Abstract
The clinical application of the frailty phenotype and frailty index still has some limitations, and whether the classification of frailty based on metabolites is beneficial to the management of the frailty population remains unclear. This study analyzed 160,407 UK Biobank participants to define frailty subtypes using metabolic profiles. Based on 251 biomarkers, machine learning identified 11 key metabolites, leading to four novel frailty subtypes. Subtypes III and IV, characterized by adverse metabolic features such as high GlycA and low LA/FA, were designated as high-risk groups. These subtypes showed significantly increased risks for 13 chronic diseases and all-cause mortality compared to lower-risk subtypes. Adherence to a healthy diet was associated with risk reduction in the high-risk groups. These findings demonstrate the heterogeneity of frailty and suggest that metabolite-based subtyping could improve prognostic precision and guide targeted dietary interventions in clinical practice.