Dementia risk prediction in individuals with prediabetes or diabetes: A novel multi-protein score with biological...

Individuals with prediabetes or diabetes face elevated dementia risk, yet robust prediction tools and mechanistic insights remain limited.

This study aimed to develop and validate a protein-based risk score for dementia prediction in this high-risk population while elucidating underlying biological pathways and therapeutic targets.

Utilising data from 10 433 UK Biobank participants with prediabetes or diabetes and proteomic profiling (2911 plasma proteins measured), we developed a dementia protein risk score in a training set (n = 6514) and validated it in testing (n = 2790) and external cohorts (n = 1129).

In the training set, 23 out of 2911 proteins were selected. In the testing set, compared with the basic model (age and sex, C-index: 0.78; 95% confidence interval [CI] 0.74-0.82), the dementia protein risk score (C-index: 0.84; 95% CI 0.81-0.88) significantly improved the performance in predicting incident dementia (C-index increase: 0.06; 95% CI 0.02-0.12), while cardiovascular risk factors, ageing and dementia incidence risk factors (C-index: 0.80; 95% CI 0.76-0.83) and apolipoprotein E (APOE; age and sex included, C-index: 0.81; 95% CI 0.77-0.85) had no significant improvement. Six key proteins (glial fibrillary acidic protein [GFAP], neurofilament light polypeptide [NEFL], Brevican core protein [BCAN], protein MENT [MENT], APOE and growth/differentiation factor 15 [GDF15]) captured the most predictive power. Pathway analyses implicated extracellular matrix remodelling and cholesterol metabolism, whereas Mendelian randomisation identified causal roles for APOE, haematopoietic prostaglandin D synthase (HPGDS), BAG family molecular chaperone regulator 3 (BAG3) and GDF15. Nine proteins were prioritised as druggable targets, including HPGDS, with existing Food and Drug Administration-approved drugs.

This study establishes a highly accurate protein-based risk score for dementia prediction (including 6-23 proteins) in individuals with prediabetes or diabetes, uncovering actionable biological pathways and therapeutic targets. The findings enable precision risk stratification and accelerate translational opportunities for dementia prevention in this population.