Longitudinal plasma p-tau217 as a marker for tracking progression and predicting cognitive decline in Alzheimer’s disease Authors: Jehyun Ahn, Jihwan Yun, Michael W. Weiner, Daeun Shin, Heekyoung Kang, Sohyun Yim, Seonghyeon Kim, Heekyoung Park, Hyemin Jang, Min Young Chun, Eun Hye Lee, Hee Jin Kim, Duk L. Na, Jun Pyo Kim, Sang Won Seo, the ADNI study group Published: 2026 ABSTRACT BackgroundPlasma phosphorylated tau217 (p-tau217) is a promising biomarker for Alzheimer’s disease (AD) with strong cross-sectional concordance with beta-amyloid (Aβ) and tau pathology. However, its longitudinal behavior across disease stages and assay platforms, and its prognostic value for cognitive decline according to Aβ status, remain insufficiently characterized. This study aimed to evaluate stage- and assay-specific longitudinal changes in plasma p-tau217 and to determine their utility for predicting cognitive decline.MethodsWe analyzed longitudinal data from 393 participants in the NA-ADNI, including cognitively unimpaired (CU) and cognitively impaired (CI) individuals. Plasma p-tau217 was measured using five assay platforms. Linear mixed-effects models were used to assess longitudinal biomarker trajectories, with time as the primary predictor and random intercepts and slopes. Aβ status and time-by-Aβ interactions were included to examine differential trajectories. Annualized biomarker changes were derived and related to longitudinal cognitive outcomes using models of Clinical Dementia Rating-Sum of boxes (CDR-SB) scores. Marginal R2 values quantified the variance explained.ResultsPlasma p-tau217 levels increased significantly over time across all assay platforms in both CU and CI groups (all q < 0.05). The largest standardized time effects were observed for the ALZpath-Simoa assay (βstd = 0.112 in CU; βstd = 0.100 in CI). Aβ + individuals showed steeper longitudinal increases than Aβ- individuals, with interaction effects ranging from βint = 0.102 to 0.164 in CU and from βint = 0.070 to 0.113 in CI. Annualized changes in p-tau217 were associated with cognitive decline in a stage- and assay-dependent manner. In CU, change in %p-tau217 measured by C2N-MS showed the strongest association with CDR-SB progression (β = 0.075, q < 0.001). In CI, absolute p-tau217 change measured by C2N-MS showed the strongest association (β = 0.238, q < 0.001), particularly among Aβ + individuals. Adding annualized change of other plasma biomarkers provided modest and stage-dependent improvements, with GFAP yielding the greatest incremental value in CI.ConclusionsLongitudinal plasma p-tau217 is a robust marker of AD progression, but its optimal prognostic application depends on cognitive stage, Aβ status, and assay platform. These findings support stage-specific strategies for using serial plasma biomarkers in clinical monitoring and therapeutic trials. Introduction Alzheimer’s disease (AD) is a progressive neurodegenerative disorder characterized by the accumulation of beta-amyloid (Aβ) plaques and tau neurofibrillary tangles, which precede clinical symptoms by years or even decades [1, 2]. In vivo detection of these pathologies has been revolutionized by positron emission tomography (PET) imaging and cerebrospinal fluid (CSF) analyses [3–6]. More recently, plasma biomarkers have emerged as minimally invasive, scalable tools for detecting Aβ and tau pathology, showing high concordance with PET and CSF measures [7–9]. Among them, plasma phosphorylated tau217 (p-tau217) has shown particularly strong diagnostic performance across the AD continuum [10–13]. Beyond cross-sectional detection, longitudinal monitoring of disease progression has become increasingly important, especially in the context of disease-modifying therapies [14, 15]. Serial plasma biomarker measurements may capture dynamic pathophysiological changes and offer practical tools for tracking disease activity and treatment response [16]. However, factors influencing biomarker trajectories—including demographic and genetic variables—may differ by cognitive stage, and longitudinal patterns may vary according to baseline Aβ status, particularly in cognitively unimpaired (CU) individuals [17–19]. Equally important is whether longitudinal plasma biomarker changes predict subsequent cognitive decline. Biomarkers that track disease progression and correlate with clinical outcomes could serve as accessible prognostic tools and surrogate endpoints in clinical trials [20, 21]. Although p-tau217 shows promise in this regard [22], comparative data across assay platforms remain limited, and it is unclear whether the optimal assay differs by cognitive stage. Moreover, the incremental prognostic value of combining p-tau217 with other plasma biomarkers reflecting neurodegeneration or astroglial activation, such as neurofilament light chain (NfL) and glial fibrillary acidic protein (GFAP) [23, 24], has not been systematically evaluated across clinical stages. In this study, we investigated the longitudinal changes and prognostic utility of plasma p-tau217 in a well-characterized cohort spanning CU and cognitively impaired (CI) individuals. We specifically aimed to characterize stage- and Aβ-dependent trajectories of p-tau217, to determine whether the prognostic value of longitudinal p-tau217 change varies by assay platform and cognitive stage, and to evaluate the incremental benefit of combining p-tau217 with other plasma biomarkers. Methods Participants We analyzed data from the North America Alzheimer's Disease Neuroimaging Initiative (NA-ADNI), a multisite longitudinal biomarker study initiated in 2003 across North America. Full inclusion/exclusion criteria are described in detail at http://www.adni-info.org. Briefly, all participants were between 55 and 90 years of age, had completed at least 6 years of education, and were free of any other significant neurologic diseases. At baseline, participants were classified as cognitively unimpaired (CU), mild cognitive impairment (MCI), or dementia due to AD according to ADNI diagnostic criteria, which have been described previously [25, 26]. Cognitive impairment (CI) in the present study comprised participants with MCI or dementia [25]. The present analysis included 393 participants who met the following criteria: (1) availability of amyloid PET imaging using ¹⁸F-florbetapir; (2) at least one plasma sample collected within 6 months of the florbetapir PET scan; and (3) at least two longitudinal plasma biomarker measurements obtained over a minimum follow-up interval of one year. The ADNI study received approval from the institutional review boards of all participating institutions, and all participants provided written informed consent. Aβ PET acquisition and quantification Pre-processed Aβ (florbetapir) PET standardized uptake value ratio (SUVR) data were directly downloaded from the ADNI website. The detailed protocols regarding ADNI PET acquisition and image processing can be found in previously published studies [27]. Amyloid-positive individuals were defined as having a florbetapir cortical SUVR of 1.11 or higher, using the whole cerebellum as a reference region [28]. Plasma collection and processing Pre-processed biomarker data and plasma sample information for this study were obtained from the ADNI database (filename: All_Subjects_FNIHBC_BLOOD_BIOMARKER_TRAJECTORIES_10Oct2024). Plasma collection, processing, and storage were performed by the ADNI core according to standardized protocols, which have been described in detail previously [29, 30]. All laboratories conducting the assays were blinded to participant information. Overall, data from two Aβ42/Aβ40 assays, four p-tau217 assays, one %p-tau217 (p-tau217/np-tau217), and single assays for p-tau181, GFAP, and NfL were included. Plasma p-tau217 was quantified using multiple assay platforms. At the Quanterix Accelerator Laboratory, p-tau217 was measured using the ALZpath antibody on the Simoa® HD-X platform (ALZpath-Simoa). At C2N Diagnostics (St. Louis, MO, USA), plasma p-tau217 was quantified using a liquid chromatography–tandem mass spectrometry (LC–MS/MS) method (C2N-MS p-tau217). Using the same LC–MS/MS data, a ratio-based biomarker (%p-tau217) was derived as the ratio of phosphorylated to non-phosphorylated tau217 (C2N-MS %p-tau217). In addition, p-tau217 was measured using a chemiluminescent enzyme immunoassay on the Lumipulse® G1200 platform with a proprietary antibody (Fujirebio-Lumipulse). In parallel, Janssen measured p-tau217 using a Simoa®-based assay employing a Janssen-developed antibody (Janssen-Simoa). Intra-assay coefficients of variation were < 10% for most assays; however, for the Fujirebio-Lumipulse p-tau217 assay, coefficients of variation were 11% and 34% for the high and low quality-control samples, respectively. Additional methodological details are available in the ADNI study documentation (adni.loni.usc.edu). Longitudinal follow-up The mean follow-up duration for plasma biomarker measurements was 5.4 years (range, 1.8–10.9 years). All participants underwent at least two Clinical Dementia Rating (CDR) assessments, with a mean clinical follow-up of 5.5 years (range, 1.1–10.9 years). For baseline cognitive assessments, test results obtained closest to the date of plasma collection were used. The mean time interval between the baseline plasma collection and the corresponding cognitive assessment was 29.5 days (range, 0–239 days). Statistical analysis We used linear mixed-effects models (LMMs) with random intercepts and random slopes over time to assess longitudinal changes in plasma biomarkers, with a particular focus on p-tau217. All plasma biomarker values were Z-transformed using the mean and standard deviation of the baseline values within each subgroup (CU, CI, CU Aβ +, and CI Aβ +). We stratified our analyses by amyloid positivity rather than adjusting for continuous amyloid burden to avoid severe multicollinearity between Aβ PET SUVR and plasma p-tau217, thereby isolating the stage-dependent prognostic value of the plasma biomarkers. For longitudinal trajectory analyses, Z-transformed plasma biomarker values were treated as outcomes, with time included as the primary fixed-effect predictor. Marginal R2 values were computed for each model to squantify the proportion of variance explained by fixed effects. To examine whether biomarker trajectories differed by amyloid status, amyloid PET status and its interaction with time were added to the models (fixed effects: amyloid status, time, and amyloid status × time). To evaluate the contribution of covariates expected to influence biomarker trajectories, baseline age, sex, APOE ε4 carrier status, and years of education were sequentially added to the models, with marginal R2 calculated at each step to quantify the incremental contribution of each covariate. To address potential concerns regarding the order of covariate entry in the stepwise LMM analyses, we conducted an additional penalized regression analysis using a Least Absolute Shrinkage and Selection Operator (LASSO) approach. The subject-level slopes of ptau217 were used as the dependent variable in LASSO regression models including baseline age, sex, APOE ε4 carrier status, and education as predictors. The penalty parameter (λ) was selected using tenfold cross-validation. To assess robustness of variable selection, 200 bootstrap resampling iterations were performed at the subject level, and selection frequency was calculated. To assess the association between plasma biomarker changes and cognitive decline, additional LMMs were constructed using the CDR-SB as the outcome. Annualized changes in plasma biomarkers were estimated for each individual as the slope of biomarker change over time and subsequently Z-transformed. The primary predictor in these models was the interaction between time and annualized biomarker change (fixed effects: time, annualized biomarker change, and time × annualized biomarker change). Covariates (baseline age, sex, APOE ε4 carrier status, and education) were sequentially added to assess their contribution to variance explained. To determine whether other plasma biomarkers provided incremental prognostic value beyond annualized change in p-tau217, annualized changes in Aβ42/Aβ40, GFAP, and NfL were sequentially added to models that included annualized change in p-tau217 and demographic covariates. Baseline age was consistently included as a covariate in all models evaluating individual and combined plasma biomarkers, including GFAP, to account for age-related increases. Continuous variables are expressed as mean (standard deviation), and categorical variables as frequency (percentage). Statistical analyses were performed using R version 4.5.1 (R Foundation for Statistical Computing, Vienna, Austria). A false discovery rate (FDR) of 0.05 was applied to adjust for multiple comparisons where indicated, with FDR-adjusted p-values reported as q values. Unless otherwise specified, statistical significance was defined as q < 0.05. Results Baseline characteristics of the study population Baseline clinical characteristics are summarized in Table 1. Among 393 participants, 209 (53.2%) were CU and 184 (46.8%) were CI (173 MCI and 11 dementia). Aβ PET positivity was observed in 65 (31.1%) CU and 85 (46.2%) CI participants. The mean age was 74.1 ± 6.6 years in CU and 71.6 ± 7.8 years in CI. The proportion of females was 54.1% (n = 113) in CU and 44.0% (n = 81) in CI, while APOE ε4 carriers accounted for 29.7% (n = 62) and 38.0% (n = 70), respectively. Mean years of education were 16.5 ± 2.6 in CU and 16.1 ± 2.7 in CI. Mean MMSE scores were 29.1 ± 1.2 in CU and 27.8 ± 2.3 in CI, and mean CDR-SB scores were 0.07 ± 0.25 and 1.45 ± 1.13 at the baseline, respectively.Table 1Baseline characteristics of study participantsCU (N = 209)CI (N = 184)Age, years, mean (SD)74.1 (6.6)71.6 (7.8)Sex, female, n (%)113 (54.1)81 (44.0)APOE ε4 carriers, n (%)62 (29.7)70 (38.0)Education, years, mean (SD)16.5 (2.6)16.1 (2.7)Diagnosis CU/MCI/Dementia, n (%)209/0/0(100.0/0.0/0.0)0/173/11(0.0/94.0/6.0)MMSE, mean (SD)29.1 (1.2)27.8 (2.3)CDR-SB, mean (SD)0.07 (0.25)1.45 (1.13)Amyloid positivity†, n (%)65 (31.1)85 (46.2)Follow-up time, years, mean (SD)5.7 (1.8)5.2 (1.8)Baseline biomarker values, mean (SD) p-tau217 (Alzpath-Simoa), pg/mL0.37 (0.26)0.49 (0.37) p-tau217 (C2N-MS), pg/mL1.94 (2.04)2.65 (2.80) %p-tau217 (C2N-MS)3.56 (2.98)4.78 (4.22) p-tau217 (Fujirebio-Lumipulse), pg/mL0.16 (0.24)0.23 (0.26) p-tau217 (Janssen-Simoa), pg/mL0.05 (0.04)0.07 (0.06) Aβ42/Aβ40 ratio (Quanterix)0.06 (0.01)0.06 (0.01) p-tau181 (Quanterix), pg/mL19.87 (12.65)21.30 (12.08) GFAP (Quanterix), pg/mL150.75 (81.39)150.29 (92.26) NfL (Quanterix), pg/mL20.07 (9.43)20.47 (12.40)Race, n (%) Asian2 (1.0%)3 (1.6%) Black or African American9 (4.3%)5 (2.7%) White195 (93.3%)174 (94.6%) More than one race2 (1.0%)2 (1.1%) Unknown1 (0.4%)0 (0%)Ethnicity, n (%) Hispanic or Latino10 (4.8%)9 (4.9%) Not Hispanic or Latino197 (94.2%)175 (95.1%) Unknown2 (1.0%)0 (0%)Abbreviations: APOE Apolipoprotein E, CU Cognitively unimpaired, MCI Mild Cognitive Impairment; p-tau217, phosphorylated-tau217, MMSE Mini-Mental State Examination, CDR-SB Clinical Dementia Rating-Sum of Boxes, SD standard deviation†Positive amyloid PET scan is defined as 18F-florbetapir (FBP) PET SUVR ≥ 1.11 Table 1Baseline characteristics of study participantsCU (N = 209)CI (N = 184)Age, years, mean (SD)74.1 (6.6)71.6 (7.8)Sex, female, n (%)113 (54.1)81 (44.0)APOE ε4 carriers, n (%)62 (29.7)70 (38.0)Education, years, mean (SD)16.5 (2.6)16.1 (2.7)Diagnosis CU/MCI/Dementia, n (%)209/0/0(100.0/0.0/0.0)0/173/11(0.0/94.0/6.0)MMSE, mean (SD)29.1 (1.2)27.8 (2.3)CDR-SB, mean (SD)0.07 (0.25)1.45 (1.13)Amyloid positivity†, n (%)65 (31.1)85 (46.2)Follow-up time, years, mean (SD)5.7 (1.8)5.2 (1.8)Baseline biomarker values, mean (SD) p-tau217 (Alzpath-Simoa), pg/mL0.37 (0.26)0.49 (0.37) p-tau217 (C2N-MS), pg/mL1.94 (2.04)2.65 (2.80) %p-tau217 (C2N-MS)3.56 (2.98)4.78 (4.22) p-tau217 (Fujirebio-Lumipulse), pg/mL0.16 (0.24)0.23 (0.26) p-tau217 (Janssen-Simoa), pg/mL0.05 (0.04)0.07 (0.06) Aβ42/Aβ40 ratio (Quanterix)0.06 (0.01)0.06 (0.01) p-tau181 (Quanterix), pg/mL19.87 (12.65)21.30 (12.08) GFAP (Quanterix), pg/mL150.75 (81.39)150.29 (92.26) NfL (Quanterix), pg/mL20.07 (9.43)20.47 (12.40)Race, n (%) Asian2 (1.0%)3 (1.6%) Black or African American9 (4.3%)5 (2.7%) White195 (93.3%)174 (94.6%) More than one race2 (1.0%)2 (1.1%) Unknown1 (0.4%)0 (0%)Ethnicity, n (%) Hispanic or Latino10 (4.8%)9 (4.9%) Not Hispanic or Latino197 (94.2%)175 (95.1%) Unknown2 (1.0%)0 (0%)Abbreviations: APOE Apolipoprotein E, CU Cognitively unimpaired, MCI Mild Cognitive Impairment; p-tau217, phosphorylated-tau217, MMSE Mini-Mental State Examination, CDR-SB Clinical Dementia Rating-Sum of Boxes, SD standard deviation†Positive amyloid PET scan is defined as 18F-florbetapir (FBP) PET SUVR ≥ 1.11 Baseline characteristics of study participants 209/0/0 (100.0/0.0/0.0) 0/173/11 (0.0/94.0/6.0) Abbreviations: APOE Apolipoprotein E, CU Cognitively unimpaired, MCI Mild Cognitive Impairment; p-tau217, phosphorylated-tau217, MMSE Mini-Mental State Examination, CDR-SB Clinical Dementia Rating-Sum of Boxes, SD standard deviation †Positive amyloid PET scan is defined as 18F-florbetapir (FBP) PET SUVR ≥ 1.11 Longitudinal changes in plasma biomarkers Using linear mixed-effects models, both CU and CI groups showed significant longitudinal increases over time in all five plasma p-tau217 assays (q < 0.05 after FDR correction) (Fig. 1). Among these, p-tau217 measured by the ALZpath-Simoa assay showed numerically higher standardized time effects in both groups (CU: βstd = 0.112, 95% CI [0.092, 0.132]; CI: βstd = 0.100, 95% CI [0.069, 0.131]), although the CIs overlapped across the assay platforms, indicating no significant difference. When stratified by Aβ status, Aβ + CU participants exhibited significantly steeper increases in p-tau217 over time compared with Aβ − individuals, with time × amyloid interaction coefficients (βint) ranging from 0.102 to 0.164 across assays. The largest interaction effect was observed for the ALZpath-Simoa assay (βint = 0.164) (Fig. 1A). Similarly, in CI, Aβ + participants showed greater temporal increases in p-tau217 than Aβ − participants, with βint ranging from 0.070 to 0.113, the strongest observed for the C2N-MS p-tau217. (βint = 0.113) (Fig. 1B).Fig. 1Longitudinal trajectories of plasma p-tau217 levels by amyloid status in cognitively unimpaired and cognitively impaired individuals. Longitudinal changes in plasma p-tau217 levels measured by five different assays over time, stratified by cognitive status and amyloid PET positivity. (A) CU participants (B) CI participants. Each panel represents a different p-tau217 assay platform (ALZpath-Simoa, C2N-MS p-tau217, C2N-MS %p-tau217, Fujirebio-Lumipulse, and Janssen-Simoa). Lines represent expected trajectories calculated from the LMM for the total cohort (black), Aβ + (blue), and Aβ- (red) subgroups. Standardized beta coefficient of time (βstd) and interaction beta coefficient (βint) are shown in each panel. Asterisks denote statistical significance of βstd (* q < 0.05, ** q < 0.01, *** q < 0.001). Shaded areas indicate 95% confidence intervals. Abbreviations: Aβ, beta-amyloid; Aβ +, amyloid-positive; Aβ-, amyloid-negative; CU, Cognitively unimpaired; CI, Cognitively impaired; LMM, Linear mixed-effects model; PET, Positron emission tomography Longitudinal trajectories of plasma p-tau217 levels by amyloid status in cognitively unimpaired and cognitively impaired individuals. Longitudinal changes in plasma p-tau217 levels measured by five different assays over time, stratified by cognitive status and amyloid PET positivity. (A) CU participants (B) CI participants. Each panel represents a different p-tau217 assay platform (ALZpath-Simoa, C2N-MS p-tau217, C2N-MS %p-tau217, Fujirebio-Lumipulse, and Janssen-Simoa). Lines represent expected trajectories calculated from the LMM for the total cohort (black), Aβ + (blue), and Aβ- (red) subgroups. Standardized beta coefficient of time (βstd) and interaction beta coefficient (βint) are shown in each panel. Asterisks denote statistical significance of βstd (* q < 0.05, ** q < 0.01, *** q < 0.001). Shaded areas indicate 95% confidence intervals. Abbreviations: Aβ, beta-amyloid; Aβ +, amyloid-positive; Aβ-, amyloid-negative; CU, Cognitively unimpaired; CI, Cognitively impaired; LMM, Linear mixed-effects model; PET, Positron emission tomography Linear mixed-effects models were used to evaluate the contribution of covariates (baseline age, sex, APOE ε4, and education) to the variance explained in longitudinal p-tau217 trajectories across assay platforms. Across all platforms, covariate adjustment resulted in modest increases in marginal R2, with %p-tau217 (C2N-MS) showing the largest gain in both CU (0.027 to 0.114) and CI (0.025 to 0.126) (Fig. 2). Overall patterns of individual covariate contributions were consistent across platforms and are presented in Supplementary Fig. 1. In general, age contributed more prominently in CU, whereas APOE ε4 showed relatively greater contributions in CI; however, these effects were modest and consistent across platforms.Fig. 2Effect of covariate adjustment on longitudinal Δp-tau217. Marginal R2 values are depicted in each panel, visualizing additional contributions of demographic and genetic covariates on LMMs for predicting plasma p-tau217 trajectory. The analysis was performed in (A) CU participants and (B) CI participants, stratified by the total cohort and the amyloid positivity. Abbreviations: Aβ, Beta-amyloid; Aβ +, amyloid-positive; CU, Cognitively unimpaired; CI, Cognitively impaired; LMM, Linear mixed-effects model Effect of covariate adjustment on longitudinal Δp-tau217. Marginal R2 values are depicted in each panel, visualizing additional contributions of demographic and genetic covariates on LMMs for predicting plasma p-tau217 trajectory. The analysis was performed in (A) CU participants and (B) CI participants, stratified by the total cohort and the amyloid positivity. Abbreviations: Aβ, Beta-amyloid; Aβ +, amyloid-positive; CU, Cognitively unimpaired; CI, Cognitively impaired; LMM, Linear mixed-effects model Predictive value of annualized plasma biomarker change for cognitive decline Annualized changes in plasma p-tau217 were associated with longitudinal cognitive decline, with distinct patterns across cognitive and amyloid-positive subgroups (Table 2, Fig. 3). In CU, significant associations with CDR-SB progression were observed for p-tau217 measured by the ALZpath-Simoa, C2N-MS, and Fujirebio-Lumipulse assays, as well as for C2N-MS %p-tau217, with standardized β values ranging from 0.038 to 0.075; the largest effect was observed for C2N-MS %p-tau217 (q < 0.05 after FDR correction). In Aβ + CU, %p-tau217 (C2N) showed a nominal association with CDR-SB progression (β = 0.11, uncorrected p = 0.026), but this association did not remain significant after FDR correction. In CI, all p-tau217 assays except ALZpath-Simoa were significantly associated with CDR-SB progression after FDR correction, with standardized β values ranging from 0.134 to 0.238; the strongest association was observed for C2N-MS p-tau217. In Aβ + CI, significant associations were identified for p-tau217 measured by the C2N-MS p-tau217 and Fujirebio-Lumipulse p-tau217, with standardized β values of 0.245 and 0.207, respectively (q < 0.05). Consistent findings were observed in sensitivity analyses excluding individuals with C2N-MS p-tau217 values below the lower limit of quantification (Supplementary Table 1, Supplementary Fig. 2).Table 2Association between annualized change in Plasma p-tau217 and longitudinal cognitive declineCUCIβ [95% CI] (q-value)β [95% CI] (q-value)Total p-tau217 (Alzpath-Simoa)0.049 [0.016, 0.082] (0.010)0.076 [−0.008, 0.16] (0.078) C2N-MS p-tau2170.050 [0.015, 0.085] (0.010)0.238 [0.156, 0.32] (< 0.001) C2N-MS %p-tau2170.075 [0.040, 0.110] (< 0.001)0.134 [0.048, 0.22] (0.003) p-tau217 (Fujirebio-Lumipulse)0.038 [0.003, 0.073] (0.042)0.136 [0.054, 0.218] (0.003) p-tau217 (Janssen-Simoa)0.030 [−0.005, 0.065] (0.091)0.157 [0.075, 0.239] (0.001)Aβ(+) p-tau217 (Alzpath-Simoa)0.037 [−0.059, 0.133] (0.450)0.021 [−0.132, 0.174] (0.791) C2N-MS p-tau2170.049 [−0.051, 0.149] (0.426)0.245 [0.096, 0.394] (0.009) C2N-MS %p-tau2170.110 [0.016, 0.204] (0.130)0.097 [−0.06, 0.254] (0.284) p-tau217 (Fujirebio-Lumipulse)0.050 [−0.044, 0.144] (0.426)0.207 [0.054, 0.360] (0.023) p-tau217 (Janssen-Simoa)0.055 [−0.041, 0.151] (0.426)0.103 [−0.05, 0.256] (0.284)Abbreviations: CU Cognitively unimpaired, CI Cognitively impaired, Aβ Beta-amyloid, β Standardized beta coefficient, 95% CI 95% confidence interval, p-tau Phosphorylated tauFig. 3Association between annualized change in plasma p-tau217 and longitudinal cognitive decline. The plot is derived from LMM demonstrating the association between annualized change in plasma p-tau217 levels and longitudinal trajectories of CDR-SB scores in (A) CU and CU Aβ + participants, and (B) CI and CI Aβ + participants. Lines represent predicted CDR-SB trajectories for participants with different levels of Δp-tau217: −1SD (red), mean (blue), and + 1SD (green) from the mean change. Each panel represents a different p-tau217 assay platform (ALZpath-Simoa, C2N-MS p-tau217, C2N-MS %p-tau217, Fujirebio-Lumipulse, and Janssen-Simoa). Asterisks denote statistical significance of β coefficient of Δp-tau217. (* q < 0.05, ** q < 0.01, *** q < 0.001). Shaded areas indicate 95% confidence intervals. Abbreviations: Aβ, Beta-amyloid; Aβ +, amyloid-positive; CDR-SB, Clinical Dementia Rating-Sum of boxes; CU, Cognitively unimpaired; CI, Cognitively impaired; Δp-tau217, Annualized change of plasma p-tau217; LMM, linear mixed-effects model; n.s., not significant; SD, standard deviation Table 2Association between annualized change in Plasma p-tau217 and longitudinal cognitive declineCUCIβ [95% CI] (q-value)β [95% CI] (q-value)Total p-tau217 (Alzpath-Simoa)0.049 [0.016, 0.082] (0.010)0.076 [−0.008, 0.16] (0.078) C2N-MS p-tau2170.050 [0.015, 0.085] (0.010)0.238 [0.156, 0.32] (< 0.001) C2N-MS %p-tau2170.075 [0.040, 0.110] (< 0.001)0.134 [0.048, 0.22] (0.003) p-tau217 (Fujirebio-Lumipulse)0.038 [0.003, 0.073] (0.042)0.136 [0.054, 0.218] (0.003) p-tau217 (Janssen-Simoa)0.030 [−0.005, 0.065] (0.091)0.157 [0.075, 0.239] (0.001)Aβ(+) p-tau217 (Alzpath-Simoa)0.037 [−0.059, 0.133] (0.450)0.021 [−0.132, 0.174] (0.791) C2N-MS p-tau2170.049 [−0.051, 0.149] (0.426)0.245 [0.096, 0.394] (0.009) C2N-MS %p-tau2170.110 [0.016, 0.204] (0.130)0.097 [−0.06, 0.254] (0.284) p-tau217 (Fujirebio-Lumipulse)0.050 [−0.044, 0.144] (0.426)0.207 [0.054, 0.360] (0.023) p-tau217 (Janssen-Simoa)0.055 [−0.041, 0.151] (0.426)0.103 [−0.05, 0.256] (0.284)Abbreviations: CU Cognitively unimpaired, CI Cognitively impaired, Aβ Beta-amyloid, β Standardized beta coefficient, 95% CI 95% confidence interval, p-tau Phosphorylated tau Association between annualized change in Plasma p-tau217 and longitudinal cognitive decline Abbreviations: CU Cognitively unimpaired, CI Cognitively impaired, Aβ Beta-amyloid, β Standardized beta coefficient, 95% CI 95% confidence interval, p-tau Phosphorylated tau Association between annualized change in plasma p-tau217 and longitudinal cognitive decline. The plot is derived from LMM demonstrating the association between annualized change in plasma p-tau217 levels and longitudinal trajectories of CDR-SB scores in (A) CU and CU Aβ + participants, and (B) CI and CI Aβ + participants. Lines represent predicted CDR-SB trajectories for participants with different levels of Δp-tau217: −1SD (red), mean (blue), and + 1SD (green) from the mean change. Each panel represents a different p-tau217 assay platform (ALZpath-Simoa, C2N-MS p-tau217, C2N-MS %p-tau217, Fujirebio-Lumipulse, and Janssen-Simoa). Asterisks denote statistical significance of β coefficient of Δp-tau217. (* q < 0.05, ** q < 0.01, *** q < 0.001). Shaded areas indicate 95% confidence intervals. Abbreviations: Aβ, Beta-amyloid; Aβ +, amyloid-positive; CDR-SB, Clinical Dementia Rating-Sum of boxes; CU, Cognitively unimpaired; CI, Cognitively impaired; Δp-tau217, Annualized change of plasma p-tau217; LMM, linear mixed-effects model; n.s., not significant; SD, standard deviation Incremental contribution of covariates and additional plasma biomarkers We next evaluated whether demographic and genetic covariates or other plasma biomarkers provided additional explanatory value beyond longitudinal p-tau217 change. Stepwise models evaluating demographic and genetic covariates showed that marginal R2 remained essentially unchanged in CU and Aβ + CU, indicating minimal covariate contribution beyond the baseline model (Fig. 4). In contrast, in CI and Aβ + CI, inclusion of education resulted in a modest increase in marginal R2, whereas other covariates had negligible effects. Consistent findings were observed in the order-independent penalized regression analysis. Using participant-specific longitudinal slopes derived from the same mixed-effects model, baseline age demonstrated the highest selection stability in CU. In CI, both age and APOE ε4 were retained under penalization, indicating that multiple factors contribute to longitudinal change in this group. In Aβ + CI, no covariate was retained under penalization, with all penalized coefficients shrunk to zero in the cross-validated model, paralleling the limited incremental variance observed in the LMM analysis (Supplementary Table 2).Fig. 4Stepwise contribution of covariates to variance explained in CDR-SB trajectories with annualized change in plasma p-tau217. Marginal R2 values from linear mixed-effects models examining the additive contribution of demographic and genetic covariates to CDR-SB trajectory prediction when combined with annualized change in plasma p-tau217 (C2N-MS assay) in (A) CU and CU Aβ + participants and (B) CI and CI Aβ + participants. The p-tau217 change was analyzed as percent change (Δ%p-tau217) for CU participants and absolute change (Δp-tau217) for CI participants. Each dot represents the marginal R2 value as covariates are sequentially added to the base model (CDR-SB ~ time × p-tau217 change). Covariates were incorporated in the following order: age, sex, APOE ε4 carrier status, and education. Specific marginal R2 values are labeled adjacent to each dot. Abbreviations: Aβ, Beta-amyloid; Aβ +, amyloid-positive; APOE, Apolipoprotein E; CDR-SB, Clinical Dementia Rating-Sum of boxes; CU, Cognitively unimpaired; CI, Cognitively impaired Stepwise contribution of covariates to variance explained in CDR-SB trajectories with annualized change in plasma p-tau217. Marginal R2 values from linear mixed-effects models examining the additive contribution of demographic and genetic covariates to CDR-SB trajectory prediction when combined with annualized change in plasma p-tau217 (C2N-MS assay) in (A) CU and CU Aβ + participants and (B) CI and CI Aβ + participants. The p-tau217 change was analyzed as percent change (Δ%p-tau217) for CU participants and absolute change (Δp-tau217) for CI participants. Each dot represents the marginal R2 value as covariates are sequentially added to the base model (CDR-SB ~ time × p-tau217 change). Covariates were incorporated in the following order: age, sex, APOE ε4 carrier status, and education. Specific marginal R2 values are labeled adjacent to each dot. Abbreviations: Aβ, Beta-amyloid; Aβ +, amyloid-positive; APOE, Apolipoprotein E; CDR-SB, Clinical Dementia Rating-Sum of boxes; CU, Cognitively unimpaired; CI, Cognitively impaired Across all groups, adding individual plasma biomarkers measured using the Simoa-HD-X Neurology 4-Plex E (N4PE) assay (ΔAβ42/40, ΔGFAP, or ΔNfL) to Δp-tau217 models (analyzed as % change for CU and absolute change for CI) modestly increased marginal R2 (Fig. 5). The largest improvements in marginal R2 were observed when adding ΔGFAP in CU (ΔR2 = 0.069) and Aβ + CU (ΔR2 = 0.079), whereas ΔNfL provided more modest gains in CI (ΔR2 = 0.018) and Aβ + CI (ΔR2 = 0.024). However, simultaneous inclusion of all biomarkers did not further improve marginal R2 compared to the best single-marker models, indicating overlapping explanatory contributions.Fig. 5Incremental value of plasma biomarkers added to p-tau217 models for CDR-SB trajectory prediction. Marginal R2 values from linear mixed-effects models examining the incremental value of additional plasma biomarkers when combined with Δp-tau217 for predicting longitudinal CDR-SB trajectories. The analysis includes (A) CU and CU Aβ + participants and (B) CI and CI Aβ + participants. Each dot represents the marginal R2 value for models incorporating different combinations of annualized biomarker changes. The base model includes annualized change in p-tau217 (analyzed as Δ%p-tau217 for CU and absolute Δp-tau217 for CI) with demographic covariates (baseline age, sex, APOE ε4, education). Subsequent models sequentially add ΔAβ42/Aβ40, ΔGFAP, ΔNfL, or all biomarkers combined. Dotted vertical lines represents the marginal R2 of base model. Abbreviations: Aβ, Beta-amyloid; Aβ +, amyloid-positive; APOE, Apolipoprotein E; CDR-SB, Clinical Dementia Rating-Sum of boxes; CU, Cognitively unimpaired; CI, Cognitively impaired; GFAP, Glial fibrillary acidic protein; NfL, Neurofilament light chain Incremental value of plasma biomarkers added to p-tau217 models for CDR-SB trajectory prediction. Marginal R2 values from linear mixed-effects models examining the incremental value of additional plasma biomarkers when combined with Δp-tau217 for predicting longitudinal CDR-SB trajectories. The analysis includes (A) CU and CU Aβ + participants and (B) CI and CI Aβ + participants. Each dot represents the marginal R2 value for models incorporating different combinations of annualized biomarker changes. The base model includes annualized change in p-tau217 (analyzed as Δ%p-tau217 for CU and absolute Δp-tau217 for CI) with demographic covariates (baseline age, sex, APOE ε4, education). Subsequent models sequentially add ΔAβ42/Aβ40, ΔGFAP, ΔNfL, or all biomarkers combined. Dotted vertical lines represents the marginal R2 of base model. Abbreviations: Aβ, Beta-amyloid; Aβ +, amyloid-positive; APOE, Apolipoprotein E; CDR-SB, Clinical Dementia Rating-Sum of boxes; CU, Cognitively unimpaired; CI, Cognitively impaired; GFAP, Glial fibrillary acidic protein; NfL, Neurofilament light chain Discussion In this study, we investigated longitudinal changes in plasma p-tau217 across five assay platforms and their prognostic value for cognitive decline in CU and CI individuals, with additional stratified analyses focusing on Aβ + subgroups. Our major findings were as follows. First, plasma p-tau217 levels increased consistently over time across all assay platforms in both CU and CI, with more pronounced longitudinal trajectories in Aβ + individuals. In supportive analyses, adding age improved explanatory power in CU, whereas APOE ε4 status contributed more meaningfully to p-tau217 trajectories in CI. Second, prognostic value of longitudinal p-tau217 changes for cognitive decline was both stage- and assay-dependent, with ΔC2N-MS %p-tau217 showing the strongest associations in CU, whereas Δp-tau217 measured by C2N-MS and Fujirebio-Lumipulse most robustly predicted CDR-SB progression in CI, particularly among Aβ + individuals. Finally, incremental prognostic value of adding other plasma biomarkers to p-tau217 models was modest and stage-dependent, with ΔGFAP providing the greatest additional explanatory value in CU—particularly in Aβ + CU—whereas in CI and Aβ + CI, ΔAβ42/40, ΔGFAP, and ΔNfL each yielded small and comparable improvements without further benefit from combining all markers. Taken together, our findings demonstrate that longitudinal plasma p-tau217 is a robust marker of disease progression across the AD continuum, but its optimal prognostic use—including assay selection and complementary biomarkers—differs by cognitive stage and amyloid status. From a clinical perspective, these results suggest that longitudinal plasma biomarker monitoring can be flexibly adapted to an individual’s disease stage and Aβ status, enabling more precise prognostication and personalized follow-up strategies in both preclinical and symptomatic AD. Our first major finding was that plasma p-tau217 levels increased consistently over time across all five assay platforms in both CU and CI individuals, with more pronounced longitudinal trajectories observed in Aβ + participants. This pattern was robust across assay platforms, supporting p-tau217 as a consistent longitudinal biomarker associated with presumed progressive AD-related pathophysiology. It is notable that less than half of the CI participants were Aβ-positive. This proportion reflects the composition of the ADNI dataset we utilized, where the CI group predominantly consisted of individuals with mild cognitive impairment (n = 173) rather than dementia (n= 11). Notably, the factors contributing to inter-individual variability in p-tau217 trajectories differed by cognitive stage and amyloid status. In CU individuals, baseline age generally resulted in the largest or comparable increase in marginal R2 across platforms, whereas APOE ε4 carrier status played a more prominent role in CI. However, this pattern was reversed in the amyloid-positive subgroup in CI, where baseline age emerged as the dominant contributor. This pattern is broadly consistent with prior observations but also highlights additional complexity [31–33]. While age-related processes were more prominent in preclinical stages and APOE ε4 status in cognitively impaired individuals, the dominance of age in amyloid-positive subgroups suggests that these relationships are further influenced by underlying pathology. Together, these findings indicate that longitudinal p-tau217 trajectories are shaped by distinct biological contexts across the AD continuum, underscoring the need for cognitive stage–specific interpretation when using p-tau217 for disease monitoring in both research and clinical settings. Our second major finding was that the prognostic value of longitudinal p-tau217 changes for cognitive decline was both stage- and assay-dependent. In CU individuals, Δ%p-tau217 measured by the C2N-MS assay showed the strongest association with CDR-SB progression (β = 0.075), whereas in CI individuals, absolute Δp-tau217 measured by C2N-MS demonstrated the highest standardized effect (β = 0.238). The superior performance of Δ%p-tau217 in CU may reflect advantages of ratio-based biomarkers in populations with low baseline p-tau217 levels, where absolute concentrations are more vulnerable to biological confounders such as kidney dysfunction or anemia, as suggested by prior studies [19, 34, 35]. In contrast, CI individuals exhibit substantially elevated p-tau217 levels driven by AD pathology, potentially diminishing the relative impact of such confounders and allowing absolute Δp-tau217 to retain strong prognostic value without normalization. Notably, subgroup analyses further showed that associations were attenuated in Aβ + CU and became restricted to fewer assays in Aβ + CI, suggesting that as Aβ pathology advances, the prognostic signal of p-tau217 becomes more assay- and context-specific. This may be directly driven by the phenomenon where different p-tau assays plateau at varying levels of cortical amyloid burden [36]. Together, these findings underscore the importance of tailoring assay selection and analytic strategies to both cognitive stage and amyloid status when using longitudinal p-tau217 for prognostic purposes. Our final major finding was that the incremental prognostic value of adding other plasma biomarkers to p-tau217 models was modest and strongly stage-dependent. In CU individuals, particularly those who were Aβ +, adding ΔGFAP to Δp-tau217 resulted in a meaningful improvement in model performance (marginal R2 from 0.087 to 0.156 in CU and from 0.162 to 0.241 in Aβ + CU), suggesting that astroglial activation captures early disease processes not fully reflected by tau changes alone. This pattern was consistent with covariate analyses showing minimal contributions from demographic or genetic factors in CU, indicating that biomarker-level information predominates at this stage. In contrast, in CI and Aβ + CI, adding ΔAβ42/40, ΔGFAP, or ΔNfL led to only modest and comparable improvements, with no further gain from combining all biomarkers. This likely reflects the convergence of multiple pathological processes in later disease stages, where p-tau217 already captures much of the variance related to cognitive decline. Together, these findings highlight the need for stage-specific biomarker integration strategies, as the incremental value of additional plasma markers depends on baseline cognitive status. Our findings align with a large body of literature, including recent work by Karlsson et al., demonstrating the modest incremental value of multi-marker fluid models. [37]. Several other studies have suggested that using multiple biomarkers related to neurodegeneration such as GFAP or NfL may be helpful in predicting cognitive decline [38, 39]. Furthermore, our head-to-head comparison builds upon recent large-scale initiatives, such as the round-robin study evaluating multiple p-tau assays [40]. Assay discrepancies might also be partially influenced by non-pathological fluctuations in plasma p-tau217, as highlighted by Brum et al. [41]. A major strength of this study is the comprehensive, head-to-head comparison of multiple plasma biomarker assays, including five distinct p-tau217 platforms, within a well-characterized longitudinal cohort. This study expands upon existing literature by providing a comprehensive, head-to-head longitudinal comparison of multiple p-tau217 assay platforms, an area where longitudinal data has been limited. Clinically, these findings suggest that assay selection should be tailored: ratio-based biomarkers (e.g., %p-tau217) may be optimal for screening early-stage individuals in clinical trials, whereas absolute concentration assays may be more effective for tracking progression in advanced clinical settings. Several limitations should nonetheless be considered. First, the follow-up duration may have been insufficient to capture longer-term biomarker trajectories. Second, subgroup analyses—particularly those stratified by Aβ status—were based on modest sample sizes, potentially limiting statistical power. Third, plasma sampling and PET imaging were not always contemporaneous, introducing possible temporal misalignment between modalities. Fourth, although key demographic and genetic covariates were accounted for, residual confounding cannot be fully excluded. Finally, these results are derived from a single, well-characterized cohort, and thus may not be fully generalizable to the broader, more diverse population. As the ADNI cohort consists solely of older adults (ages 55–90), a young reference group was unavailable for z-transformation. However, we adjusted for baseline age as a covariate in all statistical models to account for age-related biomarker increases. Future studies in heterogeneous populations are warranted. Despite these limitations, our findings provide robust evidence that longitudinal changes in plasma biomarkers, especially p-tau217, closely track disease progression and cognitive decline across assay platforms. Collectively, these results support the utility of serial plasma biomarker monitoring for both mechanistic studies and clinical prognostication in AD. In conclusion, our study demonstrates that longitudinal plasma p-tau217 is a robust and clinically meaningful biomarker of disease progression across the AD continuum, while its optimal prognostic application depends on cognitive stage, amyloid status, and assay characteristics. These findings highlight the potential of stage-tailored, assay-informed plasma biomarker strategies to support longitudinal disease monitoring, risk stratification, and biomarker-guided clinical decision-making in both research and clinical settings.