P4‐229: Improved CSF‐based discrimination between Alzheimer's disease patients and controls after correction for ventricular volumes

Abstract

Cerebrospinal fluid (CSF) and MRI biomarkers may support the diagnosis of Alzheimer's disease (AD). At equal amyloid beta (Aβ) or tau production rates, variations in CSF volume may lead to variations in CSF protein concentrations, affecting interpretation of results. We studied if the diagnostic power of AD CSF biomarker concentrations, i.e. Aβ42, total and phosphorylated tau (t-tau, p-tau), can be improved by correction for lateral ventricular volume (VV). Biomarker data and MRI scans from 12 centers were obtained from subjects (total: n=730; AD: n=175, controls: n=157), who underwent both lumbar puncture and MRI scan within 6 months of each other. CSF Aβ42, t-tau, p-tau concentrations were determined using Fujirebio ELISAs. We developed a Matlab-algorithm for standardized automated segmentation analysis of T1 scans in SPM8, determining VV as proxy for CSF volume; VV was also used as ratio with total intracranial volume (TIV). All segmentation outcomes were judged by visual inspection. Height, age, gender, diagnosis according to internationally accepted criteria, scanner type, magnetic field strength, were also recorded. Bivariate correlations were determined between the CSF biomarkers and VV/TIV ratio. Spearman correlation and linear regression analyses were performed to define optimal diagnostic models. The diagnostic power of CSF biomarkers (combinations) was determined by ROC analyses and correction for VV/TIV ratio and covariates. The AD CSF biomarkers negatively correlated weakly to lateral ventricular volume (Aβ42: p<0.0001, r2=0.037; p-tau: p=0.4645, r2=0.0063; t-tau: p=0.3450, r2=0.0059). For differentiation of AD and controls, the area under the curve (AUC) for Aβ42 (0.75), but not for p-tau and t-tau, increased after correction for VV/TIV ratio (0.81). Models constructed using logistic regression showed that the combination of Aβ42 and t-tau improved the diagnostic value (AUC: 0.88) to discriminate AD from controls, compared to the single markers. The diagnostic power was further improved by correcting Aβ42 for VV/TIV ratio (AUC: 0.91). We developed and validated a novel algorithm for automated measurement of VV in T1 images. CSF Aβ42 concentrations decreased with increasing VV; correction for differences in this volume improved the differentiation of AD vs. controls based on CSF Aβ42 alone or in combination with t-tau.