Abstract
INTRODUCTION
We examined when Alzheimer's disease biomarkers become informative by identifying age‐related breakpoints with slope‐changing trajectories.
METHODS
In 2082 Mayo Clinic Study of Aging participants, we modeled plasma amyloid beta (Aβ)42/40, phosphorylated tau (p‐tau)181, glial fibrillary acidic protein (GFAP), neurofilament light chain (NfL), amyloid/tau positron emission tomography (PET), hippocampal volume (HVa), and global cognition. Generalized additive models described age trends; Davies' test and piecewise linear regression estimated breakpoints. A C2N subsample (n = 462) provided mass‐spectrometry plasma markers (p‐tau181, p‐tau217, their ratios, Aβ42/40).
RESULTS
In the full cohort, Aβ42/40, HVa, and cognition declined with age; p‐tau181, NfL, GFAP, and amyloid/tau PET increased. We observed single breakpoints (years, 95% CI): GFAP 68.1 (63.5–71.8), NfL 70.7 (65.9–75.6), p‐tau181 67.2 (60.3–70.3), amyloid PET 62.3 (56.2–69.3), HVa 68.1 (63.1–71.9), cognition 59.8 (55.4–66.0); tau PET showed none. In the mass‐spectrometry subset, p‐tau217 and p‐tau181 broke at 72.6; their ratios and Aβ42/40 showed no breakpoints.
DISCUSSION
Breakpoints cluster near late midlife, suggesting windows for screening and monitoring.
Keywords: age‐related breakpoint thresholds, Alzheimer's disease, amyloid, amyloid/tau positron emission tomography, blood‐based biomarker, global cognition, Mayo Clinic Study of Aging, mild cognitive impairment, neurodegeneration, tau
Highlights
Breakpoint modeling identifies age thresholds in Alzheimer's disease biomarker trajectories.
Plasma glial fibrillary acidic protein, neurofilament light chain, and phosphorylated tau markers inflect at ages ≈ 68 to 72 years.
Findings replicate across Quanterix and C2N platforms.
Inflection points reveal midlife acceleration of neurodegenerative change.
Results refine optimal timing for screening and preventive interventions.
1. INTRODUCTION
Biomarkers for Alzheimer's disease (AD) have transformed the diagnosis and staging of the disease, enabling in vivo assessment of AD pathology. Recently, less invasive and less expensive blood‐based biomarker (BBM) tests have emerged as reliable tools for predicting cognitive decline 1 , 2 and measuring brain amyloid, 3 tau, 4 and neurodegeneration. 5 The integration of BBM tests with genetic, clinical, and demographic data offers a promising approach for effective AD screening. These tests may assist in selecting individuals for more comprehensive, invasive biomarker evaluations and influence treatment decisions. Notably, preclinical AD clinical trials are already incorporating BBM tests into their study protocols. 6
Many studies evaluating BBMs have primarily relied on convenience samples or healthier populations, which limits their generalizability and our understanding of ideal timing to consider screening. This leaves significant gaps in our understanding of how biomarker distribution progresses in the broader population and the optimal timing for screening. The primary goals of this study were to explore the distribution and aging patterns of plasma biomarkers, imaging biomarkers, and global cognition using cross‐sectional data. Because screening strategies depend on knowing when age‐related changes in plasma and imaging biomarkers occur, breakpoint modeling provides a way to identify threshold ages at which age‐related increases or decreases in biomarker levels become more pronounced, enabling more timely intervention. To investigate these questions, we fitted piecewise linear regression with two segments separated by a breakpoint, which represents an abrupt change in the slope of a biomarker–age trajectory. The breakpoint can be interpreted as a threshold age beyond which the association between age and biomarker levels becomes steeper in the direction of increase or decrease. Identifying such thresholds may help determine when biomarkers first become clinically informative, refine the timing of screening, and guide interventions such as enrollment in preclinical trials or initiation of preventive care. We examined the biomarkers in the overall sample including the plasma amyloid beta (Aβ) 42/40 ratio, phosphorylated tau (p‐tau) protein at threonine 181 (p‐tau181), glial fibrillary acidic protein (GFAP), neurofilament light chain (NfL), amyloid positron emission tomography (PET), tau PET, hippocampal volume adjusted for total intracranial volume (HVA), and global cognition (Pzglobal). We further identified a subsample for participants with available mass‐spectrometry measures of plasma p‐tau181, p‐tau217, p‐tau217 relative to non‐phosphorylated tau protein at threonine 217 ratio (p‐tau217/np‐tau217), and p‐tau217 relative to non‐phosphorylated tau protein at threonine 217 ratio (p‐tau181/np‐tau181).
2. METHODS
2.1. Participants
Participants were enrolled in the Mayo Clinic Study of Aging (MCSA), which is a population‐based study that examines the epidemiology of cognitive decline and risk of mild cognitive impairment (MCI) and dementia among residents living in Olmsted County, Minnesota. The study procedures have been previously described in detail. 7 Residents were randomly chosen for recruitment using the Rochester Epidemiology Project medical records‐linkage system. 8 , 9 The participants underwent detailed clinical visits including neuropsychological testing, a physician examination, and blood draws at ≈ 15‐month intervals for those aged ≥ 50, and at ≈ 30‐month intervals for those < 50. A subset of participants volunteered for neuroimaging. For this analysis, we included 2082 participants with available plasma AD BBMs, spanning cognitively unimpaired, MCI, and late‐onset dementia. Demographic information included self‐reported age and sex and were collected as part of the clinical evaluation. The study was approved by Mayo Clinic and Olmsted Medical Center Institutional Review Boards. Written informed consent was obtained from all participants.
2.2. Cognition function assessments and clinical diagnosis
Neuropsychological testing was administered by a trained psychometrist and included nine tests across four domains: memory, language, executive function, and visuospatial skills. A global cognition z score was computed as previously described (referenced to the 2012 MCSA cognitively unimpaired cohort and weighted back to the Olmsted County population). 7 , 10 A summary score representing global cognition was estimated from the z transform of the mean of the four domain z scores.
A clinical consensus committee, blind to the diagnosis of the previous study visit, assessed each participant to determine cognitive diagnoses. Cognitive performance was compared to age‐adjusted scores using Mayo's Older Americans Normative Studies. 11 , 12 The operational definition of MCI was based on clinical judgment, considering the participant's and informant's history and cognitive performance. Published criteria were used for the diagnosis of MCI: cognitive complaint, cognitive function not normal for age, essentially normal functional activities, no dementia. 13 Participants who performed in the normal range and did not meet criteria for MCI or dementia were categorized as cognitively unimpaired.
2.3. Plasma assays
Ethylenediaminetetraacetic acid plasma samples were collected from participants after an overnight fast. The samples were centrifuged, and 500 µL of plasma was aliquoted into polypropylene tubes and stored at –80°C until testing. The Simoa Neurology 4‐Plex E Advantage kit (N4PE, item #103,670) was used to measure plasma Aβ40, Aβ42, GFAP, and NfL. The Simoa p‐tau181 Advantage V2 kit (item #103,714) was used to measure p‐tau181. Both kits were used according to the manufacturer's instructions and were run on a Quanterix HD‐X analyzer (Quanterix).
After thawing and mixing, plasma samples were centrifuged for 5 minutes at 4000 g. Samples were diluted 1:4 and tested in singlet. Calibration curves consisting of seven points were used for p‐tau181, while eight‐point calibration curves with 1/y2 weighting were used for NfL, GFAP, Aβ1‐40, and Aβ1‐42. Concentrations of the samples and calibration curves were determined using the Simoa HD‐X Analyzer software. Details of quality control and calibration were previously published. 14 Furthermore, a selection of the participant plasma samples was analyzed by the C2N Diagnostics commercial laboratory with an immunoprecipitation mass spectrometry assay and measurements on p‐tau181, n p‐tau181, p‐tau217, and np‐tau181 were available.
2.4. Imaging
2.4.1. Structural MRI
Neuroimaging assessments were conducted at 15‐ (cognitively impaired), 30‐ (aged ≥ 50), or 60‐month (< 50) intervals. Structural magnetic resonance imaging (MRI) was performed using standardized magnetization‐prepared rapid gradient echo (MPRAGE) sequences on 3T GE scanners (GE Medical Systems) as well as Siemens scanners. Hippocampal volume was measured using FreeSurfer (version 5.3). Raw hippocampal volume for each participant was adjusted for total intracranial volume (TIV) to obtain a TIV‐adjusted hippocampal volume. 15
RESEARCH in CONTEXT
Systematic Review: Prior studies using population and clinical cohorts have characterized trajectories of plasma and imaging biomarkers across the Alzheimer's disease continuum, showing that amyloid changes precede tau and neurodegeneration. However, few have quantified threshold ages at which these changes accelerate, particularly using population‐based samples or comparing biomarker platforms (e.g., Quanterix vs. C2N).
Interpretation: Our findings demonstrate that breakpoint modeling identifies reproducible threshold ages for plasma glial fibrillary acidic protein, neurofilament light chain, and tau biomarkers. These are typically in the late 60s to early 70s and coincide with accelerated neurodegenerative change. These results extend prior research by quantifying non‐linear inflection points and confirming the robustness of plasma tau markers across assays.
Future Directions: Longitudinal studies integrating positron emission tomography, plasma, and cognitive data should test whether identified breakpoints predict future impairment. Bayesian changepoint and adaptive trend modeling may enhance detection of subtle, individualized transitions in biomarker dynamics.
2.4.2. PET imaging
PET imaging was performed following the methods detailed in previous works. 16 Amyloid PET imaging was performed with 11C‐Pittsburgh compound B while tau PET was performed with AV‐1451, synthesized on site with precursor supplied by Avid Radiopharmaceuticals. Late‐uptake amyloid PET images were obtained 40 to 60 minutes after injection and tau PET 80 to 100 minutes after injection with computed tomography used for attenuation correction.
We conducted the analysis on PET imaging biomarkers using a fully automated, in‐house image‐processing pipeline, which extracted image voxel values from regions of interest (ROIs) that were automatically labeled and propagated from the MCALT template. 17 The standardized uptake value ratio (SUVR) values of amyloid and tau PET were determined by normalizing target ROIs to the cerebellar crus gray matter. 18 A cut‐point of > 1.48 SUVR was used to categorize participants as amyloid positive. 18
A voxel number‐weighted average of the median tau PET uptake from previously established ROIs was calculated to create a tau PET meta‐ROI. 18 , 19 Tau PET positivity was defined by meta‐ROI SUVR ≥ 1.25 as previously described. 20 PET image quantifications were conducted using participants’ MRI scans, though the PET data lacked partial volume correction.
2.5. Statistics
Demographic, clinical, and imaging characteristics were summarized via medians and interquartile ranges for continuous variables, and counts and percentages for categorical variables, for both the full sample with Quanterix plasma biomarkers and the subsample which limits to participants with mass spectrometry C2N plasma biomarkers. Missingness for tau PET reflects its later introduction in the study protocol, while missingness for global cognition z scores (6% in the full sample) primarily reflects incomplete neuropsychological testing at the time of visit and was not systematic, for example, due to fatigue, scheduling constraints, sensory or motor limitations, or other practical factors.
2.5.1. Associations of the biomarkers and cognition with age
Smooth curves for the biomarkers and cognition versus age were estimated and plotted using thin plate regression splines in generalized additive models (GAMs), fitted using the R mgcv package, 21 and with degree of smoothness selected using the generalized cross validation (GCV) criteria by minimizing the GCV score to balance model complexity and overfitting. GAMs model smooth, non‐linear relationships across different data segments without requiring prior knowledge of the data. Standard errors for the predicted curves were based upon the Bayesian posterior covariance matrix of the fitted GAM object. The GAMs with cognition as the outcome were adjusted for cycle number as a covariate, which indexes the sequential MCSA evaluation cycles at which participants were seen to account for potential visit‐related and procedural differences across study waves, rather than to model individual calendar time. The smooth curves were generated using all the data but were plotted only from ages 45 to 90 to prevent showing predictions in cases in which the data were sparse.
2.5.2. Breakpoint models
Breakpoint regression models for the biomarkers and cognition versus age were estimated using the R segmented package. 22 , 23 The breakpoint model with cognition as the outcome was again adjusted for cycle number as a covariate. Breakpoint models fit straight lines to data, allowing the slopes of these lines to differ before and after the breakpoints, which capture critical transitions in age‐related patterns at the population level. The segmented package in R estimates the optimal locations of these breakpoints by iteratively adjusting the points where the slope changes and then fitting piecewise linear functions. It also provides 95% confidence intervals (CIs) for the breakpoint locations, offering a statistical measure of uncertainty around these critical points in the model. The Davies' test was used to determine whether zero, one, or two breakpoints were warranted for each regression. Adjusted R 2 values for the smooth curves and breakpoint models were computed to describe goodness of fit. For each biomarker we also added a linear model fit with age as the covariate to provide a comparison to the breakpoint model to verify comparative advantage of the latter in model performance.
As a sensitivity analysis, we repeated the GAM and breakpoint models in samples restricted to cognitively unimpaired participants in the full Quanterix cohort and the C2N subsample.
3. RESULTS
3.1. Participant demographics
Based on the participant characteristics from the full Quanterix immunoassay sample (n = 2082) in Table 1, participants had a median (interquartile range [IQR]) age of 71 (62, 79) years and 1120 (54%) were male. At the time of biomarker evaluation, 1821 (87%) were cognitively unimpaired, 239 (11%) had MCI, and 22 (1.1%) were diagnosed with dementia. Imaging biomarkers included a median (IQR) amyloid PET SUVR of 1.39 (1.32, 1.55) and a tau PET SUVR of 1.18 (1.13, 1.24) among the 628 participants with available data. Global cognition z scores had a median of 0.22 (–0.61, 0.92), and hippocampal volume had a median of –0.30 (–0.74, 0.07).
TABLE 1.
Baseline characteristics for the full Quanterix biomarker sample (N = 2082).
| Characteristic | N | Median [Q1, Q3]; n (%) |
|---|---|---|
| Age | 2082 | 71 [62, 79] |
| Clinical status | 2082 | |
| Cognitively unimpaired | 1821 (87%) | |
| MCI | 239 (11%) | |
| Dementia | 22 (1.1%) | |
| Male | 2082 | 1120 (54%) |
| Education | 2081 | 14.00 [12.00, 16.00] |
| APOE ε4 | 2050 | 582 (28%) |
| Plasma Aβ42/40 | 2082 | 0.060 [0.052, 0.068] |
| Plasma GFAP, pg/mL | 2082 | 94 [61, 141] |
| Plasma NfL, pg/mL | 2082 | 21 [14, 33] |
| Plasma p‐tau181, pg/mL | 2082 | 1.73 [1.29, 2.54] |
| Amyloid PET, SUVR | 2082 | 1.39 [1.32, 1.55] |
| Tau PET, SUVR | 628 | 1.18 [1.13, 1.24] |
| Global cognition, z score | 1955 | 0.22 [–0.61, 0.92] |
| Hippocampal volume, z score | 2082 | −0.30 [–0.74, 0.07] |
Abbreviations: Aβ, amyloid beta; APOE, apolipoprotein E; GFAP, glial fibrillary acidic protein; MCI, mild cognitive impairment; NfL, neurofilament light chain; PET, positron emission tomography; p‐tau, phosphorylated tau; SUVR, standardized uptake value ratio.
For the subsample (n = 462) with data from the C2N assay, Table 2 shows that participants’ median (IQR) age was slightly higher compared to the full sample at 73 (65, 80) years, and 250 (54%) were male. At the time of the biomarker evaluation and compared to those with the Quanterix assays, slightly more were cognitively unimpaired (93% vs. 87%), and fewer had MCI (7.4% vs. 11.0%). Quanterix plasma biomarker levels remained comparable to the full cohort. Compared to the full Quanterix sample, participants in the C2N subsample had slightly lower global cognition (median z score 0.16 vs. 0.22) and reduced hippocampal volume (–0.36 vs. –0.30). Median amyloid and tau PET SUVRs in this group were 1.40 and 1.20, respectively.
TABLE 2.
Baseline characteristics for the subsample with C2N.
| Characteristic | N | Median [Q1, Q3]; n (%) |
|---|---|---|
| Age | 462 | 73 [65, 80] |
| Clinical status | 462 | |
| Cognitively unimpaired | 428 (93%) | |
| MCI | 34 (7.4%) | |
| Male | 462 | 250 (54%) |
| Education | 462 | 14.00 [12.00, 16.00] |
| APOE ε4 | 457 | 141 (31%) |
| Plasma Aβ42/40 | 462 | 0.058 [0.050, 0.066] |
| Plasma GFAP, pg/mL | 462 | 98 [66, 140] |
| Plasma NfL, pg/mL | 462 | 23 [15, 34] |
| Plasma p‐tau181, pg/mL | 462 | 1.85 [1.38, 2.54] |
| Amyloid PET, SUVR | 462 | 1.40 [1.34, 1.53] |
| Tau PET, SUVR | 101 | 1.20 [1.16, 1.26] |
| Global cognition, z score | 441 | 0.16 [–0.64, 0.80] |
| Hippocampal volume, z score | 462 | −0.36 [–0.81, 0.01] |
| C2N p‐tau217, pg/mL | 416 | 0.79 [0.25, 1.38] |
| C2N p‐tau217/np‐tau217 | 415 | 0.008 [0.004, 0.014] |
| C2N p‐tau181, pg/mL | 412 | 12 [9, 16] |
| C2N p‐tau181/np‐tau181 | 411 | 0.16 [0.13, 0.19] |
| C2N Aβ42/40 | 452 | 0.091 [0.084, 0.101] |
Abbreviations: Aβ, amyloid beta; APOE, apolipoprotein E; GFAP, glial fibrillary acidic protein; MCI, mild cognitive impairment; NfL, neurofilament light chain; PET, positron emission tomography; p‐tau, phosphorylated tau; SUVR, standardized uptake value ratio.
3.2. Age‐related biomarker patterns in the population
To study the patterns of biomarkers based on age in the general population, we plotted smooth curves to visualize the relationship between the plasma, imaging biomarkers and cognition versus age as the left column of the paired subplots in Figure 1. The curvature in the spline varies over age, with different ages at which each biomarker shows a more pronounced age‐related change. The scatter points represent per‐participant observations, and the shaded region gives the 95% CI. In the full sample, plasma Aβ42/40, hippocampal volume, and global cognition declined with age. Plasma p‐tau181, NfL, and GFAP increased with age, with a sharper rise beginning at age ≈ 70. Amyloid PET also increased with age, but the rate of increases occurs earlier at age ≈ 60, while tau PET increases without apparent change in rate. Across biomarkers, the GAM explained the largest proportion of variance for NfL (adjusted R 2 = 0.382), with GFAP (0.368) and global cognition (0.345) also > 0.30.
FIGURE 1.
Plotted smooth curves versus cutpoints for plasma and MRI biomarkers for full Quanterix sample. Left subplot: GAM fit with splines. Right subplot: Breakpoint model fit. Aβ, amyloid beta; GAM, generalized additive model; GFAP, glial fibrillary acidic protein; MRI, magnetic resonance imaging; NfL, neurofilament light chain; PET, positron emission tomography; p‐tau, phosphorylated tau.
Fitting splines with GAM using the subsample with C2N observations from the left column of the paired subplots in Figure S1 in supporting information, we observe that hippocampal volume and global cognition showed a steady decline with age, with a slightly higher rate of decrease in cognition as age increases. Plasma p‐tau181, NfL, and GFAP had upward trajectories with age, with a sharper rise at age ≈ 70. Amyloid PET and tau PET increased without apparent change in rate. Plasma Aβ42/40 ratio exhibited a non‐linear trend with age, remaining relatively stable until at age ≈ 75, after which it showed a marked increase. Corresponding model fit statistics are provided in Table S1 in supporting information.
In analyses limited to the C2N subsample (Figure S2 in supporting information), both p‐tau217 and p‐tau181 exhibited non‐linear increases with age that steepened after ≈ 72 years. In contrast, the ratio measures p‐tau217/np‐tau217 and p‐tau181/np‐tau181 showed more linear upward trends. Model fit summaries are given in Table S2 in supporting information.
3.3. Breakpoints
In Table 3 using the full sample, breakpoints were supported for plasma Aβ42/40 (P = 0.002), GFAP, NfL, p‐tau181, amyloid PET, hippocampal volume, and global cognition (all P < 0.001). Tau PET did not show evidence for a breakpoint (P = 0.143), and its analysis was based on a smaller subsample (n = 628) than the other biomarkers. Inflection points were identified in plasma GFAP (68.1 years, 95% CI: 63.5–71.8), NfL (70.7, 65.9–75.6), p‐tau181 (67.2, 60.3–70.3), and amyloid PET (62.3, 56.2–69.3), each showing sharper increases later in life (Figure 1). Plasma Aβ42/40 (48.5, 41.6–58.7) showed an earlier inflection point before age 50, followed by consistent and steady decline. Global cognition was the second earliest slope change (59.8, 55.4–66.0). Among the biomarkers modeled with breakpoints, the adjusted R 2 values were highest for NfL (0.380), GFAP (0.373), global cognition (0.331), and amyloid PET (0.225), indicating that the one‐breakpoint model provides a good fit for their trajectories (Table 4). Tau PET was not modeled due to lack of a breakpoint. The patterns observed in Figure 1 align with these findings, with Aβ42/40 showing the lowest model fit (R 2 = 0.081). The performance of the breakpoint models was comparable to that of the GAM models and consistently superior to that of the corresponding simple linear regression (Table 4), supporting the robustness of the identified breakpoints.
TABLE 3.
Breakpoint model testing for the full Quanterix sample.
| Outcome | P value for 1 breakpoint vs. 0 | Estimated first age breakpoint (95% CI) | P value for 2 breakpoints vs. 1 |
|---|---|---|---|
| Plasma Aβ42/40 | 0.002 | 48.50 (40.61–56.40) | 0.140 |
| Plasma GFAP | <0.001 | 68.07 (65.87–70.27) | 0.074 |
| Plasma NfL | <0.001 | 70.68 (68.75–72.60) | 0.092 |
| Plasma p‐tau181 | <0.001 | 67.16 (63.83–70.48) | 0.923 |
| Amyloid PET | <0.001 | 62.30 (58.55–66.05) | 0.268 |
| Tau PET | 0.143 | – | – |
| Hippocampal volume | <0.001 | 68.07 (65.29–70.85) | 0.869 |
| Global cognition | <0.001 | 59.80 (56.47–63.14) | 0.408 |
Abbreviations: Aβ, amyloid beta; CI, confidence interval; GFAP, glial fibrillary acidic protein; NfL, neurofilament light chain; PET, positron emission tomography; p‐tau, phosphorylated tau.
TABLE 4.
Adjusted R 2 values for the GAM, linear, and breakpoint models on the full Quanterix sample.
| Outcome | GAM splines | Linear model | Breakpoint model |
|---|---|---|---|
| Plasma Aβ42/40 | 0.084 | 0.075 | 0.081 |
| Plasma GFAP | 0.374 | 0.333 | 0.373 |
| Plasma NfL | 0.382 | 0.333 | 0.380 |
| Plasma p‐tau181 | 0.163 | 0.140 | 0.163 |
| Amyloid PET | 0.226 | 0.206 | 0.225 |
| Tau PET | 0.136 | 0.130 | – |
| Hippocampal volume | 0.199 | 0.167 | 0.199 |
| Global cognition | 0.351 | 0.309 | 0.331 |
Abbreviations: Aβ, amyloid beta; GAM, generalized additive model; GFAP, glial fibrillary acidic protein; NfL, neurofilament light chain; PET, positron emission tomography; p‐tau, phosphorylated tau.
In the C2N subsample for biomarkers shared with the full Quanterix sample, one breakpoint was supported for plasma Aβ42/40 (P < 0.001), GFAP (P = 0.002), NfL (P < 0.001), and p‐tau181 (P = 0.023), while the remaining biomarkers showed no evidence of a breakpoint (Table 5). Breakpoint ages shifted to later in life compared to the Quanterix sample for plasma Aβ42/40 (at 88.4 years, 95% CI: 85.9–90.1) and p‐tau181 (72.6, 66.6–78.6). Age‐related changes for these markers, summarized in Figure S1, show visible slope shifts around ages 69 to 73 consistent with the estimated breakpoints. Across all breakpoint models on this sample, NfL showed the strongest model fit (adjusted R 2 = 0.383, Table S1). Plasma Aβ42/40 stands out in having a low adjusted R 2 (0.067), reflecting poor age‐explained variance despite a significant breakpoint. In addition to tau PET in the full Quanterix sample, no inflection points were observed in hippocampal volume, global cognition, or amyloid PET. Breakpoint models performed comparably to GAM models and consistently better than the corresponding linear regression over age, which further supports the validity of the identified breakpoints (Table S1).
TABLE 5.
Breakpoint model testing for biomarkers shared in the C2N subsample.
| Outcome | P value for 1 breakpoint vs. 0 | Estimated first age breakpoint (95% CI) | P value for 2 breakpoints vs. 1 |
|---|---|---|---|
| Plasma Aβ42/40 | <0.001 | 88.35 (85.89–90.81) | 0.217 |
| Plasma GFAP | 0.002 | 68.58 (63.59–73.57) | 0.575 |
| Plasma NfL | <0.001 | 69.51 (65.53–73.49) | 0.959 |
| Plasma p‐tau181 | 0.023 | 72.61 (66.59–78.63) | 0.445 |
| Amyloid PET | 0.210 | – | – |
| Tau PET | 0.591 | – | – |
| Hippocampal volume | 0.327 | – | – |
| Global cognition | 0.114 | – | – |
Abbreviations: Aβ, amyloid beta; GFAP, glial fibrillary acidic protein; NfL, neurofilament light chain; PET, positron emission tomography; p‐tau, phosphorylated tau.
Among the plasma biomarkers unique to the C2N subsample, a single breakpoint was identified for both p‐tau217 (P = 0.003) and p‐tau181 (P < 0.001), each occurring at an estimated age of 72.6 years (Table 6). As shown in Figure S2, both markers displayed clear inflection points, with steeper increases emerging later in life, consistent with stronger late‐life age‐related increases in tau‐related signal. These visual patterns were consistent with the relatively strong model fits (Table S2 in supporting information), which performed similarly to the GAM fits and consistently better than the simple linear regression. In contrast, the ratio measures p‐tau217/np‐tau217, p‐tau181/np‐tau181, and Aβ42/40 in the C2N subsample showed no evidence of breakpoints.
TABLE 6.
Breakpoint model testing for unique C2N biomarkers in the C2N subsample.
| Outcome | P value for 1 breakpoint vs. 0 | Estimated first age breakpoint (95% CI) | P value for 2 breakpoints vs. 1 |
|---|---|---|---|
| C2N p‐tau217 | 0.003 | 72.61 (67.44–77.79) | 0.358 |
| C2N p‐tau217/np‐tau217 | 0.184 | – | – |
| C2N p‐tau181 | <0.001 | 72.61 (68.41–76.82) | 0.427 |
| C2N p‐tau181/np‐tau181 | 0.106 | – | – |
| C2N Aβ42/40 | 0.434 | – | – |
Abbreviations: Aβ, amyloid beta; CI, confidence interval; p‐tau, phosphorylated tau.
In Figure 2, we visualize all breakpoints from both the full Quanterix sample and the C2N subsample and report their estimates and 95% CI. For biomarkers present in both the full sample and the C2N subsample, we observe consistency between the estimated breakpoints when they are detected in both samples.
FIGURE 2.
Plotted breakpoints for full Quanterix sample and C2N subsample with 95% CI over age. Green: Breakpoints for biomarkers in the full Quanterix sample that are also present in the C2N subsample. Orange: Breakpoints for biomarkers in the C2N subsample that are also present in the full Quanterix sample. Red: Breakpoints for biomarkers unique to the C2N subsample. Aβ, amyloid beta; CI, confidence interval; GFAP, glial fibrillary acidic protein; NfL, neurofilament light chain; PET, positron emission tomography; p‐tau, phosphorylated tau.
In sensitivity analyses restricted to cognitively unimpaired participants, we repeated the GAM and breakpoint models in the full Quanterix cohort and the C2N subsample, with corresponding sample characteristics summarized in Table S3 in supporting information (Quanterix cohort) and Table S4 in supporting information (C2N subsample). Plasma Aβ42/40, GFAP, NfL, p‐tau181, amyloid PET, hippocampal volume, and global cognition in the full Quanterix cohort continued to exhibit breakpoints which are essentially unchanged and within the CIs of those from the original cohort (Tables S5 and S6 and Figures S3 and S4 in supporting information). The largest shift was observed for NfL, for which estimated breakpoint moved from ≈ 70.7 years (95% CI: 65.9–75.6) in the original cohort to 61.6 years (57.2–65.9) in the cognitively unimpaired subset. In the C2N subsample, breakpoints for plasma Aβ42/40, GFAP, and NfL were broadly similar between the full and cognitively unimpaired subsets, whereas Quanterix p‐tau181 and C2N p‐tau217 no longer identified breakpoints, while C2N p‐tau181 retained a breakpoint near 72.6 years (Tables S7–S10 and Figures S5 and S6 in supporting information).
4. DISCUSSION
Biomarkers of AD, neurodegeneration, and cognition are increasingly used in research settings to support risk stratification and clinical practice for diagnosis. To facilitate more personalized and timely interventions, understanding the age‐dependent dynamics of these biomarkers within a population‐based context is essential. Leveraging cross‐sectional data from a large community‐based cohort, we characterized the trajectories of plasma, imaging, and cognitive biomarkers across the adult lifespan using non‐linear modeling. Subsequently, we used breakpoint modeling to assess whether biomarkers follow distinct inflection points and to estimate their timing, guided by preliminary trajectory patterns identified through GAMs.
Compared to conventional linear regression, generalized additive and breakpoint models explained more variance in relation to age, with modest but consistent gains in adjusted R 2 across most biomarkers. Findings were consistent in both the full Quanterix sample and the C2N subsample, except for the AB42/40 ratio, which serves as internal validation and reinforces the robustness of our findings. Inflection points for plasma GFAP and NfL consistently occur at ages ≈ 68 and 70 in both samples, respectively. Similar inflection points were also identified for the two new C2N biomarkers, p‐tau217 and p‐tau181. This strong age association for NfL is consistent with prior work showing that circulating NfL concentrations increase by ≈ 2% to 3% per year of age in both men and women. 24 In a community‐based cohort such as the MCSA, age‐related increases in NfL likely reflect a combination of non‐specific axonal injury, and neurodegenerative and cerebrovascular processes in addition to the preclinical AD; therefore, the breakpoint is best interpreted as a population‐level threshold where these processes become more common rather than a specific transition point for AD. 24 , 25 , 26 The single breakpoint for global cognition at age 59.8 likely marks the age at which average cognition in this community cohort begins to decline more steeply, with implications for early cognitive monitoring, which would help inform timely cognitive monitoring and interventions. This intensified cognitive deterioration likely reflects the cumulative impact of multiple cerebral pathologies and comorbid conditions at later life stages. 27 Because these breakpoints are estimated in calendar age from cross‐sectional data, they do not map directly onto disease time, so an earlier breakpoint for cognition should not be interpreted as cognition changing before biomarker abnormalities within individuals. Segmented regression breakpoints instead indicate the ages at which the population slope of each marker changes most strongly and are not a temporal ordering of the biomarker cascade or of clinical events, which would require longitudinal data. Tau PET followed a primarily linear trend without a breakpoint, although interpretation should be cautious given the smaller tau PET subsample, which may have limited power to detect breakpoints.
Given the growing use of these biomarkers in clinical practice for risk stratification, diagnosis, and monitoring, prior studies have shown that AD has a prolonged preclinical stage, with Aβ deposition reaching a threshold of positivity ≈ 17 years before dementia onset, while hippocampal atrophy and memory impairment emerge closer to diagnosis, at ≈ 4.2 and 3.3 years prior, respectively. 28 This is consistent with our finding that Aβ42/40 shows the earliest breakpoint, followed by cognitive and then hippocampal changes. However, unlike GFAP, NfL, and p‐tau181, which showed consistent breakpoints across both the full sample and the C2N subsample, several other markers were less stable. For global cognition, amyloid PET, and hippocampal volume, breakpoints detected in the full sample were not replicated in the subsample, reflecting their dependence on younger age ranges. The discrepancy was most pronounced for Quanterix Aβ42/40, for which an early breakpoint (≈ 48 years) in the full sample shifted to a late breakpoint (≈ 88 years) in the subsample, likely due to the absence of younger participants and the weak variance in this ratio explained by age; no breakpoints were detected for C2N Aβ42/40. The absence of breakpoints for C2N p‐tau217/np‐tau217 and p‐tau181/np‐tau181 likely reflects the fact that both phosphorylated and non‐phosphorylated tau species increase with age, so the ratios are dampened relative to the individual p‐tau trajectories and show flatter age‐related trends. We observe the clustering of detected breakpoints for hippocampal volume, GFAP, NfL as well as p‐tau181 and p‐tau217 in a 5‐year interval at age ≈ 70. This suggests a critical window during which age‐related biomarker abnormalities become more pronounced, possibly marking a transition to elevated clinical risk for cognitive decline associated with AD. The timing of p‐tau inflection points in this study aligns with prior evidence suggesting tau pathology emerges after amyloid positivity but before overt neurodegeneration, reinforcing its role as a key transitional biomarker in the AD continuum. If ongoing clinical trials such as the AHEAD345 29 trial or preclinical investigations conducted by TRAILBLAZER 3 or TRAILRUNNER 30 demonstrate that early intervention targeting preclinical amyloid or tau pathology improves outcomes, the inflection patterns observed in this study may help refine the timing and population screening for these treatments.
Current results show that the identified breakpoint sequence in some biomarkers aligns with the established models of AD pathophysiology: beginning with amyloid biomarker changes (e.g., amyloid PET), followed by tau pathology (p‐tau181), and leading to neurodegeneration markers such as hippocampal volume loss and elevated NfL. 31 , 32 This ordering is consistent with findings from plasma biomarker studies. 33 While the identified inflection points reflect population‐level averages and do not capture precise temporal ordering within individuals, they nonetheless offer meaningful guidance for optimizing biomarker measurement timing and screening strategies. The breakpoint patterns suggest a broad ordering in which amyloid PET and cognition inflect earliest, followed by tau and neurodegeneration markers.
Amyloid PET and cognitive change around age 60 may signal the transition into preclinical disease, whereas GFAP and NfL breakpoints in the late 60s to early 70s appear to reflect subsequent strengthening of age‐related associations with glial and axonal injury. Within the C2N subsample, p‐tau217 and p‐tau181 converged on a common breakpoint, reinforcing their potential as key markers of emerging tau pathology. Considering that Quanterix p‐tau181 showed an earlier breakpoint in the larger full cohort, this difference underscores that absolute ages may vary by assay platform and cohort composition. The ordering of the breakpoints (Figure 2) provides guidance for distinguishing early versus later phases of disease evolution. However, the unexplained variability in the response variables and the relatively modest gains in adjusted R 2 for the GAM and breakpoint models highlight that age, while correlated, is insufficient for predicting individual‐level biomarker values. Much of the residual variation is likely attributable to underlying pathologies and medical comorbidities. 34 , 35
The sensitivity analyses restricted to cognitively unimpaired participants are also consistent with biological and statistical expectations. Removing the clinically impaired, typically older participants removes the most abnormal biomarker and cognitive values at older ages, which tends to flatten slopes over age and can shift breakpoint estimates earlier in the cognitively unimpaired subset. For NfL, for example, the earlier breakpoint in the sensitivity analysis highlights that strong age‐related changes in this marker are already evident before clinical impairment. By contrast, in the C2N subsample, age‐related increases in some tau measures appear to be more concentrated in clinically impaired individuals and the sample size is small. Exclusion of MCI participants reduces curvature and statistical power, and some breakpoints, such as those for Quanterix p‐tau181 and C2N p‐tau217, are no longer statistically supported.
Several limitations should be considered when interpreting these findings. Because the participants were primarily cognitively unimpaired or mildly impaired with only a small proportion meeting criteria for dementia, the results may not generalize to individuals in later stages of disease. Additionally, although global cognition z scores were missing for a small proportion of participants due to incomplete neuropsychological testing, we do not expect bias because it is unrelated to cognitive status or overall health. Further stratification could be performed based on family history, which may impact the ideal screening window. The absence of participants with advanced dementia and high AD neuropathological burden may attenuate or obscure associations between biomarker trajectories and age in the later phases of the disease. In addition, the demographic composition of the cohort limits generalizability, as results may not extend to population groups that were underrepresented or absent in the current sample.
Future research could address these gaps by validating biomarker trajectories in cohorts with higher proportions of later‐stage dementia, and by integrating newer AD‐specific markers to refine the mapping and staging of the disease course. Our cognitive and biomarker profiles reflect a community‐based rather than dementia clinic–referred cohort, but because of the demographics of the MCSA, the findings may not fully generalize to more diverse populations. Studies focused on early‐changing biomarkers such as amyloid PET and Aβ42/40 may inform preventive strategies aimed at delaying or mitigating disease onset. However, given that Aβ42/40 breakpoints varied across subsamples and assay platforms, further replication using alternative assays is needed to confirm the optimal timing for screening. The consistency of age‐related trajectories across the C2N and Quanterix platforms offers internal validation of the findings, although further external validation in independent, population‐representative cohorts is warranted to account for potential selection bias in the C2N subsample. Finally, emerging Bayesian changepoint methods, such as adaptive trend filtering and decoupled dynamic linear modeling, offer promising tools for identifying nuanced temporal inflections while accounting for heteroskedasticity and local outliers. These approaches could improve the inference of non‐linear trajectories by distinguishing smooth biological drift from discrete pathological shifts.
CONFLICT OF INTEREST STATEMENT
Dr. Terry Therneau reports research support from NIH paid to his institution. Dr. David S. Knopman reports research support from Alector paid to his institution. He serves on data safety monitoring boards for the Alzheimer Network Treatment Unit at Washington University in St. Louis and the University of Kentucky SMART‐HS Clinical Trial. Dr. Val J. Lowe reports research support from Siemens Medical Imaging, AVID Radiopharmaceuticals, and the NIH, paid to his institution. He has received honoraria from PeerView Institute for Medical Education and has accepted medical supplies for research purposes on behalf of his institution from AVID Radiopharmaceuticals. Dr. Ronald C. Petersen reports research support to his institution from the National Institute on Aging and the National Institute of Neurological Disorders and Stroke; he receives royalties from Oxford University Press and Up to Date; and honoraria from Medscape. He reports consulting fees from Roche, Genentech, Eli Lilly, Novo Nordisk, and Novartis; he consults for Eisai without reimbursement. Dr. Alicia Algeciras‐Schimnich serves on scientific advisory boards for Roche Diagnostics, Fujirebio Diagnostics, and Siemens Healthineers. Dr. Clifford R. Jack Jr. reports research support to his institution from the National Institute on Aging and the GHR Foundation. Dr. Nikki H. Stricker reports research support to her institution from the National Institutes of Health. Dr. Michelle M. Mielke reports research support paid to her institution from the National Institute on Aging, Alzheimer's Association, and the Department of Defense. She consulted for the following entities with payment to her institution: Acadia, Althira, Beckman Coulter, Biogen, Cognito Therapeutics, Eisai, Lilly, Merck, Neurogen Biomarking, Novo Nordisk, Roche, and Siemens Healthineers. She also reports honoraria paid directly to her from the following: Roche, Novo Nordisk, Biogen, and Medscape. Dr. Prashanthi Vemuri reports research support paid to her institution by the National Institutes of Health. Dr. Jonathan Graff‐Radford reports research funding paid to his institution from the National Institute on Aging and the GHR Foundation; he serves as a site PI for trials sponsored by Eisai and Cognition Therapeutics. He reports honoraria from IMPACT AD Faculty and the American Academy of Neurology and travel support from the Alzheimer's Association. He serves on a data safety monitoring board for StrokeNET NINDS and holds committee positions with the American Academy of Neurology and the Alzheimer's Association. The other authors report no disclosures. Author disclosures are available in the supporting information.
CONSENT STATEMENT
The study was approved by Mayo Clinic and Olmsted Medical Center Institutional Review Boards. Written informed consent was obtained from all participants.
Supporting information
Supporting Information
Supporting Information
ACKNOWLEDGMENTS
Avid Radiopharmaceuticals, Inc., a wholly owned subsidiary of Eli Lilly and Company, enabled use of the 18F‐flortaucipir tracer by providing precursor, but did not provide direct funding and was not involved in data analysis or interpretation. This work was supported by National Institute on Aging grants R01 AG34676, RF1 AG069052, U01 AG006786, P30 AG0062677, R01 AG041851, R01 AG0588738 and the GHR Foundation.
DATA AVAILABILITY STATEMENT
The MRI and other data from the Mayo Clinic Study of Aging are available to academic and industry researchers under restricted access per study and institutional review board data sharing policies. Access can be obtained by submitting a request to the MCSA and Mayo ADRC Executive Committee (https://www.mayo.edu/research/centers‐programs/alzheimers‐disease‐research‐center/research‐activities/mayo‐clinic‐study‐aging/for‐researchers/data‐sharing‐resources).
REFERENCES
- 1. Cullen NC, Leuzy A, Janelidze S, et al. Plasma biomarkers of Alzheimer's disease improve prediction of cognitive decline in cognitively unimpaired elderly populations. Nat Commun. 2021;12(1):3555. doi:10.1038/s41467‐021‐23746‐0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Janelidze S, Mattsson N, Palmqvist S, et al. Plasma P‐tau181 in Alzheimer's disease: relationship to other biomarkers, differential diagnosis, neuropathology and longitudinal progression to Alzheimer's dementia. Nat Med. 2020;26(3):379‐386. doi:10.1038/s41591‐020‐0755‐1 [DOI] [PubMed] [Google Scholar]
- 3. Tosun D, Veitch D, Aisen P, et al. Detection of β‐amyloid positivity in Alzheimer's Disease Neuroimaging Initiative participants with demographics, cognition, MRI and plasma biomarkers. Brain Commun. 2021;3(2):fcab008. doi:10.1093/braincomms/fcab008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Mielke MM, Hagen CE, Xu J, et al. Plasma phospho‐tau181 increases with Alzheimer's disease clinical severity and is associated with tau‐ and amyloid‐positron emission tomography. Alzheimers Dement. 2018;14(8):989‐997. doi:10.1016/j.jalz.2018.02.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Marks JD, Syrjanen JA, Graff‐Radford J, et al. Comparison of plasma neurofilament light and total tau as neurodegeneration markers: associations with cognitive and neuroimaging outcomes. Alzheimers Res Ther. 2021;13(1):199. doi:10.1186/s13195‐021‐00944‐y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Rafii MS, Aisen PS. Detection and treatment of Alzheimer's disease in its preclinical stage. Nat Aging. 2023;3(5):520‐531. doi:10.1038/s43587‐023‐00410‐4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Roberts RO, Geda YE, Knopman DS, et al. The Mayo Clinic Study of Aging: design and sampling, participation, baseline measures and sample characteristics. Neuroepidemiology. 2008;30(1):58‐69. doi:10.1159/000115751 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Rocca WA, Grossardt BR, Brue SM, et al. Data resource profile: expansion of the Rochester Epidemiology Project medical records‐linkage system (E‐REP). Int J Epidemiol. 2018;47(2):368‐368. doi:10.1093/ije/dyx268 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. St Sauver JL, Grossardt BR, Yawn BP, et al. Data resource profile: the Rochester Epidemiology Project (REP) medical records‐linkage system. Int J Epidemiol. 2012;41(6):1614‐1624. doi:10.1093/ije/dys195 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Vemuri P, Lesnick TG, Przybelski SA, et al. Association of lifetime intellectual enrichment with cognitive decline in the older population. JAMA Neurol. 2014;71(8):1017‐1024. doi:10.1001/jamaneurol.2014.963 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Smith GE, Ivnik RJ. Normative neuropsychology. In Mild Cognitive Impairment: Aging to Alzheimer's Disease. Oxford University Press; 2003:63‐88. [Google Scholar]
- 12. Harris ME, Ivnik RJ, Smith GE. Mayo's Older Americans Normative Studies: expanded AVLT Recognition Trial norms for ages 57 to 98. J Clin Exp Neuropsychol. 2002;24(2):214‐220. doi:10.1076/jcen.24.2.214.995 [DOI] [PubMed] [Google Scholar]
- 13. Petersen RC. Mild cognitive impairment as a diagnostic entity. J Intern Med. 2004;256(3):183‐194. doi:10.1111/j.1365‐2796.2004.01388.x [DOI] [PubMed] [Google Scholar]
- 14. Bermudez C, Graff‐Radford J, Syrjanen JA, et al. Plasma biomarkers for prediction of Alzheimer's disease neuropathologic change. Acta Neuropathologica. 2023;146(1):13‐29. doi:10.1007/s00401‐023‐02594‐w [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Jack CR, Jr. , Wiste HJ, Weigand SD, et al. Age‐specific population frequencies of cerebral β‐amyloidosis and neurodegeneration among people with normal cognitive function aged 50‐89 years: a cross‐sectional study. Lancet Neurol. 2014;13(10):997‐1005. doi:10.1016/s1474‐4422(14)70194‐2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Graff‐Radford J, Arenaza‐Urquijo EM, Knopman DS, et al. White matter hyperintensities: relationship to amyloid and tau burden. Brain. 2019;142(8):2483‐2491. doi:10.1093/brain/awz162 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Senjem ML, Gunter JL, Shiung MM, Petersen RC, Jack CR Jr. Comparison of different methodological implementations of voxel‐based morphometry in neurodegenerative disease. Neuroimage. 2005;26(2):600‐608. doi:10.1016/j.neuroimage.2005.02.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Jack CR, Jr. , Wiste HJ, Weigand SD, et al. Age‐specific and sex‐specific prevalence of cerebral beta‐amyloidosis, tauopathy, and neurodegeneration in cognitively unimpaired individuals aged 50‐95 years: a cross‐sectional study. Lancet Neurol. 2017;16(6):435‐444. doi:10.1016/S1474‐4422(17)30077‐7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Jack CR, Jr. , Wiste HJ, Weigand SD, et al. Defining imaging biomarker cut points for brain aging and Alzheimer's disease. Alzheimers Dement. 2017;13(3):205‐216. doi:10.1016/j.jalz.2016.08.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Jack CR, Jr. , Wiste HJ, Therneau TM, et al. Associations of amyloid, tau, and neurodegeneration biomarker profiles with rates of memory decline among individuals without dementia. JAMA. 2019;321(23):2316‐2325. doi:10.1001/jama.2019.7437 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Wood SN. Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models. J Roy Stat Soc B. 2011;73:3‐36. doi:10.1111/j.1467‐9868.2010.00749.x [Google Scholar]
- 22. Muggeo VM. Estimating regression models with unknown break‐points. Stat Med. 2003;22(19):3055‐3071. doi:10.1002/sim.1545 [DOI] [PubMed] [Google Scholar]
- 23. Fasola S, Muggeo VMR, Küchenhoff H. A heuristic, iterative algorithm for change‐point detection in abrupt change models. Computation Stat. 2018;33(2):997‐1015. doi:10.1007/s00180‐017‐0740‐4 [Google Scholar]
- 24. Bornhorst JA, Figdore D, Campbell MR, et al. Plasma neurofilament light chain (NfL) reference interval determination in an age‐stratified cognitively unimpaired cohort. Clin Chim Acta. 2022;535:153‐156. doi:10.1016/j.cca.2022.08.017 [DOI] [PubMed] [Google Scholar]
- 25. Kagerer SM, Awasthi S, Ripke S, et al. Polygenic risk for Alzheimer's disease is associated with neuroaxonal damage before onset of clinical symptoms. Alzheimers Dement (Amst). 2024;16(1):e12504. doi:10.1002/dad2.12504 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Raghavan S, Graff‐Radford J, Hofrenning E, et al. Plasma NfL and GFAP for predicting VCI and related brain changes in community and clinical cohorts. Alzheimers Dement. 2025;21(6):e70381. doi:10.1002/alz.70381 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Kapasi A, DeCarli C, Schneider JA. Impact of multiple pathologies on the threshold for clinically overt dementia. Acta Neuropathol. 2017;134(2):171‐186. doi:10.1007/s00401‐017‐1717‐7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Villemagne VL, Burnham S, Bourgeat P, et al. Amyloid beta deposition, neurodegeneration, and cognitive decline in sporadic Alzheimer's disease: a prospective cohort study. Lancet Neurol. 2013;12(4):357‐367. doi:10.1016/S1474‐4422(13)70044‐9 [DOI] [PubMed] [Google Scholar]
- 29. Rafii MS, Sperling RA, Donohue MC, et al. The AHEAD 3‐45 Study: design of a prevention trial for Alzheimer's disease. Alzheimers Dement. 2023;19(4):1227‐1233. doi:10.1002/alz.12748 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Yaari R, Holdridge K, Ferguson M, Wessels A, Sims J. Decentralized approaches in TRAILBLAZER‐ALZ 3 (P7‐6.006). Neurology. 2023;100(17 Suppl 2):3010. doi:10.1212/WNL.0000000000202956 [Google Scholar]
- 31. Jack CR, Jr. , Knopman DS, Jagust WJ, et al. Tracking pathophysiological processes in Alzheimer's disease: an updated hypothetical model of dynamic biomarkers. Lancet Neurol. 2013;12(2):207‐216. doi:10.1016/S1474‐4422(12)70291‐0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Bateman RJ, Xiong C, Benzinger TL, et al. Clinical and biomarker changes in dominantly inherited Alzheimer's disease. N Engl J Med. 2012;367(9):795‐804. doi:10.1056/NEJMoa1202753 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Bilgel M, An Y, Walker KA, et al. Longitudinal changes in Alzheimer's‐related plasma biomarkers and brain amyloid. Alzheimers Dement. 2023;19(10):4335‐4345. doi:10.1002/alz.13157 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Graff‐Radford J, Mielke MM, Hofrenning EI, et al. Association of plasma biomarkers of amyloid and neurodegeneration with cerebrovascular disease and Alzheimer's disease. Neurobiol Aging. 2022;119:1‐7. doi:10.1016/j.neurobiolaging.2022.07.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Mielke MM, Dage JL, Frank RD, et al. Performance of plasma phosphorylated tau 181 and 217 in the community. Nat Med. 2022;28(7):1398‐1405. doi:10.1038/s41591‐022‐01822‐2 [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
Supporting Information
Supporting Information
Data Availability Statement
The MRI and other data from the Mayo Clinic Study of Aging are available to academic and industry researchers under restricted access per study and institutional review board data sharing policies. Access can be obtained by submitting a request to the MCSA and Mayo ADRC Executive Committee (https://www.mayo.edu/research/centers‐programs/alzheimers‐disease‐research‐center/research‐activities/mayo‐clinic‐study‐aging/for‐researchers/data‐sharing‐resources).