Diagnostic Performance and Clinical Applicability of Blood-Based Biomarkers in a Prospective Memory Clinic Cohort

Abstract

Background and objectives: Blood-based biomarkers have emerged as minimally invasive options for evaluating cognitive impairment. Most studies to date have assessed them in research cohorts, limiting their generalization to everyday clinical practice. We evaluated their diagnostic performance and clinical applicability in a prospective, real-world, memory clinic cohort.

Methods: All patients referred with suspected cognitive impairment between July 2019 and June 2021 were prospectively invited to participate. Five plasma biomarkers (tau phosphorylated at threonine 181 [p-tau181], glial fibrillary acidic protein [GFAP], neurofilament light chain [NfL], total tau [t-tau], and ubiquitin C-terminal hydrolase L1 [UCH-L1]) were determined with single-molecule array. Performance was assessed in comparison to clinical diagnosis (blinded to plasma results) and amyloid status (CSF/PET). A group of cognitively unimpaired (CU) controls was also included.

Results: Three hundred forty-nine participants (mean age 68, SD 8.3 years) and 36 CU controls (mean age 61.7, SD 8.2 years) were included. In the subcohort with available Alzheimer disease (AD) biomarkers (n = 268), plasma p-tau181 and GFAP had a high diagnostic accuracy to differentiate AD from non-neurodegenerative causes (area under the receiver operating characteristic curve 0.94 and 0.92, respectively), with p-tau181 systematically outperforming GFAP. Plasma p-tau181 levels predicted amyloid status (85% sensitivity and specificity) with accurate individual prediction in approximately 60% of the patients. Plasma NfL differentiated frontotemporal dementia (FTD) syndromes from CU (0.90) and non-neurodegenerative causes (0.93), whereas the discriminative capacity with AD and between all neurodegenerative and non-neurodegenerative causes was less accurate. A combination of p-tau181 and NfL identified FTD with 82% sensitivity and 85% specificity and had a negative predictive value for neurodegenerative diagnosis of 86%, ruling out half of the non-neurodegenerative diagnoses. In the subcohort without AD biomarkers, similar results were obtained. T-tau and UCH-L1 did not offer added diagnostic value.

Discussion: Plasma p-tau181 predicted amyloid status with high accuracy and could have potentially avoided CSF/amyloid PET testing in approximately 60% of subjects in a memory clinic setting. NfL was useful for identifying FTD from non-neurodegenerative causes but behaved worse than p-tau181 in all other comparisons. Combining p-tau181 and NfL improved diagnostic performance for FTD and non-neurodegenerative diagnoses. However, the 14% false-negative results suggest that further improvement is needed before implementation outside memory clinics.

Classification of evidence: This study provides Class I evidence that plasma p-tau181 correlates with the presence or absence of AD and a combination of plasma p-tau181 and NfL correlates moderately well with a diagnosis of FTD.

Figure 1. Plasma Concentrations of P-Tau181, GFAP, NfL, and T-Tau in the Whole Cohort per Diagnostic Syndromic Group

(A-D) Box-and-whisker plots with the central horizontal box line showing the median plasma concentration (p-tau181 [A]; GFAP [B]; NfL [C]; t-tau [D]) in each diagnostic group and lower and upper box boundaries showing the 25th and 75th percentiles, respectively. Plasma biomarker concentrations between diagnostic groups were log10 transformed and compared using an analysis of covariance, adjusting for age and sex, and pairwise comparisons were assessed with the Bonferroni correction method. For visualization purposes, raw (not log10 transformed) biomarker concentrations were plotted in the graph, and the scale of the upper segment of the y axis was adjusted, as marked by a discontinuous line. Participants were represented by a different color depending on amyloid status (not performed/negative/positive). Two additional horizontal discontinuous lines were represented in A, corresponding to a plasma p-tau181 concentration of 0.89 and 1.92, which were the ones with an optimized negative and positive predictive value for β-amyloid status discrimination, respectively (see the Results section). *p < 0.05, **p < 0.01, and ***p < 0.001. Brackets indicate the diagnostic groups compared and a left-pointing arrow marks that all diagnostic categories to the left have the same statistical significance. Aβ−/+ = β-amyloid negative/positive; AD = Alzheimer disease; CJD = Creutzfeldt-Jakob disease; CU = cognitively unimpaired; FTD = frontotemporal dementia; GFAP = glial fibrillary acidic protein; LBD = dementia with Lewy bodies; MCI = mild cognitive impairment; NfL = neurofilament light chain; p-tau181 = tau phosphorylated at threonine 181; SCD = subjective cognitive decline; t-tau = total tau.

Figure 2. Plasma P-Tau181 and NfL Concentrations in the Subcohort With AD Biomarkers Performed per Diagnostic Group (Left) and Area Under the ROC Curve for Discrimination Between Diagnostic Etiologic Categories (Right)

(A and C) Box-and-whisker plots with the central horizontal box line showing the median plasma concentration (p-tau181 [A]; NfL [C]) in each diagnostic group and lower and upper box boundaries showing the 25th and 75th percentile, respectively. Plasma biomarker concentrations between diagnostic groups were log10 transformed and compared using an analysis of covariance, adjusting for age and sex, and pairwise comparisons were assessed with the Bonferroni correction method. For visualization purposes, raw (not log10 transformed) biomarker concentrations were plotted in the graph, and the scale of the upper segment of the y axis was adjusted for plasma NfL (C), as marked by a discontinuous line. Participants were represented by a different color depending on β-amyloid positivity or negativity, which was defined using our center definitions (see the Methods section). Two additional horizontal discontinuous lines were represented in A, corresponding to a plasma p-tau181 concentration of 0.89 and 1.92, which were the ones with an optimized negative and positive predictive value for β-amyloid status discrimination, respectively (see the Results section). (B and D) Diagnostic accuracy of plasma biomarkers to differentiate between each pair of diagnoses is represented by a heatmap of each AUC, with a value of 0.5 meaning no discrimination and a value of 1 a perfect discrimination. *p < 0.05, **p < 0.01, and ***p < 0.001. Brackets indicate the diagnostic groups compared and a left-pointing arrow marks that all diagnostic categories to the left have the same statistical significance. Aβ−/+ = β-amyloid negative/positive; AD = Alzheimer disease; AUC = area under the ROC curve; CU = cognitively unimpaired; FTD = frontotemporal dementia; LBD = dementia with Lewy bodies; NfL = neurofilament light chain; p-tau181 = tau phosphorylated at threonine 181; ROC = receiver operating curve; SND = suspected nondegenerative cognitive impairment.

Figure 3. Receiver Operating Characteristic Plots of Predictive Models and Individual Plasma Biomarkers for Distinct Clinical Scenarios and Scatter Plot for Clinical and β-Amyloid Status Classification Using Plasma NfL and P-Tau181

(A, B, and C) Receiver operating characteristic plots showing the area under the curve of the predicted probability of distinct logistic regression models and individual plasma biomarkers in the proposed practical scenarios, with 95% CI in brackets. The complete model included age, sex, APOE, MMSE, and the 5 plasma biomarkers, whereas the model without biomarkers included only age, sex, APOE, and MMSE. (D) Scatter plot showing the distribution of diagnoses (represented by shape) and β-amyloid status (represented by color) by plasma p-tau181 and NfL concentrations. Three cutoffs are represented in the scatter plot by dotted lines: plasma p-tau181 1.37 pg/mL (balanced cutoff for β-amyloid status discrimination), plasma NfL 11.3 pg/mL (balanced cutoff for FTD diagnosis when p-tau181 is below 1.37 pg/mL, for optimal differentiation of FTD from SND), and plasma NfL/p-tau181 ratio of 10.3 (balanced cutoff for FTD diagnosis when p-tau181 is equal or higher than 1.37 pg/mL, for optimal differentiation of FTD from AD). For visualization purposes, the scale of the upper segment of the y axis was adjusted, as marked by a discontinuous line. Aβ−/+ = β-amyloid negative/positive; AD = Alzheimer disease; CU = cognitively unimpaired; FTD = frontotemporal dementia; GFAP = glial fibrillary acidic protein; LBD = dementia with Lewy bodies; MMSE = Mini-Mental State Examination; NfL = neurofilament light chain; PM = parsimonious model; p-tau181 = tau phosphorylated at threonine 181; SND = suspected nondegenerative cognitive impairment.

Figure 4. Proposed Algorithm for Amyloid Status Prediction Using Plasma P-Tau181

Aβ−/+ = β-amyloid negative/positive; p-tau181 = tau phosphorylated at threonine 181.

Figure 5. Proposed Algorithm for FTD Diagnosis Using Plasma P-Tau181 and NfL

AD = Alzheimer disease; FTD = frontotemporal dementia; LBD = dementia with Lewy bodies; NfL = neurofilament light chain; p-tau181 = tau phosphorylated at threonine 181; SND = suspected nondegenerative cognitive impairment.