Clinical performance and robustness of blood‐based biomarkers for early detection of amyloid pathology associated with Alzheimer’s disease

Abstract

AbstractBackgroundAccurate and robust blood‐based biomarkers (BBBMs) of Alzheimer’s disease (AD) are required for identification of symptomatic patients with low likelihood of amyloid pathology before confirmatory diagnostic evaluation. Further evidence on the clinical performance and robustness of BBBMs is required to identify patients for clinical trials and in routine clinical practice. We evaluated the clinical performance and robustness of amyloid‐β 1–42 (Aβ42), amyloid‐β 1–40 (Aβ40), apolipoprotein E4 (ApoE4), phosphorylated‐tau 181 (pTau181), glial fibrillary acidic protein (GFAP), and neurofilament light chain (NFL) as BBBMs of AD.MethodAnalyses were performed retrospectively using plasma samples from AIBL, BioFINDER, and CREAD cohorts, comprising cognitively normal individuals and patients with subjective/objective cognitive impairment, mild AD, or amyloid screen failure. BBBMs were measured at two laboratories using automated Elecsys® prototype immunoassays (cobas e 601 and e 411 analyzers; all Roche Diagnostics International Ltd). Clinical performance of single BBBMs and combinations of BBBMs (using logistic regression) was assessed using area under the receiver operating characteristic curve (AUC‐ROC) analysis. Negative percent agreement (NPA), prevalence‐adjusted negative and positive predictive values (NPV and PPV), and screen‐out rate were compared at 85% positive percent agreement (PPA). Robustness, defined as the change in clinical performance when adding ±10% bias and random variability, was calculated at 85% PPA.ResultAcross cohorts, the best performing single BBBM was pTau181 (AUC: 81.6–89.3). The best performing combined BBBM models across cohorts were pTau181+Aβ42 (AUC: 85.6–93.4), pTau181+ApoE4 (AUC: 83.7–91.8), and pTau181+Aβ40 (AUC: 84.8–90.2); clinical performance for these combined BBBM models in terms of NPV and PPV was comparable, providing high NPVs (>90%) and screen‐out rates (>50%). For combined BBBM models including Aβ42 or Aβ40, robustness at 85% PPA was lower compared with models not including Aβ42 or Aβ40 across cohorts. Considering both clinical performance and robustness, pTau181 and pTau181+ApoE4 were the best performing single and combined BBBM models, respectively, across cohorts.ConclusionAcross cohorts, pTau181+ApoE4 was the best performing BBBM model for detecting patients with low likelihood of amyloid pathology. These findings support the suitability of these BBBMs to inform diagnostic assessments for patients with low likelihood of AD.