A Two-Stage Plasma Biomarker Algorithm Reduces Amyloid PET Use and the p-Tau217 Gray Zone

Objective: Plasma p-Tau217 leaves 30-50% of patients in an indeterminate "gray zone" often requiring A{beta}-PET. We developed GRAD (Gatekeeper-Reflex for Alzheimer's Disease), a two-stage A{beta}+ prediction model to improve classification while reserving A{beta}-PET for residual uncertainty. Methods: GRAD was developed in 145 cognitively impaired ADNI participants and externally validated in 1,644 cognitively unimpaired A4/LEARN participants. Stage 1 used p-Tau217 alone; probabilities between 0.25-0.75 were considered indeterminate. Stage 2 used L2-regularized logistic regression combining plasma-biomarkers, genetic and demographic features. A{beta}-PET served as the reference standard (ADNI florbetapir SUVR>1.11; A4 Centiloid[&ge;]20). Results: Full model AUC was 0.915 (95% CI, 0.859-0.961) internally and 0.879 (95% CI, 0.861-0.895) externally. In the external gray zone (n=630), Stage 2 improved discrimination over p-Tau217 alone (AUC 0.793 vs 0.719; P<.001), increased correct classification without PET from 41.4% to 56.2% and reduced PET referral from 35.9% to 24.3%. In the p-Tau181 subset (n=256), Stage 2 outperformed p-Tau181 (AUC 0.799 vs 0.597; P<.001). Overall, 89.0% of ADNI and 90.7% of A4/LEARN participants were classified without PET; modeled testing costs were 44-52% lower than p-Tau217 screening with PET confirmation of the gray zone. Interpretation: Reflex multimarker testing improves indeterminate p-Tau217 classification while retaining A{beta}-PET for residual uncertainty. GRAD may reduce PET use and testing costs in memory-clinic triage, prevention-trial enrichment, and disease-modifying therapy workup.