The Short Version
The 79% accuracy figure is real but significantly mischaracterized. UEA researchers developed a preliminary machine-learning model using six gut-derived blood metabolites that classifies participants across three study groups — healthy, mild cognitive impairment (MCI), and impaired — with 79% accuracy. This is not a validated clinical blood test, and MCI is a precursor state, not equivalent to "early-stage dementia." Even UEA-affiliated coverage describes the work as research that "could pave the way" for a future test, not a deployable diagnostic tool.