Rates of clinical progression according to biological Alzheimer's disease stages

Abstract

INTRODUCTION

Predicting the rate of cognitive decline and the likelihood of progression to dementia remains a critical unmet need in clinical settings.

METHODS

We assessed progression to mild cognitive impairment (MCI) and all‐cause dementia in 492 individuals from the TRIAD, ADNI, and HABS‐HD cohorts followed for an average of 2.49 years. Amyloid‐positive participants were staged according to the Alzheimer's Association biological staging framework (A+T₂‐/A+T_2MTL+/A+T_2MOD+/A+T_2HIGH+).

RESULTS

Cognitively unimpaired (CU) individuals in the A+T_2MTL+, A+T_2MOD+, and A+T_2HIGH+ biological Alzheimer's disease (AD) stages were at significantly higher risk of clinical progression compared to non‐AD CU individuals. In individuals with MCI, advanced tau stage was associated with an 83% likelihood of developing dementia over 4 years. Biological AD staging demonstrated superior accuracy in predicting clinical progression compared to amyloid‐PET (positron emission tomography) status, tau‐PET status, and demographic information. All tau‐PET‐positive individuals showed a significantly faster rate of cognitive decline than non‐AD controls, with the A+T_2HIGH+ stage showing the steepest rate of decline (p < 0.001).

DISCUSSION

Our results highlight the prognostic value of biological AD staging.

Highlights

• Cognitively unimpaired (CU) individuals in all tau‐PET (positron emission tomography)–positive biological Alzheimer's disease (AD) stages were at significantly higher risk of clinical progression compared to individuals without AD.

• In individuals with mild cognitive impairment (MCI), only the A+T_2HIGH+ stage reached a point where 50% of individuals had progressed to all‐cause dementia, after 2.36 years.

• Biological AD staging demonstrated superior accuracy in predicting clinical progression to dementia compared to other PET biomarkers and demographic information.

• All tau‐PET‐positive individuals showed a significantly faster rate of cognitive decline than individuals without AD, with the A+T_2HIGH+ stage showing the steepest rate of decline.

1. BACKGROUND

Cognitive decline is a growing clinical concern among older adults. ¹ Addressing this concern by estimating likelihood and rate of more severe clinical cognitive impairment remains a significant challenge in clinical practice. ² Recently, several studies have reported that amyloid‐ and tau‐PET (positron emission tomography) biomarkers are associated with risk for clinical decline. ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² These studies have demonstrated that elevated amyloid beta (Aβ) and tau burden are linked to faster progression to mild cognitive impairment (MCI) or dementia, as well as steeper decline in longitudinal cognitive trajectories. Notably, tau‐PET measured in temporal lobe composite regions of interest has consistently shown stronger associations with cognitive and clinical outcomes than amyloid‐PET alone. Furthermore, recent studies have reported that the risk of clinical progression to dementia is related to tau pathology outside the medial temporal lobe. ⁵ , ⁹ , ¹¹

Using these biomarkers, the Alzheimer's Association recently proposed a biological staging framework for AD, in which Aβ abnormality (Aβ+) defines the presence of AD, and in which disease severity can be staged according to tau‐PET. ¹³ In this four‐stage framework, AD can be staged as (1) initial (Aβ+ in the absence of tau), (2) early (Aβ+ with restricted medial temporal lobe (MTL) tau‐PET uptake), (3) intermediate (Aβ+ and moderate neocortical tau‐PET uptake), and advanced (Aβ+ with advanced neocortical tau‐PET uptake). This new framework differs from the approach of previous reports by integrating both Aβ status and tau‐PET stage, applying tau‐PET staging to Aβ‐positive individuals only.

Understanding the prognostic value of this staging system is essential for improving risk evaluation for patients with cognitive concerns. Furthermore, the value of this biological staging system in individuals who are cognitively unimpaired (CU) versus individuals with MCI is not known. Here, we test the prognostic utility of the Alzheimer's Association biological AD staging framework in predicting clinical progression in individuals without dementia in three observational cohort studies.

2. METHODS

2.1. Participants

This study included 492 participants from three cohorts with at least one follow‐up visit: the Translational Biomarkers in Aging and Dementia (TRIAD) cohort, the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort, and the Health and Aging Brain Study‐Health disparities (HABS‐HD) cohort. TRIAD is a longitudinal cohort that was launched in 2017 as part of the McGill Centre for Studies in Aging. The TRIAD cohort recruits individuals who are CU, have MCI, or have dementia. Full information regarding the TRIAD inclusion/exclusion criteria is described elsewhere. ¹⁴ The ADNI cohort is a multicenter longitudinal cohort study launched in 2004, with tau‐PET introduced in 2015. Full inclusion/exclusion criteria of the ADNI cohort are available online. ¹⁵ The HABS‐HD cohort is a community‐based study of aging and dementia focused on the diverse recruitment of African American, Hispanic, and non‐Hispanic White participants. Inclusion/exclusion criteria for the HABS‐HD study have been described previously. ¹⁶ Exclusion criteria common to all three cohorts included recent/current diagnosis of cancer, current severe mental illness, recent traumatic brain injury with loss of consciousness, current or recent alcohol or substance abuse, or contraindications to PET or magnetic resonance imaging (MRI). Participants included in this study had amyloid‐PET, tau‐PET, and MRI assessments at baseline, in addition to clinical evaluations at baseline and at least one follow‐up time point. The TRIAD study was approved by the Montreal Neurological Institute (MNI) PET working committee and the Douglas Mental Health University Institute Research Ethics Board. The ADNI study was approved by the institutional review boards of all participating institutions. The HABS‐HD study was approved by the Institutional Review Board of University of North Texas Health Science Center. All participants or their caregivers provided informed written consent to participate in the respective studies. Due to the sparsity of follow‐up data after 4 years, data beyond the 4‐year follow‐up visit were excluded from all analyses to ensure consistency across time points.

2.2. Neuroimaging

In the TRIAD cohort, the acquisition and processing of amyloid‐PET and tau‐PET data have been described previously. ¹⁷ Briefly, [¹⁸F]AZD4694 and [¹⁸F]MK6240 PET scans were obtained using a brain‐dedicated Siemens High Resolution Research Tomograph (HRRT). [¹⁸F]AZD4694 standardized uptake volume ratio (SUVR) images were generated by normalizing to the whole cerebellum, and [¹⁸F]MK6240 SUVR images were generated by normalizing to the inferior cerebellar gray matter. In all participants, an overall composite amyloid‐PET SUVR was calculated by averaging the SUVR from the precuneus, prefrontal, orbitofrontal, parietal, temporal, and cingulate cortices. Individuals with an [¹⁸F]AZD4694 PET SUVR above 1.55 ¹⁷ were considered amyloid‐PET positive. This threshold was validated in a previous study from our group using converging analytic and biomarker techniques, including cerebrospinal fluid (CSF) Aβ42/40 measurements, Gaussian mixture modeling, and receiver‐operating characteristic (ROC) curve analyses, all of which identified 1.55 as the optimal cutoff for amyloid positivity. ¹⁷ A binary tau‐PET status was derived from tau‐PET meta‐ROI SUVR value; a previously validated threshold of 1.24 was considered tau‐PET positive. ¹⁸ This threshold was derived based on a reference group of Aβ– CU young adults (age <25). ¹⁸

In the ADNI cohort, the amyloid‐PET tracer was either [¹⁸F]florbetapir or [¹⁸F]florbetaben. Amyloid‐PET positivity in ADNI was defined using a [¹⁸F]florbetapir threshold of 1.11 or [¹⁸F]florbetaben threshold of 1.08 SUVR. Amyloid‐PET SUVR positivity thresholds were established by the ADNI PET Core, ¹⁹ and they have been applied consistently across ADNI studies. Briefly, the [¹⁸F]florbetapir SUVR cutoff of 1.11 was derived from the upper 95% confidence interval (CI) of mean cortical [¹⁸F]florbetapir SUVR in young healthy controls, and aligns with post‐mortem findings showing no AD pathology below this threshold. ²⁰ , ²¹ , ²² The [¹⁸F]florbetaben SUVR threshold of 1.08 was derived using Gaussian mixture modeling in an independent sample of young controls, consistent with the [¹⁸F]florbetapir cutoff approach, to ensure methodological congruence across tracers. ²³ Tau‐PET was performed using [¹⁸F]flortaucipir. [¹⁸F]flortaucipir SUVR images were generated by normalizing to the inferior cerebellar gray matter. Details of imaging acquisition parameters, as well as imaging processing information, have been described previously (see ADNI PET methods available online ¹⁵ ). A binary tau‐PET status was derived from tau‐PET meta‐ROI SUVR value; we used a cutoff of 1.32, which was obtained previously with ROC curves maximizing classification accuracy between amyloid‐negative cognitively unimpaired (Aβ− CU) and amyloid‐positive older adults with MCI (Aβ+ MCI). ²⁴ , ²⁵ A similar threshold was independently derived in our sample by calculating two standard deviations (SDs) above the mean SUVR of amyloid‐negative CU individuals.

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed the literature (PubMed and Google scholar) using traditional sources. Several recent observational studies have reported that tau‐PET (positron emission tomography) uptake conveys prognostically meaningful information. Recently, the Alzheimer's Association workgroup proposed a biological Alzheimer's disease (AD) staging system. Relevant papers are cited in the manuscript.

2. Interpretation: In asymptomatic individuals, all AD stages were associated with higher risk of progression as compared to the amyloid‐negative group. In individuals with mild cognitive impairment, advanced tau stage was associated with an 83% likelihood of developing dementia over 4 years. Biological staging was more accurate at predicting clinical progression than dichotomous biomarkers and demographic information.

3. Future directions: Future studies with longer follow‐up durations are needed to better characterize the natural history of AD‐associated clinical decline. Furthermore, studies evaluating the accuracy of staging models using more accessible biomarkers are needed.

In the HABS‐HD study, amyloid‐PET was acquired using [¹⁸F]florbetaben, and tau‐PET was acquired using [¹⁸F]PI2620 normalized to the inferior cerebellar gray matter. [¹⁸F]florbetaben amyloid‐PET scans were considered positive if the SUVR (normalized by the whole cerebellum) exceeded a previously validated threshold of 1.08. ²⁶ This cutoff is consistent with the ADNI PET Core's [¹⁸F]florbetaben SUVR threshold. ¹⁵ Details of imaging acquisition parameters, as well as imaging processing information, have been described previously for the HABS‐HD cohort. ²⁶ A binary tau‐PET status was derived from tau‐PET meta‐ROI SUVR value. No previously validated tau‐PET positivity threshold using [¹⁸F]PI2620 has been established for the temporal meta‐ROI. Therefore, we derived a cutoff of 1.33 using the mean +2 SD in Aβ− CU older adults. This approach is consistent with the methods applied previously for [¹⁸F]PI2620. ²⁷ Because tau‐PET staging methods used the mean +2.5 SD from the Aβ− CU older adult group, a sensitivity analysis was conducted employing a threshold of 1.38 SUVR, which was derived based on 2.5 SD from the Aβ− CU older adult group. 

FIGURE 1.

Schematic representation of biological AD staging. Aβ+ individuals were classified as A+T₂– (initial stage) if they were tau‐PET negative, A+T_2MTL+ (early stage) if they were positive in Braak I–II regions only, A+T_2MOD+ (intermediate stage) if they were positive in Braak III–IV regions, and A+T_2HIGH+ (advanced stage) if they were positive in Braak V–VI regions.

2.3. Clinical assessment

To classify participants as CU or with MCI, we used the Clinical Dementia Rating (CDR) scale, which was available in all datasets. The CDR was used at each time point to assess conversion to MCI or dementia. Individuals with a CDR score of 0 were categorized as CU, indicating no clinically significant cognitive impairment. Individuals with a CDR score of 0.5 were categorized as having MCI. Participants with dementia were identified based on a CDR score of 1 or 2. This classification scheme is consistent with established criteria for diagnosing MCI and dementia in clinical and research settings. ²⁸

2.4. Thresholding methods for biological AD staging

Biological AD stages were operationalized using tau‐PET Braak staging methods, as described previously, ²⁹ , ³⁰ , ³¹ which represents a slight variation from the Alzheimer's Association staging criteria because two different ROIs are employed to differentiate between the T_2MOD+ and T_2HIGH+ stages. Two methods were used to determine tau‐PET stage: (1) a common approach to all cohorts and radiotracers, based on the mean +2.5 SD of CU Aβ− individuals younger than age 70, and (2) previously validated cohort‐ and tracer‐specific thresholds (Figure S1). The previously validated thresholding methods in the TRIAD cohort using the [¹⁸F]MK‐6240 tau‐PET positivity were defined as SUVRs greater than 2.5 SD above the mean SUVR of young individuals for each Braak ROI. ³² , ³³ In the HABS‐HD cohort, the previously validated [¹⁸F]PI‐2620 tau‐PET used SUVRs greater than the mean +2.5 SD of the Aβ− CU older adults, which has shown a theoretical false‐positive rate of ≈1.2%. ²⁷ In the ADNI cohort, the previously validated [¹⁸F]Flortaucipir cutoffs were defined using SUVRs of 2 SD above the CU Aβ− older adults mean for each ROI, based on a published systematic review of empirical thresholding approaches using [¹⁸F]Flortaucipir. ²⁴ Stage‐specific threshold values and previously validated methods for all tracers and cohorts are provided in Table S1.

Biological AD severity was staged using the Alzheimer's Association biological framework. ¹³ More precisely, participants were first classified as Aβ− or Aβ+ based on their Aβ status. Aβ− individuals were included as individuals without AD (irrespective of their clinical stage). In all amyloid‐PET‐positive participants, we first staged tau‐PET using the PET‐based Braak staging framework described above. ³⁰ Aβ+ individuals were classified as T₂‐ (initial stage) if they were tau‐PET negative, T_2MTL+ (early stage) if they were positive in Braak I–II regions, T_2MOD+ (intermediate stage) if they were positive in Braak III–IV regions, and T_2HIGH+ (advanced stage) if they were positive in Braak V–VI regions (Figure 1). We observed a very small proportion of individuals (≈5%) exhibiting a “discordant” tau‐PET pattern (i.e., individuals who were positive in more advanced regions, but not early‐stage regions), similar to other reports. ⁹ Previous work from our group has shown that when participants are tau‐PET “discordant,” they are often near the threshold of abnormality in the early regions and are therefore unlikely to be truly negative. ³⁰ , ³² Thus, for these participants, we assigned a stage based on the latest regions showing tau‐PET abnormality.

2.5. Statistical methods

Statistical analyses were performed using R v.2024.04.2. Kaplan–Meier survival analyses were performed to estimate the time of progression to MCI and all‐cause dementia. Participants were stratified by biological AD stage at baseline, and survival curves were compared using the log‐rank test to evaluate differences in progression rates across groups. Cox proportional hazards models were used to examine the association between biological stage at baseline and the risk of progression to MCI or dementia over time. The models included adjustments for age, sex,  education, and cohort. Adjusting for cohort in our analyses ensured that there were no cohort‐specific effects, which could be due to the differences in demographic characteristics. Using the R package lmerTest, we performed linear mixed‐effects models to assess the relationship between changes in Mini‐Mental State Examination (MMSE) score and CDR Sum of Boxes (CDR‐SoB) and biological AD staging, including fixed effects for time, biological AD stage, and their interaction, and covariates such as age, sex, years of education, cohort, and diagnosis at baseline visit. Random intercepts and slopes were included at the participant level to account for individual differences in baseline cognitive function and rates of change over time. Using the car package in R, collinearity was assessed for all variables in the models by calculating the variance inflation factor (VIF). Area under the ROC curve (AUROC) analyses were performed to assess the ability of baseline biological AD staging to predict progression to the next clinical stage in CU and MCI individuals. To do so, we fitted Cox proportional hazards models to estimate the risk of clinical progression and extracted the linear predictors from each model. Models were adjusted for age, sex, and years of education to account for potential confounders. These risk scores were then used to compute AUROC values, reflecting the discriminative accuracy of each biomarker. ROC curves were generated using the pROC package, and Delong's tests were used to statistically compare AUROCs. Sensitivity analyses were conducted to assess the influence of different thresholding methods on the main results (i.e., thresholds based on mean +2.5 SD from the mean of CU Aβ− adults younger than age 70 years or using previously validated thresholds in each cohort for each tau‐PET imaging agent). Similarly, sensitivity analyses were conducted using a binary tau‐PET thresholding method similar to the method used for tau‐PET staging. Finally, we conducted sensitivity analyses excluding the ≈5% of subjects with “discordant” tau‐PET scans (i.e., evidence of tau in regions comprising a later stage but with only subthreshold tau deposition in earlier stage regions, who often have atypical clinical presentations ³⁴ ).

TABLE 1.

Demographics of all sample.

	Aβ–	A+T2–	A+T2MTL+	A+T2MOD+	A+T2HIGH+	Overall
All cohorts
No.	277	85	49	29	52	492
Age, mean (SD), y	69.3 (8.43)	72.6 (7.84) a	74.5 (6.12) a , e	72.0 (6.25)	68.6 (8.98)	70.4 (8.28)
Male, no. (%)	125 (45.1%)	34 (40.0%)	18 (40.9%)	12 (40.0%)	25 (44.6%)	214 (43.5%)
Education, mean (SD), y	15.9 (3.35)	15.3 (3.33)	15.1 (3.36)	15.8 (3.67)	15.1 (2.77)	15.6 (3.31)
MMSE, mean (SD)	28.9 (1.59) c , d , e	28.3 (2.38) e	27.4 (2.83)	27.1 (2.94)	25.5 (4.00)	28.2 (2.57)
	228 (43/6) b , c , d , e	65 (13/7)	23 (24/2)	10 (16/3)	8 (34/10)	334 (130/28)
Follow‐up time, mean (SD), y	2.43 (0.895)	2.63 (0.955)	2.51 (0.999)	2.55 (0.999)	2.47 (0.811)	2.49 (0.912)

Abbreviations: CU, cognitively unimpaired; MCI, mild cognitive impairment; MMSE, Mini‐Mental State Examination; SD, standard deviation.

^a Significantly higher than Aβ–

^b Significantly higher than A+T₂–

^c Significantly higher than A+T_2MTL+

^d Significantly higher than A+T_2MOD+

^e Significantly higher than A+T_2HIGH+

Data availability

Data from the ADNI and HABS‐HD cohorts can be accessed online. ³⁵ Requests for analyzed TRIAD cohort data and materials will be reviewed promptly by McGill University to determine whether they are subject to any intellectual property or confidentiality obligations. The study's senior author will provide anonymized TRIAD data upon request from a qualified academic investigator for the sole purpose of replicating the procedures and results presented in this article. A material transfer agreement will be used to release any data and materials that can be shared.

3. RESULTS

3.1. Participants

We included 155 participants from the TRIAD cohort (average follow‐up time: 2.13 years ± 0.850), 244 participants from the ADNI cohort (average follow‐up time: 2.84 years ± 0.958), and 93 participants from the HABS‐HD cohort (average follow‐up time: 2.14 years ± 0.468) with at least one follow‐up visit. For the HABS‐HD cohort, the data freeze 5 was used, including non‐Hispanic Whites and Mexican Americans. Demographic characteristics of all participants are listed in Table 1. Group differences in categorical variables (i.e., clinical status and sex) were assessed using chi‐square [χ ²] tests, whereas continuous demographic variables were compared using Kruskal−Wallis tests followed by Dunn's post hoc tests with Bonferroni correction. Demographic characteristics of individuals stratified by cohort and stratified according to previously validated cohort‐specific methods are summarized in Tables S2–S4. Due to the small number of participants in some subgroups, statistical comparisons were not performed for demographic variables stratified by cohort. In total, 277 individuals were classified individuals without AD (Aβ−), 85 individuals were in the A+T₂‐ stage, 49 individuals were in the A+T_2MTL+ stage, 29 individuals were in the A+T_2MOD+ stage, and 52 individuals were in the A+T_2HIGH+ stage.

3.2. Progression to MCI in CU individuals

Over the follow‐up period, 5 of 228 (2.2%) CU individuals without AD (Aβ−), 5 of 65 (7.7%) A+T₂‐ individuals, 7 of 23 (30.4%) A+T_2MTL+, 6 of 10 (60%) A+T_2MOD+, and 8 of 8 (100%) A+T_2HIGH+ converted to MCI (Figure 2A). Pairwise log‐rank tests indicated that although all groups significantly differed from the reference group (i.e., non‐AD CU individuals), only the A+T_2HIGH+ stage was found to be significantly different from the A+T₂‐ group; no statistically significant differences were found between other stages (see Table 2 for p‐values). All stages except the Aβ− and A+T₂‐ stages reached a point at which 50% of individuals within a stage had progressed to MCI (3.84 years for A+T_2MTL+, 3.91 years for A+T_2MOD+, and 2.48 years for A+T_2HIGH+). Risks of progression to MCI in percentage for each follow‐up year according to biological stage are listed in Table 3. In CU individuals, a Cox proportional‐hazards model correcting for age, sex, years of education, and cohort showed an increased risk of progression to MCI for all biological AD stages compared to the reference group (i.e., non‐AD CU individuals), whereas age, sex, and years of education were not significantly associated with risk of progression (Figure 2A). Sensitivity analyses using different thresholding methods provided nearly identical results (Tables S5,S6 and Figure S2A). Sensitivity analyses using excluding tau‐PET stage “discordant” individuals also yielded nearly identical results (Table S7 and Figure S3A).

FIGURE 2.

Progression to MCI or to all‐cause dementia according to biological stages. (A) Survival curves and table showing progression to MCI in CU individuals according to baseline biological AD stage. The dashed line indicates the median survival time for each biomarker group, that is, the point at which 50% of individuals had converted to all‐cause dementia. Below, HRs and 95% CIs for progression from CU to MCI are displayed. (B) Survival curves and table showing progression to all‐cause dementia in MCI individuals according to baseline biological AD stage. The dashed line indicates the median time, that is, the point at which 50% of individuals had converted to all‐cause dementia. Below, HRs and 95% CIs for progression from MCI to all‐cause dementia are displayed. CI, confidence interval; CU, cognitively unimpaired; HR, hazard ratio; MCI, mild cognitive impairment.

TABLE 2.

Pairwise log‐rank tests for comparison between biological AD stages in CU individuals (left) and in MCI individuals (right).

CU progression to MCI					MCI progression to all‐cause dementia
	Aβ–	A+T2–	A+T2MTL+	A+T2MOD+		Aβ–	A+T2–	A+T2MTL+	A+T2MOD+
A+T2 –	0.1052	‐	‐	‐	A+T2‐	0.628	‐	‐	‐
A+T2MTL+	1.3e‐09	0.0071	‐	‐	A+T2MTL+	0.0458	0.141	‐	‐
A+T2MOD+	7.9e‐14	0.0002	0.680	‐	A+T2MOD+	0.0067	0.102	0.628	‐
A+T2HIGH+	2e‐16	1.7e‐06	0.168	0.493	A+T2HIGH+	3.6e‐07	0.0067	0.0067	0.0745

Abbreviations: AD, Alzheimer's disease; CU, cognitively unimpaired; MCI, mild cognitive impairment.

TABLE 3.

Risk of conversion to MCI and all‐cause dementia.

Risk of conversion to MCI (%)					Risk of conversion to all‐cause dementia (%)
	1 year	2 years	3 years	4 years		1 year	2 years	3 years	4 years
Aβ–	0	1	3.51	6.96	Aβ‐	0	2.9	2.9	2.9
A+T2 –	0	5.6	11.7	11.5	A+T2‐	0	0	0	0
A+T2MTL+	0	5.8	19.2	72.7	A+T2MTL+	0	0	34.7	34.7
A+T2MOD+	0	10	48.6	82.9	A+T2MOD+	0	16.6	16.6	44.7
A+T2HIGH+	0	12.5	62.5	100	A+T2HIGH+	0	31.2	66.2	83.1

Abbreviations: MCI, mild cognitive impairment.

3.3. Progression to all‐cause dementia in individuals with MCI

Over the follow‐up period, 1 of 43 (2.33%) non‐AD patients with MCI (Aβ–), 0 of 13 (0%) A+T₂‐, 4 of 24 (16.6%) A+T_2MTL+, 5 of 16 (31.3%) A+T_2MOD+, and 17 of 34 (50.0%) A+T_2HIGH+ converted to all‐cause dementia (Figure 2B). Pairwise log‐rank tests showed no significant difference between the A+T₂‐ and reference group but revealed that the A+T_2HIGH+ stage had a significantly higher risk of conversion to dementia compared to all other stages (see Table 2 for p‐values). Only the A+T_2HIGH+ stage reached a point at which 50% of individuals within the stage had progressed to all‐cause dementia, after 2.36 years. Risks of progression to MCI in percentage for each follow‐up year according to biological stage are listed in Table 3. The Cox proportional‐hazards model in MCI individuals required collapsing of Aβ− (i.e., non‐AD MCI individuals) and A+T₂‐ groups due to insufficient events in this stage, thereby limiting statistical power to estimate hazard ratios accurately. Correcting for age, sex, years of education, and cohort, this model revealed an increased risk of progression to AD for A+T_2MTL+, A+T_2MOD+, and A+T_2HIGH+ stages (Figure 2B). Age, sex, and years of education were not significantly associated with risk of progression. Sensitivity analyses using different thresholding methods provided nearly identical results (Tables S5,S6 and Figure S2B). Sensitivity analyses excluding tau‐PET stage “discordant” individuals also yielded nearly identical results (Table S7 and Figure S3B).

3.4. Comparison of biomarker stratification systems to predict MCI and dementia

In CU individuals, we observed that biological AD staging accurately identified clinical progression to MCI (AUROC: 0.861, 95% CI: 0.78–0.95). Similarly, binarized tau‐PET status also predicted clinical progression to MCI (AUROC: 0.804, 95% CI: 0.71–0.90). Amyloid‐PET status had lower discriminative accuracy to predict development of MCI (AUROC: 0.78, 95% CI: 0.69–0.87). In contrast, demographic information had poor discriminative ability to identify individuals who would progress to MCI (AUROC: 0.605, 95% CI: 0.50–0.71) (Figure 3A). Delong's tests revealed that biological AD staging had significantly higher performance in predicting clinical progression to MCI over demographic information (p < 0.0001) and amyloid‐PET status (p = 0.0036), but not compared to tau‐PET status (p = 0.11). Tau‐PET biological staging methods using fewer stages showed lower discriminative accuracy compared to biological AD stages to predict progression to MCI (AUROC: 0.805, 95% CI: 0.73–0.89). Delong's tests revealed that this three‐stage method had a significantly higher performance in predicting progression to MCI than amyloid‐PET status (p = 0.016) and demographic information (p < 0.0001), but had statistically similar performance to tau‐PET binary status (p = 0.10) and biological AD stages (p = 0.11). P‐values from Delong's tests are reported in Table 4. Sensitivity analyses using different thresholding methods yielded nearly identical results (Table S8 and Figure S4A). Sensitivity analyses using excluding tau‐PET stage “discordant” individuals also provided nearly identical results (Table S9 and Figure S5A).

FIGURE 3.

AUROC analyses to identify CU and MCI individuals who will progress to the next clinical dementia stage. All models were corrected for age, sex, and years of education. AUROC, area under the receiver‐operating characteristic curve; CU, cognitively unimpaired; MCI, mild cognitive impairment.

TABLE 4.

Pairwise comparisons of the area under the ROC curves (AUROCs) using DeLong's test.

CU progression to MCI		MCI conversion to all‐cause dementia
Comparison	p‐value	Comparison	p‐value
Demographics versus Aβ‐PET status	0.0002	Demographics versus Aβ‐PET status	0.0013
Demographics versus tau‐PET status	0.0030	Demographics versus tau‐PET status	0.0063
Demographics versus biological AD stage	8.29e‐06	Demographics versus biological AD stage	4.53e‐05
Aβ‐PET status versus tau‐PET status	0.5725	Aβ‐PET status versus tau‐PET status	0.6278
Aβ‐PET status versus biological AD stage	0.0036	Aβ‐PET status versus biological AD stage	0.0114
Tau‐PET status versus biological AD stage	0.0134	Tau‐PET status versus biological AD stage	0.0294
Demographics versus 3‐stage tau‐PET status*	2.96e‐05	Demographics versus 3‐stage tau‐PET status*	0.0987
Aβ‐PET status versus 3‐stage tau‐PET status*	0.0167	Aβ‐PET status versus 3‐stage tau‐PET status*	0.0353
Tau‐PET status versus 3‐stage tau‐PET status*	0.1006	Tau‐PET status versus 3‐stage tau‐PET status*	0.0771
Biological AD stage (AA framework) versus 3‐stage tau‐PET status*	0.1160	Biological AD stage (AA framework) versus 3‐stage tau‐PET status*	0.0004

Abbreviations: Aβ, amyloid beta; AA, Alzheimer's Association; CU, cognitively unimpaired; MCI, mild cognitive impairment; PET, positron emission tomography.

Each p‐value reflects whether the discriminative accuracy of one model differs significantly from another in predicting clinical progression from MCI to AD.

p <  0.05 (in bold) indicates a statistically significant difference.

In individuals with MCI, we observed that the biological AD staging system accurately identified clinical progression to dementia (AUROC: 0.875, 95% CI: 0.80–0.95). Binarized tau‐PET status also had good discriminative accuracy to predict dementia over 4 years (AUROC: 0.814, 95% CI: 0.73–0.91). Amyloid‐PET status had a more modest ability to predict clinical progression to dementia (AUROC: 0.791, 95% CI: 0.70–0.89). Demographic information had poor discriminative ability to identify individuals who would progress to dementia (AUROC: 0.678, 95% CI: 0.55–0.78) (Figure 3B). Delong's tests revealed that biological AD staging had significantly higher performance in predicting clinical progression to dementia over amyloid‐PET status (p = 0.01), tau‐PET status (p = 0.03), and demographic information (p = 0.0001). Tau‐PET biological staging methods using three stages showed a modest ability to predict progression to dementia (AUROC: 0.734, 95% CI: 0.63–0.84). Delong's tests revealed that this method still performed better than amyloid‐PET status and demographic information in predicting progression to dementia, but was significantly outperformed by the four‐stage biological AD staging framework (p = 0.0004). P‐values from Delong's tests are reported in Table 4. Sensitivity analyses using different thresholding methods provided nearly identical results (Table S10 and Figure S4B). Sensitivity analyses using excluding tau‐PET stage “discordant” individuals also provided nearly identical results (Table S11 and Figure S5B).

3.5. Cognitive trajectories

We employed linear mixed‐effects models controlling for age, sex, years of education, and diagnosis at baseline visit to assess the effect of biological AD stage at baseline and the interaction with visit year on cognition, as measured by the CDR‐SoB and the MMSE. A VIF smaller than 1.4 for both models suggested no inflation of coefficients due to collinearity. Using the CDR‐SoB, the interaction of biological AD stage with time in years showed that individuals in the A+T_2MTL+ (β _STD = 0.21, 95% CI: 0.06–0.37; p = 0.01), A+T_2MOD+ (β _STD = 0.28, 95% CI: 0.15–0.42; p = 0.004), and A+T_2HIGH+ (β _STD = 1.11, 95% CI: 0.97–1.25; p < 0.0001) stages experienced significantly faster cognitive decline over time compared to the reference group (Aβ−), whereas the A+T₂‐ stage (β _STD = 0.04, 95% CI: −0.09 to 0.16; p = 0.55) did not differ significantly from the reference group (Figure 4A). Using the MMSE, the interaction of biological AD stage with time indicated that only individuals in the A+T_2MOD+ (β _STD = −0.37, 95% CI: −0.69 to −0.07; p = 0.03) and A+T_2HIGH+ (β _STD = −1.65, 95% CI: −1.93 to −1.37; p < 0.0001) stages experienced significantly faster cognitive decline compared to the reference group (Aβ−), whereas the A+T₂‐ (β _STD = 0.05, 95% CI: −0.04 to 0.14; p = 0.55) and A+T_2MTL+ (β _STD = −0.14, 95% CI: −0.26 to 0.13; p = 0.32) and the stages did not differ significantly (Figure 4B). Sensitivity analyses using different thresholding methods provided nearly identical results (Figure S6). Sensitivity analyses using excluding tau‐PET stage “discordant” individuals also provided nearly identical results (Figure S7).

FIGURE 4.

Cognitive decline over time according to baseline biological AD stage. Cognitive trajectories of CDR‐SoB (A) and MMSE (B) over time according to baseline biological AD stage.

4. DISCUSSION

This study investigated the patterns of clinical progression to MCI and to all‐cause dementia according to biological AD stage in three independent cohorts with distinct demographic characteristics. We observed that the risk of clinical progression from CU to MCI was high for individuals at the A+T_2MTL+, A+T_2MOD+, or A+T_2HIGH+ stages of AD. Furthermore, we observed that clinical progression from MCI to all‐cause dementia was significantly higher for individuals at advanced tau‐PET stages, supporting the notion that advanced AD pathology is strongly associated with clinical dementia. Taken together, our findings suggest that biological AD staging is useful for predicting clinical progression, and confers more prognostic information than dichotomized tau‐PET status in individuals with MCI.

MCI is a clinical syndrome that is highly heterogeneous in terms of etiology and prognosis. ³⁶ We observed that in individuals with MCI, the risk of developing all‐cause dementia over 4 years differed significantly according to baseline biological AD stage. Specifically, the risk of developing dementia over the study period was significantly elevated in the A+T_2HIGH+ group as compared to the A+T₂‐ and A+T_2MTL+ stages, but only trended toward significance when compared to the A+T_2MOD+ stage. Nonetheless, the biological AD staging approach also outperformed classification methods using fewer stages (i.e., binary or three‐stage methods) for predicting progression to all‐cause dementia, underscoring its additional prognostic value. These results highlight that the clinical dementia phase of AD is strongly associated with advanced AD pathology. ³⁰ , ³⁷ Finally, these results support the recently published amyloid‐PET and tau‐PET appropriate use criteria, which deemed using tau‐PET to predict prognosis in individuals with MCI as clinically appropriate. ³⁸

In CU individuals, all tau‐positive AD stages were predictive of clinical progression to MCI compared to CU individuals without AD. Specifically, a baseline biological AD stage of A+T_2MTL+, A+T_2MOD+, or A+T_2HIGH+ was associated with a risk of over 50% for developing MCI over 4 years. In our study, increasing biological AD stages was associated with a numerically higher risk of clinical progression to MCI, although some between‐stage comparisons were not statistically significant. Despite including data from three cohort studies, only a small number of CU individuals were at the A+T_2MTL+, A+T_2MOD+, or A+T_2HIGH+ stages. The close relationship between tau pathology and cognitive impairment ¹³ , ³⁰ , ³⁹ , ⁴⁰ , ⁴¹ explains the low prevalence of these biological AD stages in asymptomatic individuals, and this may have decreased the statistical power of our study to detect between‐stage differences in CU individuals. However, a recent multicenter longitudinal study of over 1300 individuals found that the progression from CU to MCI was similar in individuals who had tau‐PET in the medial temporal lobe (in our study classified as A+T_2MTL+) and individuals with tau‐PET signal in the neocortex (in our study classified either as A+T_2MOD+ or A+T_2HIGH+), ⁵ which is congruent with our findings.

In the same line of thought, we also observed that the few CU individuals exhibiting tau pathology declined more rapidly, which may reflect a state of clinical resilience at baseline. These individuals may represent a resilient subgroup who have maintained intact cognition despite substantial pathology, ⁴² but for whom clinical decline may accelerate as the disease process begins to catch up. In comparison, we observed numerically slower rates of conversion from MCI to dementia in individuals with early and intermediate tau pathology, which may be explained by the fact that dementia is typically present in individuals with advanced tau pathology. ³⁰ , ³² Furthermore, given the typically slow progression of AD, their transition to dementia may unfold over a longer period than the 4‐year follow‐up period of this study. Our results contribute to recent studies evaluating the correspondence between clinical and biological stages as proposed by the AA workgroup, ⁴³ , ⁴⁴ and emphasize the importance of identifying clinical–biological mismatch to better anticipate short‐term clinical progression.

The results of our study must be interpreted in the context of the detection thresholds of PET imaging, which may differ among tau imaging agents, ⁴⁵ and in the context of non‐AD neurodegenerative diseases, which also result in dementia. Emerging studies with [¹⁸F]flortaucipir suggest that tau‐PET scans reliably detect advanced neurofibrillary tangle pathology, ⁴⁶ but their ability to detect early tau tangles in Braak I–II regions is limited. ⁴⁷ , ⁴⁸ Moreover, neuropathological studies report that 74%–87% of asymptomatic individuals with moderate/frequent CERAD neuritic plaque scores (which correspond to a nearly 100% probability of being amyloid‐PET positive) ⁴⁹ had Braak Stage III or higher neurofibrillary tangle stage at autopsy. ¹³ Therefore, it is likely that amyloid‐PET‐positive individuals have some degree of tau tangles, which may contribute to the development of cognitive symptoms over the study period. Furthermore, it is possible that brain β‐amyloidosis also predisposes individuals to non‐AD pathologies, which contribute to their cognitive decline. ⁵⁰ For example, significant correlations have been observed between brain β‐amyloidosis and severity of α‐synuclein pathology in individuals with dementia with Lewy bodies. ⁵¹ Overall, a greater understanding of the risks for clinical progression associated with brain β‐amyloidosis, as well as identification of risk factors and biomarkers associated with non‐AD pathologies, is needed.

By employing PET‐based Braak staging, our study applied a variation of the Alzheimer's Association revised framework for AD. The Alzheimer's Association criteria framework describes that the distinction between the T_2MOD+ or T_2HIGH+ stages can be operationalized in different ways and that the best method to differentiate between T_2MOD+ and T_2HIGH+ stages is yet to be determined. A recent study that investigated the 2024 Alzheimer's Association staging framework also used different ROIs to stage individuals as T_2MOD+ (temporal lobe ROI) or T_2HIGH+ (temporal lobe & Mubada ROIs). ⁴⁴ These results were highly comparable to another recent study investigating the 2024 Alzheimer's Association staging framework, which used the same neocortical ROI to define T_2MOD+ and T_2HIGH+ stages. ⁴³ Nonetheless, it is important to interpret the findings of the present study in the context of the methodology employed to stage participants, as other staging techniques may perform better or worse at predicting clinical progression.

Some biomarker profiles were not explicitly investigated in our study. For example, all amyloid‐PET negative individuals in this study were categorized as individuals without AD, irrespective of their tau‐PET status. However, a small number of amyloid‐PET‐negative individuals display detectable tau‐PET signal, ¹⁸ , ³⁰ , ⁵² most often in medial temporal cortices. ⁵³ , ⁵⁴ Nonetheless, studies have shown that individuals with this biomarker profile do not display higher rates of cognitive decline than controls. ⁵ , ⁵³ Thus, the influence of this biomarker category on our results is likely minimal.

Findings from this study suggest that tau‐PET staging of AD in vivo provides important information about the risk of clinical decline in selected populations. If replicated in community‐based settings with less stringent inclusion/exclusion criteria (i.e., that will include more individuals likely to develop non‐AD diseases), our results highlight the need for more accessible methods to estimate risk of clinical decline. ³⁷ Recently, fluid biomarker‐based staging systems have been proposed, ¹³ , ³⁷ and have been validated in CSF, ⁵⁵ although the correspondence between fluid and PET‐based staging requires further study. Moreover, because recent studies suggest that plasma biomarkers can be used to predict amyloid‐PET and tau‐PET stages, ⁵⁶ , ⁵⁷ , ⁵⁸ , ⁵⁹ further research could lead to plasma‐based biological AD staging profiles. Furthermore, emerging data also suggest that stratifying individuals based on plasma p‐tau217 concentrations has prognostic utility at the group level, ⁶⁰ , ⁶¹ but that caution is needed when applying these data at the individual level, especially in CU individuals because of the risk of false positives. ⁶² Taken together, our study highlights the potential clinical value of AD staging system, while emphasizing the need for more accessible methods with prognostic information comparable to tau‐PET‐based biological staging.

Our study should be interpreted in the context of its methodological limitations. First, AD is a disease that is estimated to evolve over ≈20 years, and the follow‐up period of our study is short by comparison. Longer studies are needed to better elucidate the natural history of clinical progression rates according to biological AD stages. Second, the existence of four biological AD stages, in addition to the control group, may have resulted in modest statistical power to detect between‐stage differences in clinical progression and clinical decline rates. Third, the detection threshold of PET imaging is higher than neuropathological assessments, and a better understanding of the correspondence between in vivo PET stage and postmortem AD stage is needed. Fourth, neuropathological studies have demonstrated that other proteinopathies such as TAR DNA‐binding protein 43 (TDP‐43) and α‐synuclein contribute to cognitive decline in participants of the age range in our study. However, lacking biomarkers to quantify these processes, we were unable to assess their influence on clinical decline. Fifth, the current study operationalized the Alzheimer's Association staging framework according to PET‐based Braak staging methods, which have been employed widely by our group and others. ²⁹ , ³³ , ³⁹ , ⁵⁹ , ⁶³ However, it is important to note that PET‐based Braak staging may be considered as a variation of the Alzheimer's Association staging framework because it employs two different ROIs to define T_2MOD+ and T_2HIGH+ stages. It is possible that other tau‐PET staging methods may have stronger or weaker associations with clinical progression. Finally, participants in this study consisted of self‐motivated individuals who were interested in participating in a study on aging and AD and hence may not be demographically representative of the general population.

In summary, staging individuals with cognitive concerns based on tau‐PET uptake can provide patients with important information about the natural history of their disease (i.e., their prognosis in the absence of disease‐modifying intervention). ⁶⁴ This may allow for facilitation of personalized medicine approaches, which will help individuals understand how they may potentially benefit from a therapy given their disease severity or stage. Finally, understanding progression‐free survival time and likelihood of conversion to dementia can aid patients and their families in making informed decisions regarding care strategies and future planning.

CONFLICT OF INTEREST STATEMENT

L. Trudel reports no disclosures relevant to the manuscript. J. Therriault has served as a consultant for the Neurotorium Educational Platform and as a medical writer for Alzheon Inc, both outside of the scope of the present work. A.C. Macedo, S. Servaes, S.A. Hosseini, G. Bezgin, N. Rahmouni, T. Chan, J. Fernandez‐Arias, E. Aumont, Y.T. Wang, Y. Zheng, B. Hall, R. Hopewell, C.H. Hsiao, A.W. Toga, M.N. Braskie, K.L. Meeker, and J.P. Soucy report no disclosures relevant to the manuscript. S. Gauthier serves on scientific advisory boards for Alzeon, AmyriAD, Advantage, Eisai Canada, Enigma USA, Lilly Canada, Medesis, Lundbeck Foundation, Novo‐Nordisk Canada, Okutsa, and TauRx. H. P. Vitali serves on the scientific advisory boards for NovoNordisc, Eisai, and Lilly, and received honoraria from IntelGenx Corp. S.E. O'Bryansst and T.A. Pascoal report no disclosures relevant to the manuscript. P. Rosa‐Neto has served on scientific advisory boards and/or as a consultant for Roche, Novo Nordisk, Eisai, and Cerveau Technologies, outside of the scope of the present work. Any author disclosures are available in the Supporting Information

CONSENT STATEMENT

All human participants or their caregivers provided informed written consent to participate in the respective studies.

Supporting information

Supporting Information

Supporting Information

Trudel L, Therriault J, Macedo AC, et al. Rates of clinical progression according to biological Alzheimer's disease stages. Alzheimer's Dement. 2025;21:e70624. 10.1002/alz.70624