 years patients, whereas mixed-age analysis showed limited metabolic separation between survivors and non-survivors. Importantly, age-stratified analyses identified distinct exploratory metabolite panels: phenylalanine and glycine showed nominal positive associations with mortality in patients aged ≥ 65 years, whereas taurine, hypotaurine and niacinamide showed nominal inverse associations in those aged < 65 years. Cross-platform validation confirmed the generalizability of elderly derived panel (AUC = 0.829), and further identified phenylalanine remained significantly elevated in elderly non-survivors. Conclusions Our study establishes an integrated plasma metabolic fingerprinting framework for ARDS age‑stratified prognosis, and further identifies phenylalanine and glycine as elderly-specific metabolic risk factors for mortality, highlighting age-dependent metabolic vulnerability in ARDS. Trial registration This trial is registered at ClinicalTrials.gov (identifier: NCT07380997).