Abstract
Neurology. 2026 Nov 24;107(10):e218681. doi: 10.1212/WNL.0000000000218681. Epub 2026 Oct 23.
ABSTRACT
BACKGROUND AND OBJECTIVES: Blood-based biomarkers are strongly correlated with amyloid positivity in research settings, but their long-term performance in heterogeneous populations remains unclear. Our aim was to determine their long-term prediction accuracy for incident all-cause dementia on their own and when used alongside known risk factors in community-dwelling older adults.
METHODS: This was a retrospective study based on the UK Whitehall II longitudinal, population-based cohort study established in 1985 among civil servants that includes a clinical examination every 4-5 years. Blood samples stored at the 2013-2013 wave were assayed in 2024 for measures of amyloid-beta (Aβ) 42 and 40, glial fibrillary acidic protein (GFAP), neurofilament light (NfL), and phosphorylated tau protein 217 (p-tau217). Incident all-cause dementia, from 2012 to 2013 to 2025, was ascertained from linked medical records, and Cox regression was used to examine associations of biomarkers with dementia, followed by estimation of prediction statistics.
RESULTS: Of the 5,225 (27.2% female) participants (mean age in 2012-2013, 69.4 ± 5.7 years), 516 (30.0% female) had a dementia diagnosis over a 12.2-year median follow-up. Participants with values in the worst quartile of amyloid-beta 42 (Aβ42)/40 ratio (hazard ratio [HR] 1.88; 95% CI 1.57-2.24), p-tau217 (HR 3.74; 95% CI 3.12-4.50), NfL (HR 1.88; 95% CI 1.56-2.28), and GFAP (HR 2.25; 95% CI 1.87-2.72) had higher risk of dementia. The C-index for predicting dementia was highest for p-tau217 (0.693; 95% CI 0.672-0.714), lowest for Aβ42/40 ratio (0.587; 95% CI 0.566-0.608), and was 0.760 (0.738, 0.781) for all biomarkers together. Sensitivity (between 41.7% and 59.5%) and positive predictive value (between 16.4% and 23.5%) of all biomarkers were modest, but specificity (between 76.8% and 78.8%) and negative predictive value (92.3%-94.7%) were good. A second approach to prediction modeling showed a model containing sociodemographic, behavioral, and health-related risk factors to have a C-index of 0.765 (95% CI 0.746-0.785); adding all biomarkers to this model led to a C-index of 0.813 (95% CI 0.795-0.830).
DISCUSSION: The long-term predictive ability of blood-based biomarkers for incident dementia in community-dwelling older adults was modest due to poor sensitivity and similar to that of a model containing known risk factors.
PMID:42913444 | DOI:10.1212/WNL.0000000000218681
UK DRI Authors
