Among 7,671 UK Biobank participants with type 2 diabetes (mean age ~60), NMR-based metabolomic profiling across data-driven T2D subtypes revealed that lipid and lipoprotein remodeling dominates the metabolic landscape separating these subtypes. An HDL-remodeling principal component independently raised risk of all-cause dementia by 17% (HR 1.17), vascular dementia by 18% (HR 1.18), and all-cause mortality by 16% (HR 1.16). Paradoxically, lower scores on an LDL-enriched axis were associated with reduced dementia risk (HR 0.75) and lower mortality (HR 0.76). Mendelian randomization identified 197 metabolite-outcome associations surviving FDR correction, supporting potential causal relevance.
These findings matter because T2D is rarely a monolithic condition — its heterogeneity may explain why cardiovascular and cognitive risk varies so dramatically between patients on similar treatments. Pinpointing HDL remodeling, rather than total HDL levels, as a dementia-risk signal is mechanistically intriguing: it aligns with emerging evidence that HDL particle dysfunction, not simply low HDL concentration, drives vascular and neurological injury. The inverse LDL finding warrants caution, as residual confounding in observational data can invert expected lipid-outcome relationships. Crucially, this is a preprint posted on medRxiv and has not yet undergone peer review — findings, particularly the Mendelian randomization causal claims, should be treated as preliminary. If validated, metabolomic subtyping could refine dementia prevention strategies within T2D populations, moving clinical care toward precision risk stratification rather than one-size-fits-all management.
