Optimizing cutpoints for clinical interpretation of brain amyloid status using plasma p‐tau217 immunoassays

Abstract

INTRODUCTION

We aimed to evaluate clinical interpretation cutpoints for two plasma phosphorylated tau (p‐tau)217 assays (ALZpath and Lumipulse) as predictors of amyloid status for implementation in clinical practice.

METHODS

Clinical performance of plasma p‐tau217 against amyloid positron emission tomography status was evaluated in participants with mild cognitive impairment or mild dementia (n = 427).

RESULTS

Using a one‐cutpoint approach (negative/positive), neither assay achieved ≥ 90% in both sensitivity and specificity. A two‐cutpoint approach yielding 92% sensitivity and 96% specificity provided the desired balance of false positives and false negatives, while categorizing 20% and 39% of results as indeterminate for the Lumipulse and ALZpath assays, respectively.

DISCUSSION

This study provides a systematic framework for selection of assay‐specific cutpoints for clinical use of plasma p‐tau217 for determination of amyloid status. Our findings suggest that a two‐cutpoint approach may have advantages in optimizing diagnostic accuracy while minimizing potential harm from false positive results.

Highlights

• Phosphorylated tau (p‐tau)217 cutpoints for detection of amyloid pathology were established.

• A two‐cutpoint approach exhibited the best performance for clinical laboratory use.

• p‐tau217 assays differed in the percentage of results categorized as intermediate.

1. BACKGROUND

The introduction of Alzheimer's disease (AD) blood‐based biomarkers (BBMs) into clinical practice promises to provide non‐invasive and scalable tools for the detection of amyloid pathology. AD BBMs may be used to aid in the evaluation of individuals presenting with mild cognitive impairment or mild dementia in determining eligibility for disease‐modifying therapies in situations in which access to advanced testing, such as cerebrospinal fluid (CSF) biomarkers or positron emission tomography (PET), is limited. However, implementation of these assays as a routine clinical chemistry test requires careful evaluation of test performance and selection of cutpoints for interpretation, to avoid potential misclassification of patients that may lead to misdiagnosis and/or unnecessary treatments.

Proposed AD BBMs include plasma amyloid beta 42/40 (Aβ42/40) ratio and the phosphorylated tau (p‐tau) species—p‐tau181, p‐tau217, and p‐tau231. While useful in CSF, assays for Aβ42/40 have demonstrated limited diagnostic accuracy and poor robustness in plasma due to the small fold change between amyloid‐positive and amyloid‐negative individuals—only 10% to 15% in plasma compared to ≈ 50% in CSF. ¹ , ² Among the p‐tau species, p‐tau217 is the leading AD BBM demonstrating high diagnostic accuracy and disease specificity. Using memory clinic‐based cohorts, plasma p‐tau217 has been shown to be able to detect AD pathology with areas under the receiver operating characteristic (ROC) curve (AUC) > 0.90. ³ , ⁴ , ⁵ , ⁶ , ⁷ However, widespread replication of these findings has been limited, possibly caused by limited commercial availability of plasma p‐tau217 assays until recently.

The Alzheimer's Association's appropriate use recommendations for using BBMs in clinical practice and trials emphasized the need to compare the clinical performance of BBMs to the clinical performance of established CSF and/or PET biomarkers. The recommendations indicate that for BBMs to be standalone biomarkers for trial inclusion in studies evaluating disease‐modifying therapies, the positive predictive value of such tests should likely be at least 90% to 95%. ⁸ The revised Alzheimer's Association criteria for diagnosis and staging of AD acknowledge that plasma biomarkers are being integrated with traditional CSF and PET biomarkers; however, they also caution that diagnostic accuracy varies substantially among the various available BBMs. ⁹ Furthermore, the revised criteria introduce guidance on accuracy requirements for the use of BBMs in the diagnosis of AD. ⁹ Per their definition, plasma BBM assays that may suffice as standalone diagnostic tests for AD are those with a minimum accuracy of 90% to detect abnormal amyloid in the intended use population, which will be approximately equivalent to the diagnostic performance of US Food and Drug Administration–approved AD CSF biomarker assays.

Most recently, the Global CEO Initiative on Alzheimer's Disease published a consensus statement for the acceptance performance of BBM assays of amyloid pathology. ¹⁰ Recommendations include that a BBM used as a triaging test (i.e., individuals with a positive result are expected to undergo a second confirmatory test such as amyloid PET or CSF AD biomarker testing), should have a sensitivity ≥ 90% for prediction of amyloid PET status and a specificity ≥75% to 85%, depending on the availability of follow‐up testing. For BBM assays that are used as a confirmatory test, that is, without a follow‐up amyloid PET or CSF AD biomarker test, the performance of the assay should be equivalent to that of CSF tests, with ≈ 90% sensitivity and specificity for prediction of amyloid PET status. As more AD BBM assays become available in clinical laboratories, transparency about clinical performance in the intended use population is important to allow clinicians to make informed decisions using the most appropriate test.

In this study, we aimed to define clinical interpretation cutpoints for various plasma assays as predictors of amyloid status taking into consideration the accuracy recommendations from the Revised Criteria for Diagnosis and Staging of Alzheimer's Disease. ⁹ Cutpoints for predicting amyloid pathology based on an abnormal amyloid PET were defined in a cohort of participants from the Mayo Clinic Study of Aging and Mayo Clinic Alzheimer's Disease Research Center. Four assays were evaluated, including the ALZpath Simoa p‐tau217 assay, and the Fujirebio Lumipulse p‐tau217, Aβ42/Aβ40, and p‐tau217/Aβ42 ratio assays.

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed literature in PubMed related to plasma phosphorylated tau (p‐tau)217 immunoassays for predicting brain amyloid pathology related to Alzheimer's disease (AD). While a limited number of publications were identified, none so far had compared the diagnostic performance of a two‐cutpoint approach between leading plasma p‐tau217 immunoassays likely to be used clinically.

2. Interpretation: Cutoffs were identified using a two‐cutpoint approach that resulted in the highest possible specificity, while maintaining at least 90% sensitivity and keeping a minimal number (≤ 20%) of results in the intermediate range. Differences in the percentages of results classified as intermediate were seen between the assays. The number of results in the intermediate range were also dependent on prevalence when using predefined cutpoints to achieve 92% sensitivity at the lower cutpoint and 96% specificity at the upper cutpoint.

3. Future directions: The information provided here could serve as the basis for interpretation of these AD biomarkers in clinical practice.

2. METHODS

2.1. Participants

Participants (N = 427) were selected through the Mayo Clinic Study of Aging (MCSA, n = 312) and the Mayo Clinic Alzheimer's Disease Research Center (ADRC, n = 115) in Rochester, Minnesota. The MCSA is a population‐based study in Olmsted County, Minnesota examining long‐term cognitive aging in adults to study prevalence, incidence, and risk factors for mild cognitive impairment (MCI) and dementia with a focus on biomarkers for dementia. ¹¹ The Mayo Clinic ADRC is a longitudinal study of participants referred to the Mayo Clinic. Final clinical diagnoses of the participants were established by consensus using previously published criteria. ¹² , ¹³ The MCSA and Mayo Clinic ADRC protocols have been approved by the institutional review boards of Mayo Clinic and Olmsted Medical Center. Written informed consent was obtained from all who participated.

The plan for analysis was created after the participants were recruited for inclusion in the MCSA or the Mayo Clinic ADRC and all had received a clinical diagnosis and amyloid PET imaging. The plan included testing by the ALZpath and Lumipulse assays and comparing diagnostic accuracy data between available methods. Participants included in this study were taken from participants with PET data and plasma collections between December 2005 and March 2023, spanning a range of amyloid PET Centiloid values. Participants were required to be ≥ 50 years, have an available plasma sample drawn within 6 months of amyloid PET imaging (median [interquartile range (IQR)] = 1.7 [0.4, 2.6] months), and have a clinical diagnosis of MCI or mild dementia that could be consistent with AD.

2.2. Plasma samples

Ethylenediaminetetraacetic acid (EDTA) plasma samples were collected from participants after an overnight fast. Plasma was separated within 2 hours of collection. EDTA plasma was aliquoted into 1.5 mL polypropylene tubes, which were stored frozen at −80°C until testing. Samples were thawed to room temperature, mixed by vortexing, and centrifuged at 2000 × g for 5 minutes directly prior to analysis of all assays to maintain consistent preanalytic processing with other tests in the clinical laboratory. Samples had undergone two or fewer freeze/thaw cycles prior to testing. Up to three freeze/thaw cycles were shown not to affect analyte concentration based on internal validation studies. Samples that were not able to be analyzed by all assays due to low sample volume were excluded from all analysis. During testing, the laboratory was blinded to the clinical diagnosis and amyloid PET data.

2.3. ALZpath p‐tau217 assay

The ALZpath p‐tau217 assay is a plasma‐based Simoa immunoassay performed on the Quanterix Simoa HD‐X automated immunoassay analyzer. ³ EDTA plasma samples were analyzed in duplicate directly from the aliquot tubes. Testing was performed using the ALZpath Simoa^® p‐Tau 217 v2 Assay Kit (catalog number: 104371, lot number: 999008) between January 12, 2024 and January 25, 2024, and results are expressed in pg/mL. The analytical performance characteristics of this assay have been previously reported. ³ Manufacturer quality control (QC) material was analyzed before and after each testing run and determined to be within the manufacturer's specifications. Precision based on the QC data collected was < 14% coefficient of variation (CV).

2.4. Fujirebio Lumipulse p‐tau217, Aβ42, and Aβ40 assays

Testing was performed on the Fujirebio Lumipulse G1200 automated immunoassay analyzer using the Lumipulse^® G β‐Amyloid 1‐42 Plasma (catalog number: 81301, lot number: 3105), Lumipulse^® G β‐Amyloid 1‐40 Plasma (catalog number: 81298, lot number: 3105), and Lumipulse^® G p‐Tau 217 Plasma (catalog number: 81472, lot number: 4049) kits. EDTA plasma samples were analyzed directly from the aliquot tubes; results are expressed in pg/mL. QC material was analyzed before and after testing each day and determined to be within the manufacturer's specifications. Precision based on QC data was < 5% CV for all Lumipulse assays tested.

2.5. ¹¹C Pittsburgh compound B PET imaging

Amyloid PET imaging was chosen as the reference method due to availability of data in the cohort and its high diagnostic accuracy for detecting AD neuropathology. ¹⁴ Amyloid PET imaging was performed with ¹¹C Pittsburgh compound B (PiB), ¹⁵ and PET images were analyzed using an in‐house automated image processing pipeline as described previously. ¹⁶ An amyloid PET standardized uptake value ratio (SUVR) was calculated as the voxel‐number weighted average of the median uptake across the following target regions: prefrontal, orbitofrontal, parietal, temporal, anterior cingulate, and posterior cingulate/precuneus divided by the median uptake in a cerebellar crus gray matter reference region, ¹⁷ and SUVR was converted to Centiloid units as previously described. ¹⁸ Amyloid PET was considered positive (abnormal) if the Centiloid was ≥ 25 (SUVR ≥ 1.52) to identify an intermediate‐to‐high degree of AD neuropathological changes. ¹⁷ , ¹⁹ , ²⁰ , ²¹

2.6. Data analysis

Spearman rank correlation coefficient (rho) was calculated to estimate the correlation between each marker with amyloid PET Centiloid. ROC AUC classification performance for identifying abnormal amyloid PET (Centiloid ≥ 25) was calculated for each BBM. Differences in AUC for the individual BBMs were tested using the DeLong method. ²² Binary cutpoints for predicting amyloid PET status were derived based on the Youden index (jointly maximizing sensitivity and specificity). ²³ To optimize overall test accuracy and decrease the number of false positive results, we also developed a two‐cutpoint (three‐category) approach as previously recommended. ²⁴ , ²⁵ , ²⁶ Selected combinations of sensitivity and specificity were examined to define the lower and upper cutpoints, respectively. Individuals with assay levels between the lower and upper cutpoints were classified as intermediate risk and would be clinically referred to follow‐up testing using amyloid PET imaging. Our aim was to identify thresholds that resulted in the highest possible specificity, while maintaining at least 90% sensitivity and keeping a minimal number (≤ 20%) of results in the intermediate range.

Histogram density plots with kernel smoothers were created for each test to visualize differences in concentration distributions between positive and negative PET results. Assays were expressed as log(base 2) and then z score transformed for visualizing comparisons across assays on the same scale. Additionally, positive predictive value (PPV), negative predictive value (NPV), and accuracy at cutpoints with defined sensitivity and specificity were calculated using standard formulas for dependencies on prevalence and expressed as percentages. We note that when using lower and upper cutpoints defined by a given sensitivity and specificity, PPV, NPV, and overall test accuracy are each a function of prevalence and do not depend on the biomarker values. These measures were plotted along with the calculated percent of results expected to fall in the intermediate range, on a separate scale, at the defined cutpoints as a function of prevalence of amyloid PET positivity in the population. Estimates, 95% confidence intervals (95% CI) and figures were produced using Analyse‐it version 6.15 for Microsoft Excel, Stata MP v.18, and the R language and environment for statistical computing version 4.3.3 (R Foundation for Statistical Computing).

3. RESULTS

3.1. Participant characteristics

Table 1 summarizes the clinical characteristics and biomarker distribution of the participants. Of the 427 participants, 345 had MCI, and 82 had mild dementia based on consensus clinical diagnosis. Most of the participants in each group were male (61% among MCI, and 56% among those with mild dementia), with ages ranging from 52 to 93 years. In the MCI group, 198 (57%) were classified as having a positive amyloid PET, whereas 76 (93%) in the mild dementia group had a positive amyloid PET, with the overall prevalence of amyloid positivity being 64%.

TABLE 1.

Participant demographic characteristics by clinical diagnosis.

Characteristic	All (N = 427)	MCI (n = 345)	Dementia (n = 82)
Median age, years (range)	78 (52, 93)	79 (52, 93)	76 (53, 92)
Sex
Female	169 (40%)	133 (39%)	36 (44%)
Male	258 (60%)	212 (61%)	46 (56%)
Racial or ethnic category
Asian	2 (< 1%)	2 (< 1%)	0 (0%)
Black	4 (< 1%)	2 (< 1%)	2 (2%)
Hispanic or Latino	1 (< 1%)	1 (< 1%)	0 (0%)
White	415 (97%)	336 (97%)	79 (96%)
More than one race	3 (< 1%)	2 (< 1%)	1 (1%)
Decline to state	2 (< 1%)	2 (< 1%)	0 (0%)
Education, years	14 (12, 16)	14 (12, 16)	14 (12, 16)
Cohort study
Mayo Clinic Study of Aging	312 (73%)	293 (85%)	19 (23%)
Mayo Clinic Alzheimer's Disease Research Center	115 (27%)	52 (15%)	63 (77%)
APOE ε4 genotype
Non‐carrier	231 (56%)	202 (60%)	29 (36%)
Carrier	185 (44%)	133 (40%)	52 (64%)
Amyloid PET, SUVR	1.93 (1.40, 2.44)	1.66 (1.38, 2.31)	2.47 (2.12, 2.73)
Amyloid Centiloid	62 (15, 107)	38 (13, 96)	108 (79, 132)
Amyloid PET Centiloid ≥ 25 (SUVR ≥ 1.52)
Negative (normal)	153 (36%)	147 (43%)	6 (7%)
Positive (abnormal)	274 (64%)	198 (57%)	76 (93%)
ALZpath p‐tau217, pg/mL	0.590 (0.309, 1.00)	0.492 (0.281, 0.901)	0.986 (0.738, 1.40)
Lumipulse p‐tau217, pg/mL	0.283 (0.148, 0.597)	0.236 (0.130, 0.466)	0.709 (0.471, 1.31)
Lumipulse Aβ42/40 ratio	0.090 (0.084, 0.099)	0.092
(0.085, 0.101)	0.086 (0.080, 0.091)
Lumipulse p‐tau217/Aβ42 ratio	0.012 (0.005, 0.025)	0.010 (0.005, 0.019)	0.032 (0.022, 0.060)
CDR Global Score
0	108 (25%)	108 (31%)	0
0.5	264 (62%)	230 (67%)	34 (41%)
1	55 (13%)	7 (2%)	48 (59%)
CDR Sum of Boxes	1 (0, 2)	0 (0, 2)	4 (4, 6)
Mini‐Mental State Examination	26 (23, 27)	26 (24, 27)	22 (19, 23)

Note: Values shown are median (lower quartile, upper quartile) or number and percentage.

Abbreviations: Aβ, beta amyloid; APOE, apolipoprotein E; CDR, Clinical Dementia Rating; MCI, mild cognitive impairment; PET, positron emission tomography; p‐tau, phosphorylated tau; SUVR, standardized uptake value ratio.

3.2. BBM levels based on amyloid status

The concentrations of each of the four biomarkers stratified by amyloid positive (A+) and amyloid negative (A−) groups are shown in Figure S1 in supporting information. Fold differences between A+ and A− means were: 2.9, 3.8, 4.6, and 1.2 for ALZpath p‐tau217, Lumipulse p‐tau217, Lumipulse p‐tau217/Aβ42, and Lumipulse Aβ42/Aβ40, respectively. Corresponding to the smaller fold differences, ALZpath p‐tau217 and Lumipulse Aβ42/Aβ40 showed greater overlap in assay distributions between A+ and A− groups than Lumipulse p‐tau217 and Lumipulse p‐tau217/Aβ42 (Figure S1).

3.3. Diagnostic performance of BBMs alone for amyloid status prediction

Figure 1 ROC curves (and Figure S2 in supporting information) show high accuracy for classifying amyloid PET status (A+ vs. A−) from ALZpath p‐tau217 AUC (95% CI) of 0.91 (0.88, 0.93), Lumipulse p‐tau217 of 0.93 (0.90, 0.96), and Lumipulse p‐tau217/Aβ42 of 0.94 (0.92, 0.96), while less so for Lumipulse Aβ42/Aβ40 of 0.83 (0.78, 0.87). Differences in AUC were not statistically significant between the Lumipulse p‐tau217 and p‐tau217/Aβ42 assays (p = 0.096). Differences between Lumipulse p‐tau217 and ALZpath p‐tau217 were statistically significant (p = 0.025). The Lumipulse Aβ42/Aβ40 ratio displayed significantly lower AUC than all other assays evaluated (p < 0.001 for all). Single cutpoints for detection of amyloid positivity based on Youden index are also shown in Figure 1. Even with high AUCs, none of the tests achieved ≥ 90% in both sensitivity and specificity using a single cutpoint approach.

FIGURE 1.

ROC curves with Youden index cutpoints for prediction of amyloid PET positivity. Youden index (J), used to identify optimal binary cutpoints jointly maximizing sensitivity and specificity is provided in addition to AUCs calculated from the ROC curve for all assays. Aβ, amyloid beta; AUC, area under the receiver operating characteristic curve; CI, confidence interval; PET, positron emission tomography; p‐tau, phosphorylated tau; ROC, receiver operating characteristic

3.4. Clinical interpretation cutpoints for identification of an abnormal amyloid PET

To optimize the overall test accuracy and decrease the number of false positive results compared to a single cutpoint, we used a two‐cutpoint approach as previously proposed. ²⁴ , ²⁵ , ²⁶ For this, we focused on the two p‐tau217 assays, as the Lumipulse Aβ42/Aβ40 ratio showed significantly lower diagnostic accuracy than either p‐tau217 assay, and the Lumipulse p‐tau217/Aβ42 ratio diagnostic accuracy was not significantly better than the more straightforward Lumipulse p‐tau217 alone (p = 0.096). To meet the simultaneous goals of sensitivity and specificity > 90% and an intermediate results percentage ≤ 20%, we examined all cutpoint pairings [sensitivity/specificity] for classification that achieved these goals, with [90/90], [95/95], and [92/96] being shown in detail for the two p‐tau217 assays (Table 2). Results in the intermediate range require additional testing (CSF AD biomarkers or amyloid PET) to confirm the presence of amyloid pathology; classification results presented throughout assumed follow‐up amyloid PET testing had 100% sensitivity and specificity for BBM results falling in the intermediate range.

TABLE 2.

p‐tau217 assays performance for prediction of amyloid PET positivity at defined sensitivity and specificity cutpoints.

Assay	Sensitivity ≥ 92%, specificity ≥ 96%			Sensitivity ≥ 95%, specificity ≥ 95%			Sensitivity ≥ 90%, specificity ≥ 90%
ALZPath	A−	A+	All	A−	A+	All	A−	A+	All
Negative, # (%)	109 (71%)	22 (8%)	131 (31%)	90 (59%)	13 (5%)	103 (24%)	113 (74%)	27 (10%)	140 (33%)
Intermediate, # (%)	38 (25%)	129 (47%)	167 (39%)	56 (37%)	118 (43%)	174 (41%)	25 (16%)	50 (18%)	75 (18%)
Positive, # (%)	6 (4%)	123 (45%)	129 (30%)	7 (5%)	143 (52%)	150 (35%)	15 (10%)	197 (72%)	212 (50%)
Cutpoints (pg/mL)	Lower: ≤ 0.372, upper: > 0.920			Lower: ≤ 0.304, upper: > 0.811			Lower: ≤ 0.405, upper: > 0.590

Lumipulse	A−	A+	All	A−	A+	All	A−	A+	All
Negative, # (%)	117 (76%)	22 (8%)	139 (33%)	111 (73%)	13 (5%)	124 (29%)	120 (78%)	28 (10%)	148 (35%)
Intermediate, # (%)	30 (20%)	56 (20%)	86 (20%)	35 (23%)	57 (21%)	92 (22%)	18 (12%)	21 (8%)	39 (9%)
Positive, # (%)	6 (4%)	196 (72%)	202 (47%)	7 (5%)	204 (74%)	211 (49%)	15 (10%)	225 (82%)	240 (56%)
Cutpoints (pg/mL)	Lower: ≤ 0.185, upper: > 0.324			Lower: ≤ 0.166, upper: > 0.285			Lower: ≤ 0.194, upper: > 0.248

Note: A–, amyloid PET negative (n = 153); A+, amyloid PET positive (n = 274); All: sum of A– and A+ (N = 427).

Abbreviations: PET, positron emission tomography; p‐tau, phosphorylated tau.

For all evaluated sensitivity/specificity pairings, the ALZpath p‐tau217 assay had a consistently higher percentage of results in the intermediate range than the Lumipulse p‐tau217 assay (Table 2), with the 90% sensitivity and specificity combination being the only combination classifying ≤ 20% of results in the intermediate range. For the Lumipulse p‐tau217 assay, the combination of 92% sensitivity and 96% specificity was chosen to define lower and upper cutpoints, which provided the highest possible specificity, while maintaining > 90% sensitivity and keeping a minimal number (≤ 20%) of results in the intermediate range. Using this combination, the number of samples in the intermediate range was 20% with 60% of the samples in the intermediate range being amyloid PET positive. Figure 2 shows Lumipulse and ALZpath p‐tau217 distributions by amyloid status with positive, negative, and intermediate “gray zones” marked by assay cutpoints corresponding to the 92% sensitivity/96% specificity thresholds. Figure 2A,B show logistic probability curves for A+ status with assay values being log(base 2) and z score transformed to visualize distributions on a comparable scale, while Figure 2C,D show continuous amyloid PET Centiloid values with raw assay concentration labels and cutpoints. Notably, there is a wide range of Centiloid values in the results classified as intermediate, but the majority of samples have an amyloid PET Centiloid ≥ 25, especially for ALZpath p‐tau217 (Figure 2).

FIGURE 2.

Biomarker categorizations with 92% sensitivity and 96% specificity two‐cutpoint thresholds for the Lumipulse (A,C) and ALZpath (B,D) p‐tau217 assays (Test₁). Lower and upper cutpoints are represented with green and red dashed lines, respectively. Gray shaded areas represent the intermediate zone in which samples would require follow‐up testing. Numbers shown on the plot represent the number of samples falling into each classification category and for the bottom plot are separated by amyloid PET positivity. PET, positron emission tomography; p‐tau, phosphorylated tau

3.5. Overall test performance

While sensitivity and specificity can be determined independent of disease prevalence, important clinical considerations of PPVs and NPVs are known to depend on prevalence. Similarly, and importantly for two‐cutpoint (three‐category) tests, the proportion of intermediates is also dependent on disease prevalence. ²⁷

Figure 3 (and Figure S3 in supporting information) displays the overall PPV, NPV, and accuracy for selected sensitivity and specificity combinations as a function of the underlying population amyloid positivity prevalence, along with the intermediate test percentages for the Lumipulse p‐tau217 and ALZpath assays. Table S1 in supporting information shows PPV, NPV, accuracy, and intermediate frequency values at specific amyloid positive prevalences outlined in Figure 3. As noted above, given the use of predefined sensitivity/specificity cutpoints and amyloid PET follow‐up testing for intermediate results, both assays will have the same PPV, NPV, and accuracy. For example, using the 92% sensitivity and 96% specificity cutpoints (panel A) and the study amyloid positive prevalence rate of 64%, the overall test accuracy is 93% for both p‐tau217 assays.

FIGURE 3.

Effect of amyloid positivity prevalence on the number of results in the intermediate range for the Lumipulse (A) and ALZpath (B) p‐tau217 assays. Calculations consider follow‐up amyloid PET testing of intermediates to be 100% sensitive and specific. Dotted lines plotted at 30, 50, and 65% amyloid PET positivity prevalence represent approximations of expected prevalence in populations of individuals with subjective cognitive impairment, mild cognitive impairment, and dementia, respectively. Positive predictive value (PPV), negative predictive value (NPV), and accuracy are dependent on the amyloid positive prevalence. PET, positron emission tomography; p‐tau, phosphorylated tau

The most meaningful difference between the assays at fixed sensitivity/specificity cutpoints then is the overall number of samples classified into the intermediate range. With the ALZpath assay, the number of patients needing follow‐up would be reduced by 61% (39% of results in the intermediate range [Table 2]) while for the Lumipulse p‐tau217, the number of patients needing follow‐up would be reduced by 80% (20% intermediate results [Table 2]). Figure 3 shows that, while the intermediate result percentage remains approximately constant for the Lumipulse p‐tau217 assay across amyloid pathology prevalences, the intermediate percentage is both consistently higher and increases dramatically with prevalence for the ALZpath p‐tau217 assay.

4. DISCUSSION

In this study, we evaluated the performance of various plasma assays as predictors of amyloid pathology and established cutpoints that could be used in the clinical interpretation of these assays. These cutpoints were optimized for the detection of amyloid pathology in patients with MCI and mild dementia, which represents the population in which these results will provide actionable information about eligibility for current therapies targeting amyloid beta (Aβ) pathology. Given that these assays may be used to determine eligibility for Aβ‐targeting therapies, maximizing performance toward the clinical specificity of the assays was preferred to minimize the number of false positive results, which could otherwise lead to complications from unnecessary treatment use.

While most quantitative tests in laboratory medicine use a binary classification for test interpretation, none of the assays evaluated here would provide a sensitivity and specificity of ≥ 90% to parallel the performance of current AD CSF biomarkers when using a single cutpoint. This is not an unexpected finding as the current AD BBMs, including p‐tau217, display a higher overlap between amyloid positive and negative groups than AD CSF biomarkers. ² , ⁶ Our findings support prior observations, but also demonstrate that the degree of overlap is assay dependent, as we observed a higher overlap of p‐tau217 concentrations between groups in the ALZpath assay versus the Lumipulse assay. While the reason for these differences was not investigated, this might be related to differences in assay precision as reported for other AD BBMs. ²

A two‐cutoff approach has been suggested by others as a way of optimizing the overall accuracy for prediction of amyloid PET status. ³ , ²⁴ , ²⁵ , ²⁶ , ²⁸ However, to our knowledge this is the first report comparing the performance of two leading plasma p‐tau217 immunoassays that are likely to be used clinically using this two‐cutpoint approach. The two‐cutpoint approach provides more stringent cutpoints to obtain higher accuracy of the positive and negative results, while classifying some results as intermediate, in which patients will require additional testing for assessment of amyloid status. ⁸

Based on our findings, the use of the two‐cutpoint approach would have advantages for clinical implementation of these p‐tau217 assays, as it reduces the number of false positive and false negative results that might lead to patient misclassification with its potential for neurologic and psychosocial harms. In this study, the two‐cutpoint approach offered a minimal false positive frequency (4%) and provided a definite result (similar to a CSF test) in ≈ 80% of the cases. In contrast, the one‐cutpoint approach could not achieve at least 90% in both sensitivity and specificity for any of the examined assays. For example, using the optimal (Youden) cutpoint for the Lumipulse p‐tau217, 14% of results would be false negatives and 12% of results would be false positives. Alternatively, if a single cutpoint with high sensitivity (95%) is selected, the specificity in this dataset will be 73% resulting in 27% false positive results. In this scenario (high specificity and low sensitivity), even if the test was considered a rule‐out test for amyloid pathology, a higher number of patients will need follow‐up by amyloid PET or CSF AD biomarkers to confirm or exclude the presence of amyloid pathology. Furthermore, a two‐cutpoint approach that minimizes false positives (higher specificity at upper cutpoint) may have unique strengths in relation to the impact of an AD diagnosis as well as the resource‐intensive nature of current anti‐amyloid therapies. This approach also preserves options for further assessment, for example, when a high suspicion of AD with a negative p‐tau217 result might warrant follow‐up such as repeat clinical assessment and/or repeat blood draw, AD CSF biomarker testing, or amyloid PET.

This study was designed with a specific intended use, which was detection of amyloid pathology in individuals presenting with symptoms of cognitive decline. More specifically, we focused on individuals with MCI or mild dementia, which represents the population in which these results will provide actionable information about eligibility for current therapies targeting Aβ pathology. In a population with MCI, the expected prevalence of amyloid pathology will be ≈ 50%. ²⁹ In our cohort the amyloid pathology prevalence was 64%, which is higher than expected in an MCI cohort. The test accuracy, PPV, and NPV would be influenced by the disease prevalence (Table S1). Therefore, for an assay with 92% sensitivity and 96% specificity, the test accuracy, PPV, and NPV would be 93%, 98%, and 87%, respectively, in a population with a 64% disease prevalence. The same assay would have an accuracy, PPV, and NPV of 94%, 96%, and 92% in a population with a 50% disease prevalence.

The key differentiator between the two p‐tau217 assays was the number of results that were classified in the intermediate range. It has been proposed that the percentage of samples in the intermediate category should ideally be ≤ 20% to minimize the number of individuals that must undergo additional testing. ²⁴ In our study, however, the percentage of samples falling in the intermediate category varied between p‐tau217 assays at cutpoints with equivalent predetermined sensitivity and specificity. While the Lumipulse assay classified 9% to 22% of results as intermediate based on the various cutpoints examined at 64% prevalence of amyloid pathology in this study, the ALZpath assay classified 18% to 39% of samples as intermediate. When selecting a p‐tau217 assay for clinical implementation, it would be advisable to consider the percent of samples categorized as intermediate as part of the overall performance, as these would need additional follow‐up for assessment of amyloid pathology. Selection of an assay with a lower number of results in the intermediate range will have a number of advantages including faster and less expensive diagnosis in a larger number of patients, reduction of potential bottleneck for lumbar puncture procedures or amyloid PET imaging, and overall health‐care cost savings driven by savings in the lower number of patients needing to receive AD CSF biomarker or amyloid PET testing for the diagnosis of amyloid pathology. ³⁰

The overall test performance of an assay using two cutpoints depends on the performance of the follow‐up method for classifying results in the intermediate range; therefore, performance calculations including sensitivity and specificity must take into account sensitivity and specificity of the follow‐up test used to assess the presence of amyloid pathology for the patients with intermediate results. For example, if the follow‐up test is considered 100% sensitive and specific, as is the case for amyloid PET, the overall test accuracy of the two‐cutpoint approach (92% sensitivity/96% specificity) would be 93% based on the 64% prevalence of amyloid pathology observed in this study. If instead a follow‐up CSF biomarker test with 90% sensitivity and specificity was used to classify intermediates, the overall test accuracy would drop to 91%.

Multiple comorbidities have been reported to be associated with elevated plasma p‐tau217 concentrations, with chronic kidney disease (CKD) having one of the largest effects. ³¹ , ³² In our study, information about kidney function was not taken into consideration for patient selection. Therefore, we are not able to assess how CKD might affect the performance of the established cutpoints. However, a strength of this approach is that it is likely to be most reflective of real‐world scenarios, where it could be challenging from a practical standpoint to necessitate multiple cutpoints for interpreting plasma pTau217 assays based on medical comorbidities. In addition, other comorbidities reported to influence p‐tau217 concentrations such as myocardial infarction (MI), history of stroke, and increased body mass index (BMI) need to be taken into consideration when interpreting p‐tau217 concentrations in the context of predefined cutpoints. In the presence of comorbidities affecting p‐tau217 concentrations, the two‐cutpoint approach may reduce the likelihood of patient misclassification, as positive or negative individuals may be more likely to end up in the intermediate zone in which follow‐up testing could help reduce the risk of affected results.

One of the strengths of our study is that only individuals with MCI or mild dementia were included, as this would be the context of use population for evaluation of amyloid pathology for current treatment with amyloid‐modifying therapies. Furthermore, 73% of the samples were from the MCSA, which is a population‐based cohort that can be expected to be representative of the samples tested in a real‐world clinical setting. Unfortunately, due to the homogeneity of the sample regarding race (97% White participants), the performance and transferability of these cutpoints may not be generalizable to other racial and ethnic groups. Given that data on the effect of race and ethnicity in p‐tau217 concentrations has been conflicting, ³³ , ³⁴ future studies are needed to evaluate the performance of these cutpoints in more diverse clinical cohorts. The performance of these cutpoints in cognitively unimpaired individuals was not evaluated; however, given that no approved disease‐targeted treatment yet exists for this population, limiting the study to impaired individuals seems appropriate. Of note, performance of these cutpoints is optimized based on the collection protocol in this study; further studies are needed to determine whether other preanalytic factors would affect the use of these cutpoints. As with any test used clinically, the performance of these cutpoints will depend on lot‐to‐lot variation of the assay, which will require a protocol for testing this variation with acceptance criteria for assessment of new lots to maintain acceptable levels of variability. In our limited experience, we have seen ≤ 4% variation over three different lots of Lumipulse p‐tau217 reagent.

In conclusion, this study represents a systematic approach for selection of clinical interpretation cutpoints for two plasma p‐tau217 assays as predictors of amyloid status. In this study cohort consisting of individuals with cognitive impairment (MCI and mild dementia), the use of a two‐cutpoint approach (negative, positive, and intermediate groups) for amyloid pathology classification provided the best diagnostic accuracy and decreased patient misclassification. Head‐to‐head comparison of two p‐tau217 immunoassays showed that the percent of samples classified as intermediate significantly differs between assays at fixed sensitivity and specificity thresholds. The information provided here could be used as the basis for interpretation of these AD BBMs in clinical practice.

CONFLICT OF INTEREST STATEMENT

A. Algeciras‐Schimnich has participated on advisory boards for Roche Diagnostics, Fujirebio Diagnostics, and Siemens Healthineers. J. A. Bornhorst has participated on an advisory board for Sunbird Bio and received an honorarium from Roche Diagnostics. D. J. Figdore reports no disclosures. J. Graff‐Radford receives funding from the NIH. He serves on DSMB for STROKENet and serves as site investigator for Alzheimer's Clinical Trial Consortium studies co‐sponsored by Cognition Therapeutics and Eisai. M. Griswold reports no disclosures. C. R. Jack receives funding from the NIH and the Alexander Family Alzheimer's Disease Research Professorship of the Mayo Clinic. D. S. Knopman serves on a Data Safety Monitoring Board for the Dominantly Inherited Alzheimer Network Treatment Unit study. He was an investigator in Alzheimer's disease clinical trials sponsored by Biogen, Lilly Pharmaceuticals, and the University of Southern California, both of which have ended, and is currently an investigator in a trial in frontotemporal degeneration with Alector. He has served as a consultant for Roche, AriBio, Linus Health, Biovie, and Alzeca Biosciences but receives no personal compensation. He receives funding from the NIH. V. J. Lowe consults for Bayer Schering Pharma, Piramal Life Sciences, Eisai, Inc., AVID Radiopharmaceuticals, Eli Lilly and Company, PeerView Institute for Medical Education, and Merck Research, and receives research support from GE Healthcare, Siemens Molecular Imaging, AVID Radiopharmaceuticals, and the NIH (NIA, NCI). R. C. Petersen has consulted for Roche, Inc.; Genentech, Inc.; Eli Lilly, Inc.; Nestle, Inc.; and Eisai, Inc.; is on a DSMB for Genentech, Inc.; and receives royalties from Oxford University Press for Mild Cognitive Impairment and from UpToDate. His research funding is from NIH/NIA. V. K. Ramanan has received research funding from the NIH and the Mangurian Foundation for Lewy body disease research; has provided educational content for Medscape; has received speaker and conference session honoraria from the American Academy of Neurology Institute; is co‐PI for a clinical trial supported by the Alzheimer's Association; is site Co‐I for the Alzheimer's Clinical Trials Consortium; and is a site clinician for clinical trials supported by Eisai, the Alzheimer's Treatment and Research Institute at USC, and Transposon Therapeutics, Inc. P. Vemuri receives funding from the NIH. Author disclosures are available in the supporting information.

CONSENT STATEMENT

All human subjects provided informed consent.

Supporting information

Supporting Information

Supporting Information

Figdore DJ, Griswold M, Bornhorst JA, et al. Optimizing cutpoints for clinical interpretation of brain amyloid status using plasma p‐tau217 immunoassays. Alzheimer's Dement. 2024;20:6506–6516. 10.1002/alz.14140