This cohort study analyzes data from participants with cognitive concerns to determine whether plasma biomarkers can identify amyloid β–positive individuals who do not yet have a high tau burden in the brain and would be suitable candidates for treatment with antiamyloid immunotherapies in Alzheimer disease.
Key Points
Question
Can plasma biomarkers serve as prescreening tools to identify individuals suitable for treatment with antiamyloid immunotherapies in Alzheimer disease, ie, amyloid β (Aβ)–positive individuals who do not yet have too much tau in the brain?
Findings
This cohort study demonstrated the efficacy of using plasma p-tau217 values to both rule out Aβ-negative individuals and identify Aβ-positive individuals who are likely to have high tau burden.
Meaning
Plasma p-tau217 measurement may be used as a first-in-line screening tool in management of patients with cognitive concerns, specifically for decision-making about treatment with antiamyloid immunotherapies in Alzheimer disease while limiting unnecessary invasive and costly biomarker testing.
Abstract
Importance
Antiamyloid immunotherapies against Alzheimer disease (AD) are emerging. Scalable, cost-effective tools will be needed to identify amyloid β (Aβ)–positive patients without an advanced stage of tau pathology who are most likely to benefit from these therapies. Blood-based biomarkers might reduce the need to use cerebrospinal fluid (CSF) or positron emission tomography (PET) for this.
Objective
To evaluate plasma biomarkers for identifying Aβ positivity and stage of tau accumulation.
Design, Setting, and Participants
The cohort study (BioFINDER-2) was a prospective memory-clinic and population-based study. Participants with cognitive concerns were recruited from 2017 to 2022 and divided into a training set (80% of the data) and test set (20%).
Exposure
Baseline values for plasma phosphorylated tau 181 (p-tau181), p-tau217, p-tau231, N-terminal tau, glial fibrillary acidic protein, and neurofilament light chain.
Main Outcomes and Measures
Performance to classify participants by Aβ status (defined by Aβ-PET or CSF Aβ42/40) and tau status (tau PET). Number of hypothetically saved PET scans in a plasma biomarker–guided workflow.
Results
Of a total 912 participants, there were 499 males (54.7%) and 413 females (45.3%), and the mean (SD) age was 71.1 (8.49) years. Among the biomarkers, plasma p-tau217 was most strongly associated with Aβ positivity (test-set area under the receiver operating characteristic curve [AUC] = 0.94; 95% CI, 0.90-0.97). A 2–cut-point procedure was evaluated, where only participants with ambiguous plasma p-tau217 values (17.1% of the participants in the test set) underwent CSF or PET to assign definitive Aβ status. This procedure had an overall sensitivity of 0.94 (95% CI, 0.90-0.98) and a specificity of 0.86 (95% CI, 0.77-0.95). Next, plasma biomarkers were used to differentiate low-intermediate vs high tau-PET load among Aβ-positive participants. Plasma p-tau217 again performed best, with the test AUC = 0.92 (95% CI, 0.86-0.97), without significant improvement when adding any of the other plasma biomarkers. At a false-negative rate less than 10%, the use of plasma p-tau217 could avoid 56.9% of tau-PET scans needed to identify high tau PET among Aβ-positive participants. The results were validated in an independent cohort (n = 118).
Conclusions and Relevance
This study found that algorithms using plasma p-tau217 can accurately identify most Aβ-positive individuals, including those likely to have a high tau load who would require confirmatory tau-PET imaging. Plasma p-tau217 measurements may substantially reduce the number of invasive and costly confirmatory tests required to identify individuals who would likely benefit from antiamyloid therapies.
Introduction
The development of antiamyloid disease-modifying therapies for Alzheimer disease (AD)1,2 is revolutionizing patient management. Given their costs, adverse effects, and modest clinical benefits, these therapies should be directed to patients where they are likely to be most effective. Within this context, the phase 2 TRAILBLAZER trial in early AD (for the antiamyloid antibody donanemab) implemented an inclusion strategy where eligible participants had to be positive for amyloid β (Aβ) based on positron emission tomography (PET) and were required to not have too much tau pathology as assessed by tau PET.3 Recent results from the phase 3 trial TRAILBLAZER-ALZ2 confirm that donanemab is less effective in participants with cognitive impairment and high tau burden and that greater clinical benefit might be achieved in participants with less tau.4 Consequently, we focus on how scalable and cost-effective biomarkers can identify individuals with Aβ pathology, but without a high tau burden.
Aβ and tau may be detected with PET or cerebrospinal fluid (CSF) biomarkers, but these tests are limited by cost, low availability, and invasiveness. Blood-based biomarkers are better suited for screening5 and have been implemented in clinical trials. For example, the AHEAD 3-45 (lecanemab) used plasma Aβ42/40 to rule out individuals unlikely to be Aβ-positive and limit the number of CSF or PET. Recent studies have proposed phosphorylated tau 217 (p-tau217) for prescreening6 because it is elevated in Aβ-positive individuals compared with Aβ-negative7,8,9 and shows agreement with tau PET.7,9,10,11,12 However, there is a lack of studies testing blood-based biomarkers to specifically identify high tau-PET burden representing advanced disease stages where the benefit-risk ratio of antiamyloid immunotherapies may be lower.3
We investigated whether blood-based biomarkers (p-tau181, p-tau217, p-tau231, N-terminal tau, glial fibrillary acidic protein, and neurofilament light chain) can be used for prescreening in patients with cognitive concerns to identify those likely to (1) be Aβ-positive (a requirement for antiamyloid treatment) and (2) have high tau-PET burden (less likely to respond well to antiamyloid therapy, reducing the benefit-risk ratio).
Methods
Participants
We included participants with subjective cognitive decline (SCD), mild cognitive impairment, or dementia from the Swedish BioFINDER 2 study (NCT03174938) who were enrolled between April 2017 and March 2022. Inclusion and exclusion criteria have been described elsewhere10 and are summarized in the eMethods in Supplement 1.
All participants gave written informed consent. Ethical approval was given by the Swedish Ethical Review Authority. Approval for PET imaging was obtained from the Swedish Medicines and Products Agency and the local Radiation Safety Committee at Skåne University Hospital in Sweden. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Plasma and CSF Biomarkers
Plasma and CSF p-tau217 were measured with an Eli Lilly assay on a Meso Scale Discovery platform.10 Plasma p-tau181,13 p-tau231,14 neurofilament light chain15 and glial fibrillary acidic protein,16 and N-terminal tau17,18 were measured using Simoa assays. CSF Aβ42 and Aβ40 were measured with the Elecsys assays on a fully automated cobas e601 instrument (Roche Diagnostics). Aβ positivity was defined as a CSF Aβ42/40 level less than 0.08.19
Amyloid and Tau-PET Acquisition and Quantification
Aβ-PET imaging was performed on digital GE Discovery MI scanners, with 4 frames of 5 minutes acquired 90 to 110 minutes after the injection of approximately 185 MBq of [18F]flutemetamol. Global standardized uptake value ratios (SUVRs) were calculated with a composite reference region. Aβ-PET status was based on a cutoff of 0.693.20
[18F]RO948 tau PET was performed on the same scanner, acquired 70 to 90 minutes after injection of approximately 370 MBq of [18F]RO948.21 SUVR images were created using the inferior cerebellar cortex as reference region. Tau burden was determined within the multiblock barycentric discriminant analysis (MUBADA) region of interest, which is a voxelwise map that optimally discriminates between Aβ-positive patients with clinically defined mild cognitive impairment and AD dementia against Aβ-negative cases.22 The region of interest includes voxels from all brain lobes, with higher weights assigned to voxels in the posterior and lateral temporal lobe, parietal lobe, association regions of the occipital cortex, and the left hemisphere. Participants were classified as having low burden (SUVR-MUBADA < 1.10), intermediate burden (SUVR-MUBADA 1.10-1.46), or high burden (SUVR-MUBADA > 1.46)22 (eMethods in Supplement 1). Cases with high meningeal and low brain uptake (ratio >1.75,23 n = 16) were reclassified as low tau, irrespective of their SUVR-MUBADA. Visual assessment confirmed high meningeal and absent or low brain uptake in all reclassified cases.
Validation Cohort
We included in a validation cohort 118 participants from the Skåne University Hospital flortaucipir study for whom all data described above were available. Tau PET was performed using the [18F]flortaucipir radiotracer, acquired 80 to 100 minutes postinjection (eResults in Supplement 1).
Statistical Analysis
To predict binary outcomes, biomarkers were used as predictors in logistic regression analyses or machine learning classification models (random forest classification, extreme gradient boosting, and support vector machine with radial basis function kernel). Linear regression was used to predict continuous tau-PET uptake together with random forest and extreme gradient boosting regression models. Age, sex, and APOE ε4 status were evaluated as covariates. The caret package was used with 10-fold cross-validation on the training set to select hyper parameters (eMethods in Supplement 1). Area under the receiver operating characteristic curves (AUC) was compared using the DeLong test. Bootstrapping was used to calculate 95% CI for cut-point performance metrics (eg, sensitivities and specificities). The Youden index was used to determine optimal single cut points. R version 4.3.1 (R Foundation) was used for all analyses, with statistical significance at P < .05.
Results
A total of 912 study patients were randomly assigned to a training (80%) or test set (20%). There were 499 males (54.7%) and 413 females (45.3%), and the mean (SD) age was 71.1 (8.49) years. Additional demographic data are shown in the Table.
Table. Demographic Dataa.
| Characteristic | Aβ-negative patients, No. (%) | Aβ-positive patients, No. (%) | Overall (n = 912) | |||
|---|---|---|---|---|---|---|
| Low tau PET (n = 349) | Intermediate tau PET (n = 9) | Low tau PET (n = 293) | Intermediate tau PET (n = 145) | High tau PET (n = 116) | ||
| Clinical diagnosis | ||||||
| SCD | 120 (34.4) | 3 (33.3) | 77 (26.3) | 11 (7.6) | 4 (3.4) | 215 (23.6) |
| MCI | 150 (43.0) | 5 (55.6) | 128 (43.7) | 51 (35.2) | 21 (18.1) | 355 (38.9) |
| Dementia | 79 (22.6) | 1 (11.1) | 88 (30.0) | 83 (57.2) | 91 (78.4) | 342 (37.5) |
| MMSE, mean (SD) | 27.0 (2.93) | 27.0 (4.33) | 26.2 (3.68) | 23.7 (4.59) | 20.9 (5.03) | 25.4 (4.30) |
| Age, mean (SD), y | 68.7 (9.63) | 69.4 (6.80) | 73.6 (6.89) | 73.6 (6.62) | 69.4 (8.17) | 71.1 (8.49) |
| Sex | ||||||
| Male | 209 (59.9) | 5 (55.6) | 168 (57.3) | 73 (50.3) | 44 (37.9) | 499 (54.7) |
| Female | 140 (40.1) | 4 (44.4) | 125 (42.7) | 72 (49.7) | 72 (62.1) | 413 (45.3) |
| Comorbiditiesb | ||||||
| Cardiovascular | 178 (51.0) | 4 (44.4) | 178 (60.8) | 74 (51.0) | 51 (44.0) | 485 (53.2) |
| Diabetes | 53 (15.2) | 4 (44.4) | 45 (15.4) | 14 (9.7) | 14 (12.1) | 130 (14.3) |
| Dyslipidemia | 112 (32.1) | 5 (55.6) | 121 (41.3) | 48 (33.1) | 30 (25.9) | 316 (34.6) |
| CKD | 10 (2.9) | 0 | 20 (6.8) | 8 (5.5) | 5 (4.3) | 43 (4.7) |
| p-Tau217, mean (SD), ng/L | 0.15 (0.07) | 0.23 (0.09) | 0.33 (0.16) | 0.65 (0.31) | 0.90 (0.36) | 0.38 (0.33) |
| Tau-PET burden (MUBADA ROI), mean (SD) | 0.99 (0.06) | 1.17 (0.05) | 1.01 (0.06) | 1.24 (0.11) | 1.70 (0.19) | 1.12 (0.26) |
Abbreviations: Aβ, amyloid β; CKD, chronic kidney disease; MCI, mild cognitive impairment; MMSE, Mini-Mental State Examination; MUBADA, multiblock barycentric discriminant analysis; PET, positron emission tomography; p-tau217, phosphorylated tau 217; ROI, region of interest; SCD, subjective cognitive decline.
eTable 1 in Supplement 1 depicts demographics for training and test sets, which were comparable in terms of age, sex, and cognitive status.
Cardiovascular burden includes the presence of hypertension, ischemic heart disease, stroke, and/or use of antihypertension drugs. Dyslipidemia burden includes the presence of hyperlipidemia and/or use of drugs against hyperlipidemia. CKD is based on sex-specific cutoffs of creatinine.
Aβ Status and Tau-PET Uptake
Aβ status was defined using Aβ-PET if available (n = 506) and otherwise using CSF Aβ42/40 (n = 406). Among Aβ-negative participants (n = 358), 349 (97.5%) had low tau PET while 9 (2.5%) had intermediate tau PET. Among Aβ-positive participants (n = 554), 293 (52.9%) had low, 145 (26.2%) intermediate, and 116 (20.9%) high tau PET (training- and test-set data in eTable 1 in Supplement 1). Mean tau-PET burdens per biomarker groups are shown in eFigures 1 and 2 in Supplement 1.
Plasma Biomarkers to Identify Aβ Status
We trained plasma biomarker-based models to rule out Aβ-negative participants (Figure 1A). The best individual plasma biomarker was plasma p-tau217 (test-set AUC = 0.94; 95% CI, 0.90-0.97). The machine learning approaches using all plasma biomarkers had similar performance as p-tau217 alone (eg, extreme gradient boosting: test AUC = 0.95; 95% CI, 0.92-0.98; P = .25) (eTable 2 in Supplement 1). In the p-tau217 model, both age and APOE ε4 were significantly associated with Aβ status, but the test AUC was not reduced when p-tau217 was used as a stand-alone predictor (test AUC = 0.92; 95% CI, 0.88-0.96; P = .10). We therefore proceeded with plasma p-tau217 alone. At a Youden index–defined cut point (0.22 ng/L, defined in the training set), plasma p-tau217 had a sensitivity of 0.84 (95% CI, 0.77-0.90) and specificity of 0.88 (95% CI, 0.79-0.95) for Aβ positivity in the test set.
Figure 1. Prediction of Amyloid β (Aβ) Status With Plasma Biomarkers.
A, Predicted Aβ status was verified by CSF or PET. B, Dashed lines indicate the 2 cut points. Low plasma p-tau217 predicted Aβ negativity, and high plasma p-tau217 predicted Aβ positivity. Intermediate (gray-zone) levels would require CSF or PET to confirm Aβ status. C and D, Dashed line indicates the cut point for a positive Aβ-PET scan. E and F, Boxes show percentages of data below the lower cut point, in the gray zone, and above the upper cut point. ML indicates machine learning; RF, random forest; SVM, support vector machine; SUVR, standardized uptake volume ratio; XGB, extreme gradient boosting.
Two–Cut-Point Strategy for Plasma p-Tau217 to Define Aβ Status
A 2–cut-point approach could be used to improve the accuracy of diagnostic AD biomarkers, where values are stratified into 3 categories, ie, normal, indeterminant (a “gray zone”), and abnormal.6,24 The gray zone should include as few participants as possible (these need confirmatory testing to determine Aβ status25). We defined a gray zone by the cut points 0.159 ng/L and 0.219 ng/L with the a priori goal to minimize the gray zone while achieving at least 90% sensitivity and 90% specificity (Figure 1B and eResults in Supplement 1). The gray-zone population represented a mix of Aβ-negative and Aβ-positive cases (Figure 1C and D). The approach had a sensitivity/specificity of 0.94 (95% CI, 0.92-0.96)/0.90 (95% CI, 0.87-0.94), with only 13.3% (95% CI, 10.5%-15.7%) of participants being in the gray-zone range in the training set. In the test set, the sensitivity/specificity was 0.94 (95% CI, 0.90-0.98)/0.86 (95% CI, 0.77-0.95), and 17.1% (95% CI, 12.1%-23.2%) of participants were in the gray zone (Figure 1E and F). The strategy classified 28.2% of (test-set) participants as true negatives, 4.4% as false positives, 3.9% as false negatives, and 63.5% as true positives (Figure 1G and H). In sum, the approach identified Aβ positivity with less than 5% false positives and false negatives, with a minority (17%) requiring CSF or PET to determine Aβ status.
Prediction of Continuous Tau-PET Load Among Patients Predicted to Be Aβ Positive
Participants predicted to be Aβ-positive by p-tau217 (training set, n = 435; test set, n = 123) were included to predict the continuous tau-PET burden using linear regression models (individual plasma biomarkers) and machine learning models (eFigure 4 and eTable 3 in Supplement 1). We observed modest performances (R2 = 0.58 for p-tau217; R2 = 0.66 for random forest regression), suggesting that the biomarkers were not sufficiently accurate to estimate the continuous tau-PET burden in a robust way.
Prediction of Having a High Tau-PET Load in Patients Predicted to Be Aβ Positive
We proceeded with evaluating whether plasma biomarkers could be used to identify Aβ-positive participants with a high likelihood of high tau-PET uptake. The machine learning approaches had comparable AUCs (Figure 2A) with plasma p-tau217 alone (test-set AUC = 0.92; 95% CI, 0.86-0.97) (eTable 4 in Supplement 1). In the p-tau217 model, age was significantly associated with high tau-PET load. The test AUC was not reduced when p-tau217 was used as a stand-alone predictor (test AUC = 0.90; 95% CI, 0.84-0.95; P = .26 for difference). Therefore, p-tau217 alone was used for further analyses.
Figure 2. Prediction of High Tau via Positron Emission Tomography (PET) Imaging Among Predicted Amyloid β (Aβ)–Positive Participants.
The false-negative rate (FNR) is the proportion of positive cases with low plasma biomarker levels, and “saved scans” are the cases with low plasma biomarker levels that would not have tau PET in our hypothetical workflow. D and E, Dashed lines indicate cut points for FNRs of 5%, 10%, and 20% in the training set. AUC indicates area under the receiver operating characteristic curve; GFAP, glial acidic fibrillary protein; ML, machine learning; NFL, neurofilament light protein; NTA, N-terminal tau; RF, random forest; SVM, support vector machine; XGB, extreme gradient boosting.
Figure 2B shows rates of true and false classifications of high tau-PET load over the range of plasma p-tau217, together with the rate of saved tau-PET scans. A higher number of saved tau-PET scans in this Aβ-positive population comes at the price of more false negatives (Figure 2C). For example, the range of saved PET scans goes from 40.2% (the proportion of participants with plasma p-tau217 below the lowest level [0.376 ng/L] among those with high tau-PET load; those with lower plasma p-tau217 do not need tau PET) to 100% (at the highest measured concentration for p-tau217 [1.96 ng/L], ie, no tau PET, at the cost of many false negatives). Figure 2D and E shows histograms of plasma p-tau217 with cut points for a range of false-negative rates (FNRs) from 5% to 20%. The frequencies of saved tau-PET scans and the observed FNRs did not indicate overfitting (Figure 2F). A cut point for a 10% FNR (10% of those with high tau are below this cut point of 0.499 ng/L) in the training set resulted in 56.9% saved tau-PET scans in the test set, compared with doing tau-PET imaging for all predicted Aβ-positive participants. The eResults in Supplement 1 contains details on true- and false-positive and true- and false-negative scans.
In participants for whom CSF p-tau217 was available (training set, n = 223; test set, n = 61), similar AUC was achieved to predict high tau-PET load between the 2 fluid biomarkers (plasma: test AUC = 0.92; 95% CI, 0.85-0.99; CSF: test AUC = 0.85; 95% CI, 0.73-0.96; P = .26), showing that plasma p-tau217 was noninferior in identifying high tau PET among Aβ-positive participants.
Sensitivity Analyses, Hypothetical Cost Savings, and Validation Cohort
The results were similar when excluding patients with SCD (eResults and eFigures 5-6 in Supplement 1). Compared with a conventional workflow with Aβ-PET in all patients and tau PET in those who are Aβ-positive, our novel approach can lead to cost savings of approximately 60% to 70%, depending on PET and blood test costs. The eResults and eTable 5 in Supplement 1 provide a detailed analysis.
We validated plasma p-tau217 cut points in an independent cohort of 118 participants (eTable 6 in Supplement 1). The gray-zone strategy for Aβ status resulted in 10.2% of the cohort requiring CSF or PET to determine Aβ status, with 6.8% false negatives and 2.5% false positives. The cut point tuned for an FNR of 5% for high tau PET among the predicted Aβ-positive participants had an FNR of 5.0% in the validation set and resulted in 46.6% saved scans, compared with doing tau-PET imaging for all predicted Aβ-positive participants (eResults in Supplement 1).
Discussion
In this study, we demonstrated the efficacy of using plasma p-tau217 to both rule out Aβ-negative individuals and identify Aβ-positive individuals who are likely to have high tau burden (Figure 3). Plasma p-tau217 measurements can substantially reduce the number of invasive and costly confirmatory tests required to identify individuals who would likely benefit from antiamyloid therapies. Our findings have implications for streamlining patient management while minimizing the need for lumbar puncture and PET (Figure 4).
Figure 3. Overview of Procedures for Determining Amyloid β (Aβ) Status and High Tau via Positron Emission Tomography (PET) Imaging.
This figure summarizes information from the different steps of our approach with plasma phosphorylated tau 217 (p-tau217) as a biomarker predicting Aβ and tau. The histogram is grouped by combinations of observed Aβ status and tau-PET status. The 2 cut points used to define the Aβ gray zone are shown (gray dashed lines). Plasma p-tau217 below the lower gray-zone cut point predicts negative Aβ status. Plasma p-tau217 above the upper gray-zone cut point predicts positive Aβ status. Confirmatory testing with cerebrospinal fluid or PET will be needed to assign Aβ status for individuals in the gray zone. Cut points to determine high tau PET among predicted Aβ-positive patients are shown for false-negative rates of 5%, 10%, and 20% (black dashed lines).
Figure 4. Standard and Proposed Novel Workflows for Testing for Amyloid β (Aβ) Status With Positron Emission Tomography (PET), Cerebrospinal Fluid (CSF), and Phosphorylated Tau 217 (p-Tau217).
In a hypothetical standard workflow, all patients would undergo PET or CSF tests for Aβ status; those determined to be Aβ positive would undergo tau PET imaging to determine whether they have a suitable range of tau uptake. In our proposed workflow, all patients would undergo tests with p-tau217. Only individuals with ambiguous levels of plasma p-tau217 would undergo Aβ tests with PET or CSF. Of those Aβ-positive, only individuals likely to have high tau uptake based on plasma p-tau217 levels would undergo tau PET. Aβ-positive individuals with high tau may not benefit from treatment; the benefits and risks of treatment should be considered carefully.
As clinical evaluations have limited sensitivity and specificity, biomarker-based diagnosis is key to selecting patients for antiamyloid immunotherapies. Biomarker evidence of Aβ pathology is needed because it is a defining characteristic of AD pathophysiology26 and the target of antiamyloid immunotherapies. Patients with cognitive impairment who are Aβ-negative most likely do not have AD2,3,27,28 and should not receive antiamyloid immunotherapies but be evaluated for differential diagnoses.26 With the attributes of being noninvasive, cost-effective, and scalable, blood-based biomarkers could facilitate testing on a large scale. We built on previous literature7,8,9 showing that plasma p-tau217 can be used to identify Aβ positivity with high sensitivity and specificity. When using the novel 2–cut-point strategy in this study, only a subpopulation (~17%) of patients with ambiguous plasma p-tau217 levels would be referred to CSF or PET to determine Aβ status.
Given the risk of serious adverse events, high costs, and likely limited access to antiamyloid immunotherapies, it will be critical to identify not only Aβ-positive individuals but also those patients with AD who have the highest likelihood to respond to treatment. Successful antiamyloid immunotherapies reduce slopes of cognitive decline by 25% to 35%,2,4 but the beneficial effects are lower in patients with AD and high tau load3,4 without a change in risk for adverse effects, reducing the benefit-risk ratio. About 13% to 24% of patients given antiamyloid disease-modifying therapies develop amyloid-related imaging abnormalities (ARIA), which may lead to severe symptoms, including seizures, hospitalization, and even death. In a donanemab trial, 1.6% of treated patients had ARIA with severe outcomes.4 Three patients (among 860 with active treatment) died after serious ARIA, and similar events have happened after treatment with lecanemab.29,30 Given these risks, it is important to consider the risk-benefit ratio before starting treatment. This will be important at a group level for payers in many countries when deciding about reimbursement for these new treatments, but also at an individual level for patients and clinicians when deciding on initiating antiamyloid therapy or not. We therefore set out to identify Aβ-positive patients with AD who do not have biomarker evidence of high tau load, who will in general have a more favorable benefit-risk ratio. While plasma p-tau217 could only marginally predict continuous tau-PET SUVRs, it has the capacity to identify those likely to have high tau-PET uptake with high accuracy. These patients can subsequently be referred to confirmatory tau-PET testing and support informed decision-making for treatment initiation.
This procedure could dramatically reduce the number of required tau-PET scans (~57% saved at a 10% FNR) compared with an agnostic approach where every Aβ-positive patient undergoes tau-PET imaging. We believe the 10% FNR is acceptable because there is still a weak effect of treatment in the high tau group. In clinics with high access to tau PET, lower FNR cut points (eg, 5%) may be preferable. If there is very low access to tau PET, higher FNR (eg, 20%) may be feasible. Another issue to consider is whether patients with AD and low tau should be given treatment. In one trial, participants with no or very limited tau burden were excluded to increase the likelihood of finding treatment effects.4 However, in clinical practice, we do not believe that it is meaningful or even ethical to withhold antiamyloid treatment from patients with AD who have not yet accumulated sufficiently high tau burden.
Among several plasma biomarkers, p-tau217 was superior at the symptomatic stage of the disease to detect Aβ positivity (this may differ in presymptomatic stages31) and high tau load. Our findings add to previous examples of suitable properties for plasma p-tau217 as a screening biomarker, including low test-retest variability, increased levels in Aβ-positive individuals compared with Aβ-negative,7,8,9 agreement with tau PET,7,9,10,11,12 and associations with postmortem AD hallmarks.32 Nonetheless, combining p-tau217 with other p-tau biomarkers more specifically associated with tau-PET changes, such as p-tau205 and MTBR-tau243 for early- and late-stage changes, respectively, could further streamline eligibility workflows.33 These biomarkers are not available in plasma yet but hold promise as staging biomarkers and are of interest for future research. The increase in availability of plasma biomarkers (p-tau and other pathologies) enables us to further identify profiles of biomarker abnormality, which could be differentially related to risk of subsequent clinical progression.34
Taken together, plasma p-tau217 might be suitable for determining the presence of Aβ and the severity of tau pathology. Our algorithm reduces the need for CSF or PET by approximately 83% to identify Aβ positivity, reduces the need for tau PET by 46% at low false-negative rates (few individuals with high tau load are misclassified), and could lead to cost savings up to 70%, depending on local health care system and payment models. The ability to conduct large-scale screening using plasma therefore sets a practical trajectory for incorporating p-tau217 into clinical practice. Potential challenges of such implementations include standardization of the sample collection, storage, and analysis and the availability of the proper facilities. In addition, implementation of the 2–cut-point approach would require local calibration or deriving in-house cut points.
Limitations and Strengths
One study limitation is that the participants were consecutively recruited patients with cognitive concerns seen at our memory clinics, and our findings may not be generalizable to primary care or ethnically more diverse populations. Still, the population was heterogenous, including many patients with low education level or comorbidities. We included patients with SCD, who are typically not included in trials for amyloid-lowering therapies. However, this allowed us to capture the full clinical continuum as encountered in routine diagnostic workups. Importantly, a sensitivity analysis excluding SCD resulted in highly similar results, supporting the robustness of our findings. Another limitation is that p-tau levels may be influenced by patient-specific factors35 and performance of different assays.36 Strengths of the study include the use of a rigorous approach with independent training and testing, which promotes a high level of confidence in the results, and the inclusion of a validation cohort, which yielded similar results to the test set of the main data set, supporting our findings.
Conclusions
The strategy used in this study presents an innovative diagnostic process for conditions associated with Aβ and tau pathologies. As scalable and cost-effective alternatives, plasma p-tau217 algorithms can accurately identify Aβ-positive individuals as well as those Aβ-positive who are likely to have a high tau load and need to undergo tau-PET imaging. This procedure avoids Aβ-PET or CSF in most individuals (80%-85%) and substantially reduces the need of tau PET by 57% at a 10% FNR to identify individuals who will benefit from antiamyloid immunotherapies.
eMethods
eResults
eReferences
eTable 1. Demographics by training and test sets
eTable 2. Results for models to predict Aβ-positivity
eTable 3. Results for models to predict continuous Tau PET SUVR
eTable 4. Results for models to predict high Tau PET
eTable 5. Hypothetical cost savings by novel work-flow
eTable 6. Hypothetical cost savings by novel work-flow
eFigure 1. Gaussian mixture model for MUBADA ROI tau-PET
eFigure 2. Average [18F]RO948 tau-PET uptake across Aβ and Tau PET classes
eFigure 3. Gray-zones for plasma p-tau217 to predict Aβ-PET positivity
eFigure 4. Predict continuous tau-PET
eFigure 5. Aβ-status is predicted with plasma biomarkers, without SCD
eFigure 6. Predict high Tau-PET among predicted Aβ-positive participants, without SCD
eFigure 7. Performance of plasma P-tau217 in a validation cohort
Data sharing statement
References
- 1.Mintun MA, Larossa GN, Sheline YI, et al. [11C]PIB in a nondemented population: potential antecedent marker of Alzheimer disease. Neurology. 2006;67(3):446-452. doi: 10.1212/01.wnl.0000228230.26044.a4 [DOI] [PubMed] [Google Scholar]
- 2.van Dyck CH, Swanson CJ, Aisen P, et al. Lecanemab in early Alzheimer’s disease. N Engl J Med. 2023;388(1):9-21. doi: 10.1056/NEJMoa2212948 [DOI] [PubMed] [Google Scholar]
- 3.Mintun MA, Lo AC, Duggan Evans C, et al. Donanemab in early Alzheimer’s disease. N Engl J Med. 2021;384(18):1691-1704. doi: 10.1056/NEJMoa2100708 [DOI] [PubMed] [Google Scholar]
- 4.Sims JR, Zimmer JA, Evans CD, et al. ; TRAILBLAZER-ALZ 2 Investigators . Donanemab in early symptomatic Alzheimer disease: the TRAILBLAZER-ALZ 2 randomized clinical trial. JAMA. 2023;330(6):512-527. doi: 10.1001/jama.2023.13239 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Hansson O. Biomarkers for neurodegenerative diseases. Nat Med. 2021;27(6):954-963. doi: 10.1038/s41591-021-01382-x [DOI] [PubMed] [Google Scholar]
- 6.Cullen NC, Janelidze S, Mattsson-Carlgren N, et al. Test-retest variability of plasma biomarkers in Alzheimer’s disease and its effects on clinical prediction models. Alzheimers Dement. 2022;19(3):797-806. doi: 10.1002/alz.12706 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Doré V, Doecke JD, Saad ZS, et al. Plasma p217+tau versus NAV4694 amyloid and MK6240 tau PET across the Alzheimer’s continuum. Alzheimers Dement (Amst). 2022;14(1):e12307. doi: 10.1002/dad2.12307 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Thijssen EH, La Joie R, Strom A, et al. ; Advancing Research and Treatment for Frontotemporal Lobar Degeneration investigators . Plasma phosphorylated tau 217 and phosphorylated tau 181 as biomarkers in Alzheimer’s disease and frontotemporal lobar degeneration: a retrospective diagnostic performance study. Lancet Neurol. 2021;20(9):739-752. doi: 10.1016/S1474-4422(21)00214-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Jonaitis EM, Janelidze S, Cody KA, et al. Plasma pTau-217 in preclinical Alzheimer’s disease. medRxiv. Published online June 14, 2022. doi: 10.1101/2022.06.09.22276206 [DOI]
- 10.Palmqvist S, Janelidze S, Quiroz YT, et al. Discriminative accuracy of plasma phospho-tau217 for Alzheimer disease vs other neurodegenerative disorders. JAMA. 2020;324(8):772-781. doi: 10.1001/jama.2020.12134 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Mattsson-Carlgren N, Janelidze S, Bateman RJ, et al. Soluble P-tau217 reflects amyloid and tau pathology and mediates the association of amyloid with tau. EMBO Mol Med. 2021;13(6):e14022. doi: 10.15252/emmm.202114022 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Janelidze S, Berron D, Smith R, et al. Associations of plasma phospho-tau217 levels with tau positron emission tomography in early Alzheimer disease. JAMA Neurol. 2021;78(2):149-156. doi: 10.1001/jamaneurol.2020.4201 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Karikari TK, Pascoal TA, Ashton NJ, et al. Blood phosphorylated tau 181 as a biomarker for Alzheimer’s disease: a diagnostic performance and prediction modelling study using data from four prospective cohorts. Lancet Neurol. 2020;19(5):422-433. doi: 10.1016/S1474-4422(20)30071-5 [DOI] [PubMed] [Google Scholar]
- 14.Ashton NJ, Pascoal TA, Karikari TK, et al. Plasma p-tau231: a new biomarker for incipient Alzheimer’s disease pathology. Acta Neuropathol. 2021;141(5):709-724. doi: 10.1007/s00401-021-02275-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Mattsson N, Andreasson U, Zetterberg H, Blennow K; Alzheimer’s Disease Neuroimaging Initiative . Association of plasma neurofilament light with neurodegeneration in patients with Alzheimer disease. JAMA Neurol. 2017;74(5):557-566. doi: 10.1001/jamaneurol.2016.6117 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Cicognola C, Janelidze S, Hertze J, et al. Plasma glial fibrillary acidic protein detects Alzheimer pathology and predicts future conversion to Alzheimer dementia in patients with mild cognitive impairment. Alzheimers Res Ther. 2021;13(1):68. doi: 10.1186/s13195-021-00804-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Snellman A, Lantero-Rodriguez J, Emeršič A, et al. N-terminal and mid-region tau fragments as fluid biomarkers in neurological diseases. Brain. 2022;145(8):2834-2848. doi: 10.1093/brain/awab481 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Lantero-Rodriguez J, Tissot C, Snellman A, et al. Plasma and CSF concentrations of N-terminal tau fragments associate with in vivo neurofibrillary tangle burden. Alzheimers Dement. Published online May 15, 2023. doi: 10.1002/alz.13119 [DOI] [PubMed] [Google Scholar]
- 19.Palmqvist S, Rossi M, Hall S, et al. Cognitive effects of Lewy body pathology in clinically unimpaired individuals. Nat Med. 2023;29(8):1971-1978. doi: 10.1038/s41591-023-02450-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Pereira JB, Ossenkoppele R, Palmqvist S, et al. Amyloid and tau accumulate across distinct spatial networks and are differentially associated with brain connectivity. Elife. 2019;8:e50830. doi: 10.7554/eLife.50830 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Leuzy A, Smith R, Ossenkoppele R, et al. Diagnostic performance of RO948 F 18 tau positron emission tomography in the differentiation of Alzheimer disease from other neurodegenerative disorders. JAMA Neurol. 2020;77(8):955-965. doi: 10.1001/jamaneurol.2020.0989 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Pontecorvo MJ, Devous MD, Kennedy I, et al. A multicentre longitudinal study of flortaucipir (18F) in normal ageing, mild cognitive impairment and Alzheimer’s disease dementia. Brain. 2019;142(6):1723-1735. doi: 10.1093/brain/awz090 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Smith R, Strandberg O, Leuzy A, et al. Sex differences in off-target binding using tau positron emission tomography. Neuroimage Clin. 2021;31:102708. doi: 10.1016/j.nicl.2021.102708 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Leuzy A, Mattsson-Carlgren N, Cullen NC, et al. Robustness of CSF Aβ42/40 and Aβ42/P-tau181 measured using fully automated immunoassays to detect AD-related outcomes. Alzheimers Dement. 2023;19(7):2994-3004. doi: 10.1002/alz.12897 [DOI] [PubMed] [Google Scholar]
- 25.Hansson O, Blennow K, Zetterberg H, Dage J. Blood biomarkers for Alzheimer’s disease in clinical practice and trials. Nat Aging. 2023;3(5):506-519. doi: 10.1038/s43587-023-00403-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Jack CR Jr, Bennett DA, Blennow K, et al. ; Contributors . NIA-AA research framework: toward a biological definition of Alzheimer’s disease. Alzheimers Dement. 2018;14(4):535-562. doi: 10.1016/j.jalz.2018.02.018 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Mattsson-Carlgren N, Grinberg LT, Boxer A, et al. Cerebrospinal fluid biomarkers in autopsy-confirmed Alzheimer disease and frontotemporal lobar degeneration. Neurology. 2022;98(11):e1137-e1150. doi: 10.1212/WNL.0000000000200040 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Budd Haeberlein S, Aisen PS, Barkhof F, et al. Two randomized phase 3 studies of aducanumab in early Alzheimer’s disease. J Prev Alzheimers Dis. 2022;9(2):197-210. doi: 10.14283/jpad.2022.30 [DOI] [PubMed] [Google Scholar]
- 29.Reish NJ, Jamshidi P, Stamm B, et al. Multiple cerebral hemorrhages in a patient receiving lecanemab and treated with t-PA for stroke. N Engl J Med. 2023;388(5):478-479. doi: 10.1056/NEJMc2215148 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Solopova E, Romero-Fernandez W, Harmsen H, et al. Fatal iatrogenic cerebral amyloid-related encephalitis in a patient treated with lecanemab for Alzheimer’s disease: neuroimaging and neuropathology. medRxiv. Published online April 29, 2023. doi: 10.1101/2023.04.26.23289061 [DOI]
- 31.Ashton NJ, Janelidze S, Mattsson-Carlgren N, et al. Differential roles of Aβ42/40, p-tau231 and p-tau217 for Alzheimer’s trial selection and disease monitoring. Nat Med. 2022;28(12):2555-2562. doi: 10.1038/s41591-022-02074-w [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Salvadó G, Ossenkoppele R, Ashton NJ, et al. Specific associations between plasma biomarkers and postmortem amyloid plaque and tau tangle loads. EMBO Mol Med. 2023;15(5):e17123. doi: 10.15252/emmm.202217123 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Horie K, Salvadó G, Barthélemy NR, et al. CSF MTBR-tau243 is a specific biomarker of tau tangle pathology in Alzheimer’s disease. Nat Med. 2023;29(8):1954-1963. doi: 10.1038/s41591-023-02443-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Del Campo M, Peeters CFW, Johnson ECB, et al. CSF proteome profiling across the Alzheimer’s disease spectrum reflects the multifactorial nature of the disease and identifies specific biomarker panels. Nat Aging. 2022;2(11):1040-1053. doi: 10.1038/s43587-022-00300-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Pichet Binette A, Janelidze S, Cullen N, et al. Confounding factors of Alzheimer’s disease plasma biomarkers and their impact on clinical performance. Alzheimers Dement. 2023;19(4):1403-1414. doi: 10.1002/alz.12787 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Janelidze S, Bali D, Ashton NJ, et al. Head-to-head comparison of 10 plasma phospho-tau assays in prodromal Alzheimer’s disease. Brain. 2023;146(4):1592-1601. doi: 10.1093/brain/awac333 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eMethods
eResults
eReferences
eTable 1. Demographics by training and test sets
eTable 2. Results for models to predict Aβ-positivity
eTable 3. Results for models to predict continuous Tau PET SUVR
eTable 4. Results for models to predict high Tau PET
eTable 5. Hypothetical cost savings by novel work-flow
eTable 6. Hypothetical cost savings by novel work-flow
eFigure 1. Gaussian mixture model for MUBADA ROI tau-PET
eFigure 2. Average [18F]RO948 tau-PET uptake across Aβ and Tau PET classes
eFigure 3. Gray-zones for plasma p-tau217 to predict Aβ-PET positivity
eFigure 4. Predict continuous tau-PET
eFigure 5. Aβ-status is predicted with plasma biomarkers, without SCD
eFigure 6. Predict high Tau-PET among predicted Aβ-positive participants, without SCD
eFigure 7. Performance of plasma P-tau217 in a validation cohort
Data sharing statement