Abstract
ABSTRACT
Aim
To derive a compact plasma proteomic signature associated with incident dementia and evaluate the internal performance of the finalized score in UK Biobank.
Methods
In this prospective cohort study, we screened 2923 baseline Olink assays using Cox models adjusted for age, age squared, sex, and the first 10 genetic principal components. Proteins passing false‐discovery‐rate control underwent 2‐ and 5‐year landmark analyses, correlation‐based redundancy pruning, and repeated stability‐based feature selection. The finalized score was evaluated in basic and extended clinical models, a chronological subgroup performance assessment, proportional‐hazards testing, and landmark sensitivity analyses.
Results
Among 36 696 participants (mean age 56.84 years; 591 incident dementia events; median follow‐up 15.73 years), the final 11‐protein score was associated with incident dementia (hazard ratio per 1‐SD increase, 2.645; 95% confidence interval, 2.459–2.845) and increased Harrell's C‐index from 0.798 to 0.862. In the extended clinical cohort (23 179 participants; 441 events), C‐index increased from 0.792 to 0.861. In the later chronological subset (6954 participants; 97 events), C‐index increased from 0.801 to 0.869. Associations persisted in the 2‐ and 5‐year landmark analyses, and no statistically significant score‐specific violation of the proportional‐hazards assumption was detected (
p
= 0.105).
Conclusions
A broad plasma proteomic screen yielded a compact 11‐protein signature that was strongly associated with incident dementia and showed consistent internal performance after the score had been finalized. Independent external validation is required before clinical use.