Development and External Validation of a Plasma p-tau217-Based Model for MCI-to-Dementia Progression

-Clinical trajectories after a diagnosis of mild cognitive impairment (MCI) vary widely, so prognostic models must account for both the timing of subsequent diagnoses and unequal follow-up. Using the Alzheimer's Disease Neuroimaging Initiative (ADNI), we developed a Cox proportional-hazards model for time from an MCI-aligned plasma index visit to a subsequent dementia diagnosis, using age at study entry and log-transformed plasma p-tau217. Coefficients and baseline survival were estimated in 401 participants with 85 events and then frozen and applied without refitting to an independent cohort from the National Alzheimer's Coordinating Center (NACC). NACC p-tau217 measurements were reported on the Quanterix HD-X platform and were mapped to the ADNI development-assay scale using a prespecified transformation derived from paired ADNI measurements. The two-predictor model reached a pooled out-of-fold C-index of 0.775 in ADNI (apparent C-index, 0.778) and 0.652 (95% CI, 0.526-0.765) in 104 NACC participants with 25 incident dementia diagnoses. At two years, observed risk was 0.214 compared with a mean predicted risk of 0.089, an observed-to-expected ratio of 2.40. Neither adding APOE ε 4 count, sex, or education nor expanding the model with GFAP, NfL, and the A β 42/A β 40 ratio materially improved discrimination in ADNI; APOE ε 4 also provided no detectable incremental gain in NACC. The model showed moderate discrimination in NACC, although the wide confidence interval includes values near chance and the middle and higher risk tertiles did not clearly separate; C-indices were similar in the bridged primary and direct-scale sensitivity analyses, and mean two-year predicted risk was lower than observed only in the bridged analysis.