Abstract
Cost-effective strategies for identifying amyloid-β (Aβ) positivity in patients with cognitive impairment are urgently needed with recent approvals of anti-Aβ immunotherapies for Alzheimer’s disease (AD). Blood biomarkers can accurately detect AD pathology, but it is unclear whether their incorporation into a full diagnostic workflow can reduce the number of confirmatory cerebrospinal fluid (CSF) or positron emission tomography (PET) tests needed while accurately classifying patients. We evaluated a two-step workflow for determining Aβ-PET status in patients with mild cognitive impairment (MCI) from two independent memory clinic-based cohorts (n = 348). A blood-based model including plasma tau protein 217 (p-tau217), age and APOE ε4 status was developed in BioFINDER-1 (area under the curve (AUC) = 89.3%) and validated in BioFINDER-2 (AUC = 94.3%). In step 1, the blood-based model was used to stratify the patients into low, intermediate or high risk of Aβ-PET positivity. In step 2, we assumed referral only of intermediate-risk patients to CSF Aβ42/Aβ40 testing, whereas step 1 alone determined Aβ-status for low- and high-risk groups. Depending on whether lenient, moderate or stringent thresholds were used in step 1, the two-step workflow overall accuracy for detecting Aβ-PET status was 88.2%, 90.5% and 92.0%, respectively, while reducing the number of necessary CSF tests by 85.9%, 72.7% and 61.2%, respectively. In secondary analyses, an adapted version of the BioFINDER-1 model led to successful validation of the two-step workflow with a different plasma p-tau217 immunoassay in patients with cognitive impairment from the TRIAD cohort (n = 84). In conclusion, using a plasma p-tau217-based model for risk stratification of patients with MCI can substantially reduce the need for confirmatory testing while accurately classifying patients, offering a cost-effective strategy to detect AD in memory clinic settings.
Main
AD is the primary cause of dementia and is neuropathologically defined by the accumulation of extracellular Aβ plaques and intracellular tangles of hyperphosphorylated tau1,2,3. Established AD biomarkers are essential for patient management and will become increasingly important as disease-modifying treatments approach clinical practice4. New anti-Aβ therapies have shown promising results in clearing Aβ from the brain5,6,7, leading to approvals of aducanumab and lecanemab by the US Food and Drug Administration (FDA). Confirmation of underlying AD biomarker abnormalities will be key in determining eligibility for disease-modifying treatments in patients with cognitive impairment visiting memory clinics8. Nevertheless, the high cost, invasiveness, time-consuming nature and limited availability of CSF and PET biomarkers hamper their widespread use to screen for AD biomarker positivity in memory clinics.
Blood-based biomarkers hold promise to aid in delivering a biomarker-supported AD diagnosis in a minimally invasive and scalable manner4. Plasma p-tau species, including p-tau181, p-tau217 and p-tau231, have shown high performance to identify underlying AD9,10,11. Plasma p-tau217 (tau phosphorylated at Thr217) shows the highest fold-changes in Aβ-positive patients with cognitive impairment, thus being less susceptible to analytical variation10,12,13,14. Moreover, plasma p-tau217 is strongly associated with measures of Aβ pathology and its levels change before tau-PET abnormalities are detectable in AD progression15,16,17, making it a feasible candidate to implement as a routine clinical chemistry test to screen for Aβ positivity in memory clinics.
Nevertheless, the implementation of new AD blood biomarkers into a comprehensive diagnostic workflow for detecting Aβ positivity has received less attention, and the Alzheimer’s Association guidelines for appropriate use of AD blood biomarkers recently highlighted the need for objectively evaluating such a strategy18. Indeed, even the best-performing blood p-tau biomarkers present a higher group-level overlap than established CSF and PET biomarkers19,20. Consequently, handling their results more granularly could potentially reduce the burden of submitting most patients to confirmatory CSF or PET testing. In this context, a model-based approach for interpreting biomarkers alongside clinically relevant information, which is a common strategy in several medical areas21,22, might also be well suited when screening for AD23,24,25.
In two independent secondary memory clinic-based cohorts, we evaluated a two-step workflow for detecting brain amyloidosis (as indexed by Aβ-PET) in patients with MCI. Step 1 consisted of a diagnostic model based on plasma p-tau217, age and APOE ε4 (apolipoprotein E allele ε4) for risk stratification of Aβ-PET positivity. Step 2 was based on confirmatory testing with CSF Aβ42/Aβ40 only in those patients with uncertain outcomes at step 1. In secondary analyses, this workflow was evaluated using a different plasma p-tau217 immunoassay version in a third cohort, from a distinct geographical setting. We demonstrate that such a two-step workflow can lead to a reduction in the number of confirmatory Aβ tests needed while preserving a high overall accuracy for detecting Aβ-PET status.
Results
Participant characteristics
In total, we included 348 MCI participants from BioFINDER-1 (n = 136) and BioFINDER-2 (n = 212) (Supplementary Table 1). Frequencies of Aβ-PET positivity (BioFINDER-1, 60.3%; BioFINDER-2, 60.8%) and APOE ε4 carriership (BioFINDER-1, 49.3%; BioFINDER-2, 55.2%) were similar and both cohorts had fewer females (BioFINDER-1, 35.3%; BioFINDER-2, 42.0%). Included patients from the two cohorts presented similar Mini-Mental State Examination (MMSE) scores, ages and plasma p-tau217 levels (as measured by the Lilly Research Laboratories’ assay unless otherwise specified). Comorbidities were frequent, with frequencies in the combined population (n = 348) of 54.0% for cardiovascular disease, 15.8% for diabetes, 37.9% for dyslipidemia and 9.2% for chronic kidney disease (CKD).
Model development, validation and threshold definition
Plasma p-tau217, age and APOE ε4 status were evaluated as candidate predictors for developing a logistic regression model for Aβ-PET positivity with bootstrapped backward variable elimination in BioFINDER-1 (Supplementary Table 2). The full model, including plasma p-tau217, age and APOE ε4, was selected, presenting an optimism-corrected AUC of 89.3% (95% confidence interval (CI) = 83.7–93.8%) for Aβ-PET positivity in BioFINDER-1. At external validation in BioFINDER-2, an independent cohort, the model also presented high discriminatory performance (AUC = 94.3%, 95% CI = 91.2–97.4%). Next, three different thresholding strategies were explored to classify participants into groups with low, intermediate and high risk of Aβ-PET positivity, based on the plasma p-tau217 model-derived probabilities of Aβ-PET positivity. We defined lower probability thresholds with 90%, 95% and 97.5% sensitivity (to avoid missing detection of patients who are Aβ positive), and higher probability thresholds with 90%, 95% and 97.5% specificity (to avoid classifying patients who are Aβ negative as ‘high risk’). As the model validated well and displayed good calibration, probability thresholds were derived for the combined BioFINDER-1 and BioFINDER-2 dataset (n = 348) (Extended Data Fig. 1). Predicted probabilities of Aβ-PET positivity and the resulting thresholds are shown in Fig. 1a….
