Related Experiment Video
Updated: Sep 19, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
A scalable workflow using digital cognitive assessment and pTau217 for MCI detection
Srinivasan Vairavan1, Nicholas Griffin2, David Wilson3
1Johnson & Johnson Titusville New Jersey USA.
Introduction:
Blood biomarkers like phosphorylated tau (p-tau)217 offer high diagnostic accuracy for Alzheimer's disease (AD) but face implementation challenges in low-prevalence settings. We evaluated the Altoida NeuroMarker Platform as a digital cognitive triage tool before plasma p-tau217 testing.
Methods:
XGBoost models were trained to predict mild cognitive impairment (MCI) and p-tau217 status in 688 individuals across Australia, Spain, and the United States using the Altoida NeuroMarker. p-tau217 status was dichotomized using a threshold of 0.04 pg/mL (LucentAD). We modeled a two-step workflow (Altoida, plasma p-tau217) to enrich downstream biomarker testing, assuming 30% prevalence of AD pathology.
Results:
The Altoida NeuroMarker accurately identified MCI (receiver operating characteristic area under the curve [ROC AUC] = 0.89 ± 0.01) and p-tau217 elevation (ROC AUC = 0.77 ± 0.08). Predicted MCI was significantly associated with elevated pTau217 (p = 0.004). When used upstream, the Altoida NeuroMarker ruled out 64.4% of participants (negative predictive value 90.2%), shifting the pre-test probability of AD pathology to 66.6%, and improving the modeled positive predictive value of plasma p-tau217 from 81.7% to 95.3%.
Discussion:
This workflow outlines a scalable path to enrich blood-based biomarker testing after digital cognitive screening.
