P4‐123: Resolving Conflicting CSF Biomarker Information in Alzheimer′s Disease

Different CSF biomarker combinations can provide conflicting diagnostic information in Alzheimer′s disease (AD). This is often attributed to differences in sensitivity and specificity, at the cohort level, between CSF markers (Aβ42, t-Tau, p-Tau181, t-Tau/Aβ42, and p-Tau181/Aβ42). When these biomarkers are analyzed against the same gold standard independently, conflicting biomarker information can also result from biomarker substructures not obvious to investigators. Previous studies have not examined conflicting biomarker information at the individual level (e.g., a profile showing normal Aβ42 levels but abnormal t-Tau/Aβ42 ratio may be interprted as AD-like even though the normal Aβ42 level argues against amyloid pathology). The prevalence of these conflicts and ways to resolve them are unknown. We measured CSF AD biomarker levels in one consecutive series (n=431) from Emory University using the multiplex AlzBio3 assay and surveyed the concordance rates between CSF biomarkers at the individual level. We also compared these results with those from clinical testing through a comparable ELISA. To resolve the issue of differential sensitivity and biomarker substructure, we then analyzed CSF AD biomarker levels through two-step clustering to identify naturally existing subgroups of biomarker profiles. Finally, to determine if the cluster membership or the combination of independent biomarker information confers greater information on prognosis, we analyzed if either predicted longitudinal cognitive changes in the Alzheimer’s Disease Neuro-Imaging Initiative (ADNI, n=409). Conflicting CSF biomarker information was very common: 59% of the Emory subjects and 37% of ADNI subjects had at least one biomarker providing diagnostic information distinct from the other biomarkers. Clustering analysis revealed three groupings: one characterized by p-Tau181/Aβ42>0.131 and longitudinal cognitive decline in MCI, and two others (including one characterized by Aβ42>258.5pg/mL) associated with cognitive stability. Within each cluster, concordant or discordant biomarker findings did not further distinguish rates of longitudinal cognitive decline. Conflicting information from different CSF AD biomarkers was common. A data-driven strategy accounting for all biomarker combinations identified naturally existing groupings each characterized by similar biochemical and prognostic profiles.