Where Should I Draw the Line: PET-Driven, Data-Driven, or Manufacturer Cut-Off?

Abstract

Background: The optimal cut-off for Alzheimer’s disease (AD) CSF biomarkers remains controversial. Objective: To analyze the performance of cut-off points standardized by three methods: one that optimized the agreement between 11C-Pittsburgh compound B PET (a-PET) and CSF biomarkers (Aβ1–42, pTau, tTau, and Aβ1–42/Aβ1–40 ratio) in our population, called PET-driven; an unbiased cut-off using data from a healthy research cohort, called data-driven, and that provided by the manufacturer. We also compare changes in ATN classification. Methods: CSF biomarkers measured by the LUMIPULSE G600II platform and qualitative visualization of amyloid positron emission tomography (a-PET) were performed in all the patients. We established a cut-off for each single biomarker and Aβ1–42/Aβ1–40 ratio that optimized their agreement with a-PET using ROC curves. Sensitivity, Specificity, and Overall Percent of Agreement are assessed using a-PET or clinical diagnosis as gold standard for every cut-off. Also, we established a data-driven cut-off from our cognitively unimpaired cohort. We then analyzed changes in ATN classification. Results: One hundred and ten patients were recruited. Sixty-six (60%) were a-PET positive. PET-driven cut-offs were: pTau > 57, tTau > 362.62, Aβ1–42/Aβ1–40 < 0.069. For a single biomarker, pTau showed the highest accuracy (AUC 0.926). New PET-driven cut-offs classified patients similarly to manufacturer cut-offs (only two patients changed). However, 20 patients (18%) changed when data-driven cut-offs were used. Conclusions: We established our sample’s best CSF biomarkers cut-offs using a-PET as the gold standard. These cut-offs categorize better symptomatic subjects than data-driven in ATN classification, but they are very similar to the manufacturer’s.