A modelling approach to derive population-specific cutoff for plasma p-Tau217.

Abstract

Plasma pTau-217 shows promise for detecting Alzheimer's disease, but needs population-specific cutoffs for effective use. Conventional cutoff determination relies on invasive or costly gold-standards, limiting scalability. This study evaluated Finite Mixture Modelling (FMM) for establishing cutoffs without gold-standards. FMM was applied to derive cutoffs for Lumipulse plasma p-Tau217 and p-Tau217/Aβ42 ratio among 1039 ADNI participants, with validation conducted in a subset with amyloid PET data (n = 711). Additionally, simulations were conducted to determine the minimum sample size for reliable FMM estimation. The results showed that FMM-derived cutoffs effectively classified participants into brain amyloid-negative, -positive, and -indeterminate groups, with an indeterminate proportion <20 %, negative and positive predictive values near or above 90 %, and with p-Tau217/Aβ42 outperforming p-Tau217. These FMM-derived cutoffs demonstrated test performance that surpassed several previously-established cutoffs, including the recent FDA-approved cutoff. At least 900 samples were needed for reliable cutoff estimation. In conclusion, this study demonstrated the effectiveness of a modelling approach for estimating plasma p-Tau217 cutoffs without reliance on gold-standards. This approach simplifies the determinating of population-specific cutoffs and facilitates adoption of plasma p-Tau217 in communities lacking access to gold-standards, including some LMICs.