Reducing Alzheimer’s diagnostic uncertainty with a sequential plasma p-tau217 workflow in real-world memory clinic practice

Abstract Background Plasma p-tau217 and p-tau217/Aβ42 show high accuracy for Alzheimer’s disease (AD) amyloid pathology and are entering specialist clinical workflows, with standalone p-tau217 CE-IVDR marked in Europe and the p-tau217/Aβ42 ratio FDA-cleared in the United States. Their optimal integration into uncertainty-aware diagnostic pathways, particularly for patients near decision thresholds, remains unclear. Methods We retrospectively analyzed 562 consecutive memory-clinic patients assessed between October 2023 and April 2026. Core1 amyloid status was defined using CSF biomarkers and/or amyloid-PET according to the 2024 Alzheimer’s Association Revised Criteria. Plasma biomarkers were measured on the Lumipulse G600II platform. We compared three two-threshold workflows: p-tau217 alone; p-tau217/Aβ42 as a universal upfront ratio; and a p-tau217-first sequential workflow in which p-tau217/Aβ42 was used only for patients with intermediate p-tau217 results. Rule-out and rule-in thresholds were selected to achieve ≥ 95% sensitivity and ≥ 95% specificity, respectively. A reverse sequential workflow and internal temporal validation were used as sensitivity analyses. Results Core1 + prevalence was 71.2% ( n = 400). P-tau217 achieved an AUC of 0.935 (95%CI 0.908–0.962), outperforming p-tau181 (0.826), Aβ42/Aβ40 (0.797), GFAP (0.789) and NfL (0.566; all p < 0.001). The p-tau217 two-threshold approach yielded a 17.8% intermediate zone with 94.1% decision-zone accuracy. The universal upfront p-tau217/Aβ42 showed higher global discrimination than p-tau217 alone (AUC 0.946; p = 0.011), but reduced indeterminacy only marginally to 17.1%. The p-tau217-first sequential workflow reduced indeterminacy to 12.3%, sparing 56 confirmatory CSF or amyloid-PET tests per 1,000 patients, with sensitivity 0.937, specificity 0.935, PPV 0.973 and NPV 0.855. The reverse sequential workflow showed comparable classification performance but would require p-tau217/Aβ42 upfront in all patients. Internal temporal validation confirmed transportability despite a 15-percentage-point Core1 + prevalence shift (validation AUC 0.941; ΔAUC = + 0.018; p = 0.54). Conclusions In this real-world memory-clinic cohort, the higher global AUC of p-tau217/Aβ42 did not translate into meaningful reduction of indeterminacy when used as a universal upfront replacement. Its clinical value was greater when positioned as a second-step assay in p-tau217 intermediate-zone patients, supporting uncertainty-aware plasma biomarker workflows to prioritise confirmatory CSF or amyloid-PET testing.