Development and Clinical Validation of Blood-Based Multibiomarker Models for the Evaluation of Brain Amyloid Pathology

Abstract

Background and objectives: Plasma biomarkers provide new tools for evaluating patients with mild cognitive impairment (MCI) for Alzheimer disease (AD) pathology. Such tools are needed for anti-amyloid therapies that require efficient and accurate diagnostic evaluation to identify potential treatment candidates. This study sought to develop and evaluate the clinical performance of a multimarker combination of plasma beta-amyloid 42/40 (Aβ42/40), ptau-217, and APOE genotype to predict amyloid PET positivity in a diverse cohort of patients at a memory clinic and evaluate >4,000 results from "real-world" specimens submitted for high-throughput clinical testing.

Methods: Study participants were from the 1Florida AD Research Center. Demographics, clinical evaluations, and amyloid PET scan data were provided along with plasma specimens for model development in the intended-use cohort (MCI/AD: n = 215). Aβ42/40 and ApoE4 proteotype (reflecting high-risk APOE ɛ4 alleles) were measured by mass spectrometry and ptau-217 by immunoassay. A likelihood score model was determined for each biomarker separately and in combination. Model performance was optimized using 2 cutpoints, 1 for high and 1 for low likelihood of PET positivity, to attain ≥90% specificity and sensitivity. These cutpoints were applied to categorize 4,326 real-world specimens and an expanded cohort stratified by cognitive status (normal cognition [NC], MCI, AD).

Results: For the intended-use cohort (46.0% prevalence of PET positivity), a combination of Aβ42/40, ptau-217, and APOE4 allele count provided the best model with a receiver operating characteristic area under the curve of 0.942 and with 2 cutpoints fixed at 91% sensitivity and 91% specificity, yielding a high cutpoint with 88% positive predictive value and 87% accuracy and a low cutpoint with 91% negative predictive value and 85% accuracy. Incorporating the APOE4 allele count also reduced the percentage of patients with indeterminate risk from 15% to 10%. The cutpoints categorized the real-world clinical specimens as having 42% high, 51% low, and 7% indeterminate likelihood of PET positivity and differentiated between NC, MCI, and AD dementia cognitive status in the expanded cohort.

Discussion: Combining plasma biomarkers Aβ42/40, ptau-217, and APOE4 allele count is a scalable approach for evaluating patients with MCI for suspected AD pathology.

Figure 1. Model Performance of Single and Combined Biomarkers in Predicting PET Positivity Demonstrated by Receiver Operating Characteristic Curves

Receiver operating characteristic area under the curve (ROC-AUC) analysis of individual and combined biomarkers for prediction of PET status. Aβ = beta-amyloid; APOE4 = APOE ε4; AUC = area under the curve.

Figure 2. Comparison of Models Using Boxplots With Dashed Lines Representing Cutoffs for Predicting Low, Indeterminate, or High Likelihood of PET Positivity Using Single and Combined Biomarkers

(A) ptau-217, (B) Aβ42/40, (C) Aβ42/40 + ptau-217, (D) Aβ42/40 + ptau-217 +APOE4 carrier status, and (E) Aβ42/40 + ptau-217 + APOE4 allele count. Orange dots represent individuals with MCI; blue dots represent individuals with AD. Horizontal dashed lines indicate high-likelihood and low-likelihood cutoffs. Results between these lines indicate neither a low nor a high likelihood (i.e., an indeterminate result).

Figure 3. Comparison of the Study Cohort With 4,000+ Real-World Clinical Specimens Using Beeswarm Plots for the Distribution of Likelihood of PET Positivity Using the Full Biomarker Model

Distribution of model results for (A) the intended-use population and (B) real-world clinical specimens. Orange dots represent individuals who are amyloid PET negative; blue dots represent those who are amyloid PET positive in panel A. Gray dots represent individuals whose amyloid PET status is unknown. Horizontal dashed lines indicate high-likelihood and low-likelihood cutoffs.