Analytical agreement and platform-specific decision thresholds for plasma p-tau217 measured on Lumipulse G600II and...

Plasma phosphorylated tau at threonine 217 (p-tau217) has emerged as a highly accurate blood-based biomarker of Alzheimer's disease (AD) pathology. As disease-modifying anti-amyloid therapies enter clinical practice, scalable biomarkers with robust and clinically interpretable decision thresholds are required. However, evidence on inter-platform comparability and threshold transferability across automated assays remains limited.

We conducted a head-to-head comparison of two automated platforms - Lumipulse^® G600II and Cobas^® e801 - for plasma p-tau217 measurement in 157 consecutive patients undergoing lumbar puncture. Amyloid status was defined by the CSF Aβ42/Aβ40 ratio. Agreement was assessed using intraclass correlation coefficients and Bland-Altman analysis. Diagnostic performance was evaluated using receiver operating characteristic curves. Optimal thresholds were derived using the Youden index. Predefined rule-out (≥95 % sensitivity) and rule-in (≥95 % specificity) thresholds were explored, alongside alternative ≥90 % thresholds.

Agreement between platforms was excellent (Spearman ρ=0.922; ICC(3,1)=0.922), although Bland-Altman analysis revealed a small systematic difference in absolute concentrations. Both assays showed comparable diagnostic accuracy for amyloid positivity (AUC=0.923 for both platforms; DeLong p>0.99), but required platform-specific thresholds. Rule-out and rule-in thresholds achieved ≥95 % sensitivity and specificity, with strong likelihood ratios and excellent categorical agreement (weighted κ=0.870). Approximately 30 % of individuals were classified in the grey zone. Using ≥90 % thresholds reduced the grey zone to 9-13 % while maintaining excellent agreement.

Plasma p-tau217 demonstrates high analytical concordance and comparable diagnostic performance across automated platforms despite systematic concentration differences. Platform-specific dual-threshold strategies may support structured and clinically interpretable implementation, pending prospective multicenter validation.