Clinical Impact and Prognostic Value of Alzheimer Disease Biomarkers in the Very Old

Abstract

Background and Objectives

Alzheimer disease (AD) is increasingly viewed as a clinical-biological continuum, but the utility of biomarkers in individuals aged ≥80 years, the so-called “very old,” remains uncertain, because of concerns about its clinical usefulness. This study aimed to determine the clinical impact and prognostic value of AD biomarkers in very old individuals.

Methods

We performed a retrospective longitudinal cohort study within the Sant Pau Initiative on Neurodegeneration (SPIN, Barcelona, Spain), a memory clinic–based research cohort. Patients were referred from primary care physicians or community neurologists for evaluation within a public universal health care catchment area. We included SPIN participants evaluated between October 2013 and May 2024 who were aged ≥80 years and had mild cognitive impairment (MCI) at the baseline visit. Participants underwent standardized clinical and neuropsychological assessments and had CSF and plasma biomarker measurements available. AD biology was defined by the CSF phosphorylated tau at threonine 181/amyloid-β 42 ratio. Plasma phosphorylated tau at threonine 217 (p-Tau217) was evaluated for (1) diagnostic accuracy for AD biology and (2) prognostic associations with longitudinal cognitive change (Mini-Mental State Examination [MMSE]) and progression to dementia. Analyses used linear mixed-effects models for MMSE trajectories and Cox regression for dementia conversion.

Results

A total of 167 participants were included (mean age 82.3 years, 59% women) with a mean follow-up of 35.8 months; the ones with AD biology (n = 116, 69%) had worse baseline cognitive performance, particularly in memory, compared with those without (Cohen d = 0.34, p = 0.03). Plasma p-Tau217 showed excellent diagnostic accuracy for detecting AD biology (0.93, 95% CI 0.88–0.98; cutoff 0.19 pg/mL; sensitivity 94.5%; specificity 84%). Over time, participants with AD biology declined faster on the MMSE than those without (−0.47 vs −0.18 points per year; p < 0.01). Cox models showed an increased risk of progression to a dementia stage among individuals with higher plasma p-Tau217 concentrations (hazard ratio 1.49, 95% CI 1.05–2.13, p = 0.026).

Discussion

In very old individuals with MCI, AD biology (CSF) and plasma p-Tau217 identify individuals at higher risk of faster cognitive decline and progression to dementia. Limitations of this study include the single-center design and the modest sample size.

Introduction

Alzheimer disease (AD) is increasingly viewed as a clinical-biological continuum that can be detected decades before dementia onset through imaging and fluid biomarkers like CSF and plasma analytes.¹ Advances in these biomarkers have transformed diagnostic practice and facilitated the development and approval of disease-modifying therapies for early symptomatic stages. Notably, plasma phosphorylated tau at threonine 217 (p-Tau217) has emerged as a minimally invasive, scalable, and highly accurate marker of AD biology, with diagnostic performance comparable with CSF and amyloid PET in younger and middle-aged population.^2-6 Reflecting this evidence, the latest criteria from the Alzheimer's Association now recognize p-Tau217 as a stand-alone biomarker for AD diagnosis.⁷ With the advent of disease-modifying therapies, lecanemab and donanemab, early detection becomes increasingly critical to delay cognitive and functional decline, reduce institutionalization, and potentially lower mortality.^8-10

Despite this progress, biomarker-guided evaluation remains uncommon in individuals aged ≥80 years, who now represent the fastest-growing segment of the population affected by cognitive impairment.¹¹ Long-standing concerns originate from neuropathologic studies showing high prevalence of amyloid plaques and tau tangles in the very old, often coexisting with other age-related pathologies.^12-14 This has historically raised scepticism about the specificity and clinical value of AD biomarkers in late life. However, recent evidence challenges the notion that Alzheimer disease neuropathologic change (ADNC) represents an incidental finding in advanced age.^15-18 Several studies have shown that ADNC may contribute to cognitive decline even in the presence of multimorbidity, and growing data indicate that p-Tau217 maintains excellent diagnostic performance for AD pathology across the age spectrum.^2,19,20

Given the ongoing demographic aging and the clinical need for accurate risk stratification in very old adults (≥80 years), it is essential to clarify whether AD biomarkers remain clinically informative in this population. Using the Sant Pau Initiative on Neurodegeneration (SPIN) cohort,²¹ we evaluated AD biomarkers in memory clinic patients aged ≥80 years with cognitive complaints, focusing on CSF-defined AD biology and plasma p-Tau217. Specifically, we aimed to (1) assess the diagnostic performance of plasma p-Tau217 for CSF-defined AD biology; (2) examine associations between AD biology and baseline cognition and medical comorbidity; and (3) determine whether AD biology and plasma p-Tau217 are associated with longitudinal cognitive decline and progression to dementia.

Methods

Study Participants

Participants were recruited at the Sant Pau Memory Unit, within the Department of Neurology at Hospital de la Santa Creu i Sant Pau (Barcelona, Spain), a public, tertiary memory clinic within Catalonia's universal health care system that serves a defined reference population. Patients are typically referred by primary care physicians and community neurologists for specialized diagnostic evaluation because of cognitive and/or behavioral symptoms; self-referral is not the usual pathway. Access to evaluation is covered by the public system (i.e., no out-of-pocket costs for patients within the reference area), and first appointments are generally scheduled within approximately 2–3 months. Following routine clinical assessment, eligible and consenting individuals are invited to participate in the SPIN,²¹ an umbrella cohort including individuals with a range of neurodegenerative conditions. From the SPIN database, we included all consecutive participants aged ≥80 years who met criteria for mild cognitive impairment (MCI) at the baseline visit, defined by objective cognitive decline with preserved functional independence.²² Eligible participants were evaluated between October 2013 and May 2024, including both amnestic and nonamnestic MCI. For inclusion in this analysis, participants were required to have CSF and plasma biomarker measurements available from the diagnostic workup and to have provided written informed consent.

Key exclusion criteria include conditions that preclude reliable neuropsychological assessment (e.g., severe sensory deficits/illiteracy), contraindications to MRI and/or lumbar puncture, current use of medications that may substantially impair cognition, major neurologic disorders (e.g., stroke, epilepsy, structural brain lesions), major psychiatric illness, recent substance abuse, and active or insufficiently treated malignancy with potential CNS involvement.²¹ Neuroimaging was obtained as part of the clinical evaluation; however, cerebrovascular comorbidity was not systematically excluded in this memory clinic–based cohort.

Clinical Assessment

Baseline was defined as each participant's first visit at the Sant Pau Memory Unit (index visit), when clinical and biomarker assessments were performed; thus, baseline occurred at different calendar dates across participants (October 2013–May 2024).

At first visit, all participants underwent a full clinical assessment, including collection of demographic data, comorbidities (toxic habits, hypertension, stroke, epilepsy, history of traumatic brain injury or “TBI,” primary psychiatric disorder and neoplasia), biochemical analytical routinal parameters and ongoing pharmacologic therapies (with particular attention to drugs with potential cognitive effect). The estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation (version of 2009),²³ without a race coefficient.

Neuropsychological Assessment

Participants underwent a standardized neuropsychological battery according to the SPIN protocol.²¹ Episodic memory was assessed using the Free and Cued Selective Reminding Test (total free recall, total recall, delayed free recall, and delayed total recall) and the Rey–Osterrieth Complex Figure delayed recall. Language was assessed using the Boston Naming Test (percentage correct) and Digit Span forward and backward. Attention and executive functions were evaluated using the Trail Making Test parts A and B, semantic (animals), and phonemic verbal fluency (letter P) tasks. Visuospatial abilities were evaluated with the Rey–Osterrieth Complex Figure copy, the Poppelreuter test, and the Visual Object and Space Perception battery number location subtest. Domain-specific composite scores were calculated as the mean of available standardized z-scores within each cognitive domain. For neuropsychological outcomes, we used standardized scores derived from demographically adjusted amyloid-β–negative next-generation norms, generated using generalized additive models for location, scale, and shape. This approach was used because Spanish normative references are not available as z-score metrics across the full age/education distribution. Next-generation norms–derived standardized scores were used to build domain composite measures for statistical analyses.^24,25

The z scores were used solely to assess the correlation between biomarkers and cognition and were not used for participant classification. More details on this approach have been recently reported by our group.^24,25 The Mini-Mental State Examination (MMSE)²⁶ was used as a general measure of global cognition because of its wide availability during routine clinical follow-up and its high responsiveness to clinical decline during long follow-up periods in persons with AD biology.²⁷ Functional status was assessed with the Interview for Deterioration in Daily Living in Dementia (IDDD) scale,²⁸ specific for functional decline in dementia, and the global level of cognitive impairment was quantified by Clinical Dementia Rating Sum of Boxes (CDR-SB)²⁹ and Global Deterioration Scale.³⁰

CSF and Plasma Biomarkers

CSF and plasma samples were collected according to the SPIN cohort protocol and following international recommendations.²¹ Briefly, CSF samples were obtained via lumbar puncture, centrifuged, aliquoted, and stored at −80°C for subsequent analysis. CSF levels of Aβ1–42, Aβ1–40, and phosphorylated tau at threonine 181 (p-Tau181) were measured using the Lumipulse G600II fully automated platform with commercially available kits (Fujirebio Europe, Ghent, Belgium). All CSF measurements were scheduled twice a month and were used for routine diagnostic assessment of patients. Blood samples were drawn into EDTA-K2 tubes, processed within 2 hours of extraction, and centrifuged at 2,000 rpm for 10 minutes at 4°C. The plasma was then aliquoted and stored at −80°C until analysis. Plasma p-Tau217 concentrations were also measured in the Lumipulse fully automated platform G600II using the commercially available kits. Plasma measurements were made by laboratory technicians blinded to clinical data, and treating physicians did not have access to these measurements during follow-up.

Genetic Analyses

DNA was extracted from whole blood using standard procedures,²¹ and APOE genotyping was performed by direct sequencing of exon 4 as part of the routine assessment.

Definition of AD

AD biology was defined using the CSF p-Tau181/amyloid-β 42 (Aβ42) ratio because it has shown the best diagnostic performance for detecting amyloid PET positivity and the highest accuracy in predicting ADNC in autopsy-proven multicenter cohorts.^31,32 Participants with a CSF p-Tau181/Aβ42 ratio ≥0.068 were classified as positive AD biology (AD-positive), otherwise negative.³¹

Clinical Follow-Up

Per the SPIN protocol, participants in stages 2–6 were scheduled for annual follow-up visits, including neuropsychological assessment. Additional visits could be performed more frequently when clinically indicated, at the discretion of the treating physician.

These assessments aimed to document subtle cognitive decline over time, which may not have been detectable with brief cognitive tests or evident through functional decline. The full baseline neuropsychological battery was not systematically repeated at follow-up visits; longitudinal cognition was therefore assessed using MMSE.

At each visit, clinicians recorded the most likely etiology based on available clinical and biomarker information, including AD, frontotemporal lobar degeneration, Lewy body disease, other neurodegenerative disease, non-neurodegenerative disease, or uncertain etiology. Data on clinical diagnosis, clinical stage, and MMSE scores were prospectively collected during each visit using Research Electronic Data Capture tool.³³

Statistical Analysis

Data were explored for normality using the Shapiro-Wilk test. Between-group differences were determined with analysis of variance or t test for normally distributed variables (with false discovery rate correction for multiple comparisons), Mann-Whitney or Wilcoxon tests for non-normally distributed variables, and the χ² for dichotomous or categorical data. All tests were performed in R statistical software version 4.2.2. The α threshold was set at 0.05 for all analyses.

The diagnostic accuracy of plasma p-Tau217 in discriminating subjects with positive AD biology (AD-positive) was assessed using receiver operating characteristic (ROC) analysis. Logistic regression models were applied to test the association between plasma p-Tau 217 and CSF-defined AD biology after adjusting for age, sex, years of education, and estimated glomerular filtration rate. In additional analyses, we evaluated the diagnostic performance of baseline MMSE for CSF-defined AD biology using ROC curves and compared areas under the curve (AUCs) for MMSE vs plasma p-Tau217 using the DeLong test. We also fit logistic regression models including MMSE and p-Tau217 (adjusted for age, sex, education, and eGFR) to assess whether p-Tau217 provides incremental discrimination beyond MMSE.

Linear mixed-effects (LME) models were used to evaluate longitudinal changes in MMSE scores over time in participants with AD biology. Time, age at first assessment, sex, years of education, estimated glomerular filtration rate, biomarker of interest (CSF or plasma p-Tau217), and the interaction of AD biology and time were included as fixed effects. Random intercepts and slopes were specified to account for within-subject variability. We used bootstrap resampling to assess the robustness of the cutoffs.

In survival analyses, survival time was defined as the time from the first visit at which stage 4 was recorded. Participants who had not developed dementia by the time of analysis were censored at their last follow-up visit. Cox proportional hazards regression models were used to estimate the risk of progression to dementia, adjusted for age, sex, years of education, and estimated glomerular filtration rate as covariates. The assumption of proportional hazards was tested and confirmed. For illustrative purposes, survival curves were generated using the ggsurvplot function from the survminer package in R.

Standard Protocol Approvals, Registrations, and Participant Consents

All procedures in the study were approved by the Sant Pau ethics committee (protocol number IIBSP-DOW-2014-30) and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants.

Data Availability

The data sets generated and analyzed during this study are available from the corresponding authors upon reasonable request.

Results

Sample Composition and Demographics

Table 1 describes the baseline characteristics of the 167 participants with MCI included in the study stratified by AD biology. A total of 116 participants (70%) had AD biology, as determined by the CSF p-Tau181/Aβ42 ratio. The age, sex distribution, education, and marital/living status were similar across participants with and without AD biology (Table 1). The mean follow-up in the analytic sample was 35.8 months and did not differ significantly between AD biology–positive and AD biology–negative participants (38.1 vs 30.5 months, p = 0.13). Because the number of participants under observation decreased substantially beyond this point, figures were truncated at 5 years. As expected, participants with AD biology had a higher frequency of APOE ε4 carriership (34% vs 4.5%, Cohen d = 0.71, p < 0.01). Despite showing similar functional impairment at baseline (as measured with IDDD), participants with AD biology showed slightly greater cognitive functional impairment, as measured with the CDR-SB (2.8 vs 2.4, Cohen's d = 0.34, p = 0.042) and the MMSE (24.7 vs 25.9, Cohen's d = 0.37, p = 0.049). The frequency of the most significant comorbidities and baseline pharmacologic treatments was also similar between participants with and without AD biology. The only notable exception was a history of TBI, which was reported more often in participants without AD biology than in those with AD biology (7.8% vs 0.9%, respectively; Cramér V = 0.18; p = 0.028). Additional details on participants' comorbidities at baseline are provided in the supplementary material (eTable 1). eTable 2 presents a panel of biochemical parameters used as part of the baseline screening for cognitive impairment, but no statistically significant differences were observed.

Table 1.

General Characteristics

Characteristic	All participants (N = 167)a	Participants without AD biology (n = 51)a	Participants with AD biology (n = 116)a	p Valueb
Sociodemographic characteristics
Age, y	82.3 (1.6)	82.1 (1.7)	82.4 (1.6)	0.2
Age group				>0.9
80–85	155 (93)	48 (94)	107 (92)
>85	12 (7.2)	3 (5.9)	9 (7.8)
Biologic sex, n (%)				0.3
Male	69 (41)	24 (47)	45 (39)
Female	98 (59)	27 (53)	71 (61)
Years of education	10.0 (4.2)	10.1 (4.1)	9.9 (4.2)	0.8
Marital status				>0.9
Married	108 (65)	34 (67)	74 (64)
Divorced	2 (1.2)	0 (0)	2 (1.7)
Not married	3 (1.8)	1 (2.0)	2 (1.7)
Widow/er	40 (24)	12 (24)	28 (24)
First domicile				0.4
Lives at home with a caregiver	121 (72)	37 (73)	84 (72)
Institutionalized	1 (0.6)	1 (2.0)	0 (0)
Lives alone at home	38 (23)	10 (20)	28 (24)
Total follow-up, mo	35.8 (36.5)	30.5 (36.0)	38.1 (36.7)	0.13
Etiologic diagnosis at last visit				<0.001
AD	95 (57)	0 (0)	95 (82)
FTLD	11 (6.6)	9 (18)	2 (1.7)
LBD	17 (10)	9 (18)	8 (6.9)
Others	44 (26)	33 (65)	11 (9.5)
APOE ε4 (carriership), n (%)	36 (25)	2 (4.5)	34 (34)	0.001
Global function assessment
CDR-SB	2.7 (1.3)	2.4 (1.2)	2.8 (1.3)	0.04
Total IDDD score	38.8 (5.6)	39.2 (5.6)	38.7 (5.7)	0.5
MMSE at first visit	25.1 (3.5)	25.9 (2.8)	24.7 (3.7)	0.04
AD biomarker status
Plasma p-Tau217	0.44 (0.57)	0.16 (0.19)	0.56 (0.64)	<0.001
Baseline cognitive composite z-scores
Memory	−1.88 (0.96)	−1.65 (1.01)	−1.98 (0.92)	0.03
Executive	−1.03 (0.72)	−0.85 (0.76)	−1.11 (0.69)	0.06
Language	−0.71 (0.79)	−0.63 (0.82)	−0.75 (0.78)	0.13
Visuospatial	−0.33 (1.69)	−0.06 (1.85)	−0.44 (1.60)	0.34

Abbreviations: AD = Alzheimer disease; Aβ42 = amyloid‐β 42; CDR-SB = Clinical Dementia Rating, sum of boxes; FTLD = frontotemporal lobe degeneration; IDDD = Interview for Deterioration in Daily Living in Dementia; LBD = Lewy body disease; MMSE = Mini-Mental State Examination; p‐Tau181 = phosphorylated tau at threonine 181; p-Tau217 = phosphorylated tau at threonine 217.

The table shows the general characteristics of the 2 groups at the first visit. AD biology is defined by CSF p-Tau181/Aβ42 ratio (cutoff of ≥0.068). Values are expressed as mean (SD) or n (%) as appropriate. Continuous variables were compared by Wilcoxon rank-sum test; categorical variables were compared using Pearson χ² test of Fisher exact test when was required.

^aMean (SD); n (%).

^bWilcoxon rank sum test; Fisher exact test; Pearson χ² test.

Cognitive Differences at Baseline by AD Biology in the Very Old

Baseline domain composite scores are summarized in Table 1. As illustrated in Figure 1, memory was the most impaired cognitive domain in participants with positive AD biology (−1.98 vs −1.65; Cohen d = −0.34; p = 0.03), followed by executive function, language, and visuospatial composites (−1.12; −0.75; −0.45, respectively), but no statistically significant differences were observed between the 2 groups comparing the other domains (additional data are listed in eTables 3 and 4).

Figure 1. Comparison of Cognitive Domains by AD Biology.

Plots show the comparison of z-standardized composite scores in 4 cognitive domains (Memory, Executive Functions, Language, Visuospatial) between individuals stratified by AD biology. The Wilcoxon rank sum test is applied for comparison between groups. Effect sizes are reported as Cohen d. Statistically significant was set at p-values <0.05. AD = Alzheimer disease.

Diagnostic Performance of Plasma p-Tau217 for AD Biology in the Very Old

Next, we explored the diagnostic accuracy of plasma p-Tau217 in identifying AD biology in the very old. As shown in Figure 2, the optimal cutoff defined by Youden Index for identifying participants with AD biology was 0.19 pg/mL (sensitivity 94.5%, specificity 84%, positive predicted value [PPV] 93%, negative predicted value [NPV] 87%), and the results indicate an excellent discriminatory capacity of plasma p-Tau217 (AUC = 0.93, 95% CI 0.88–0.98). Because renal function may affect circulating biomarker concentrations, all primary models were adjusted for eGFR. Adjustment for age, sex, education, and eGFR did not materially change the diagnostic accuracy of plasma p-Tau217 for CSF-defined AD biology (eTable 5). In exploratory analyses (eFigure 1), plasma p-Tau217 concentrations were higher at lower eGFR; however, separation between AD biology–positive and AD biology–negative participants remained evident across renal function strata.

Figure 2. Diagnostic Accuracy of Plasma p-Tau217 for AD Biology in the Very Old.

In panel A, the ROC curve shows the performance of p-Tau 217 in discriminating AD biology, as defined by CSF p-Tau181/Aβ42 ratio cutoff of ≥0.068 (gold standard). The optimal cutoff was determined by the Youden Index, via bootstrapping resampling. The figure reports the AUC with its 95% CI. In panel B, the boxplot illustrates the distribution of p-Tau217 stratified by AD biology (CSF p-Tau181/Aβ42 ratio). The horizontal bar represents the group mean. The Wilcoxon rank-sum test is used for comparison between groups. Statistical significance was set at p < 0.05, and the corresponding p-value is shown in the figure. Aβ42 = amyloid-β 42; AD = Alzheimer disease; AUC = area under the curve; GS = gold standard; p-Tau217 = phosphorylated tau at threonine 217; p-Tau181 = phosphorylated tau at threonine 181; ROC = receiver operating characteristics.

Baseline MMSE showed limited discrimination for CSF-defined AD biology (AUC = 0.45, 95% CI 0.34–0.57), substantially lower than plasma p-Tau217 (AUC = 0.93, 95% CI 0.88–0.98; p < 0.001 by DeLong test for AUC difference). In multivariable models, plasma p-Tau217 remained independently associated with AD biology after including MMSE and covariates, and the combined model yielded an AUC of 0.95 (95% CI 0.89–1.00). More details are available in the supplementary material (eTable 6).

The distribution of final etiologic diagnoses is shown in eFigure 2.

Relationship Between AD Biology and Longitudinal Change in MMSE

Figure 3 illustrates MMSE trajectory predictions by AD biology, as estimated by the LME models (more details in eTable 6). Participants with AD biology (as defined with CSF biomarkers) showed a faster cognitive decline over time (β = −3.95, 95% CI −5.99 to −1.90, p < 0.001) than participants without AD biology. Participants with a positive plasma p-Tau217 (using the 0.19 pg/mL cutoff) also presented a faster cognitive decline over time (β = −0.5, 95% CI −0.98 to −0.01, p = 0.044) compared with participants with a negative p-Tau217 result at baseline. When plasma p-Tau217 levels were included as a continuous variable, higher p-Tau217 levels at baseline predicted a faster MMSE decline over time (β = −1.36, 95% CI −1.95 to −0.77, p < 0.001). For illustrative purposes, trajectory plots were constructed using the variable dichotomized at the 0.19 pg/mL threshold from ROC curves.

Figure 3. Longitudinal Change in MMSE Trajectories by AD Biology.

In panel A, AD biology is defined by the CSF p-Tau181/Aβ42 ratio using the cutoff of ≥0.068. In panel B, AD biology is based on binary value of plasma p-Tau217 (cutoff of >0.19 pg/mL, using the Youden Index). These predictions were derived from linear mixed-effects models, including AD biology, time of follow-up, age at first visit, years of education, sex, glomerular filtration rate, and the interaction between time and AD biology as fixed effects. Statistical significance was set at p < 0.05. Aβ42 = amyloid-β 42; AD = Alzheimer disease; MMSE = Mini-Mental State Examination; p-Tau181 = phosphorylated tau at threonine 181; p-Tau217 = phosphorylated tau at threonine 217.

Although the presence of increased risk genotypes of APOE ε4 was tested as an additional covariate in preliminary models (β = −0.18, 95% CI −0.90 to 0.36, p = 0.589), we excluded it from the final model because it did not improve its performance. We could not find statistical significance in age, sex, years of education, or estimated glomerular filtration rate at baseline in the cognitive trajectories.

Relationship Between AD Biology and Progression to Dementia in the Very Old

Finally, we applied Cox regression models to investigate the association between AD biology and the risk of clinical progression to dementia. The presence of AD biology in CSF was associated with an increased risk of clinical progression to dementia stages (hazard ratio [HR] 1.64, 95% CI 1.00–2.70, p = 0.051). Participants with positive plasma p-Tau217 (>0.19 pg/mL as cutoff) also showed a higher risk of dementia progression (HR 1.64, 95% CI 0.98–2.74, p = 0.060). Similarly, increased concentrations of plasma p-Tau217 analyzed as a continuous variable were significantly associated with a higher risk of dementia (HR 1.49, 95% CI 1.05–2.13, p = 0.026). Age, sex, education, and glomerular filtration rate did not show any statistical significance in the models (eTable 7). As expected, the ability of AD biology in both CSF and plasma to predict progression to dementia was significantly improved after excluding participants who progressed to non-AD dementia during follow-up from the analyses (Figure 4; eFigure 3).

Figure 4. Probability of Dementia Progression According to AD Biology.

Kaplan-Meier survival curves represent the probability of progression to dementia stage over time for participants classified by AD biology, excluding who progressed to non-AD dementia during follow-up. In panel A, AD biology is defined by the CSF p-Tau181/Aβ42 ratio using the cutoff of ≥0.068 (log-rank p = 0.003). In panel B, AD biology is based on binary value of plasma p-Tau217 (cutoff of >0.19 pg/mL, using the Youden Index) (log-rank p < 0.001). Shaded areas indicate 95% CIs. Aβ42 = amyloid-β 42; AD = Alzheimer disease; p-Tau181 = phosphorylated tau at threonine 181; p-Tau217 = phosphorylated tau at threonine 217.

Discussion

In this study, we demonstrate that AD biology retains diagnostic and prognostic value in individuals aged ≥80 years. Using longitudinal data from a memory clinic–based cohort, we found that baseline cognitive differences by AD biomarker status were modest, whereas AD biology and higher plasma p-Tau217 were associated with faster cognitive decline and a higher risk of progression to dementia. These findings provide evidence that AD biology continues to play a clinically meaningful role even in advanced age, a population traditionally perceived as less responsive to biomarker-based stratification.

Our first aim was to assess the diagnostic performance of plasma p-Tau217 in this age group. We found that its ability to detect AD biology remained excellent, comparable with that reported in younger and middle-aged populations.^2,3,6,34 Plasma biomarkers also offer practical advantages for implementation: Blood-based testing is minimally invasive, scalable, and may be more cost-effective than CSF or amyloid PET as an initial triage step. This is particularly relevant in very old adults, in whom lumbar puncture, amyloid PET, and repeated in-person follow-up can be logistically difficult. A pragmatic approach is a 2-step workflow, using plasma p-Tau217 for initial screening and reserving confirmatory testing for equivocal cases or when diagnostic uncertainty remains.^35,36 The biomarker also retained specificity in differentiating AD from other age-associated pathologies, such as primary age-related tauopathy,²⁰ reinforcing its value as a minimally invasive tool for clinical use in the very old.

A biomarker-guided approach offers an important advantage in very old adults because neuropsychological phenotypes alone have limited discriminative power in this age group. In our cohort, although participants with and without AD biology differed in cognitive performance, these differences were modest and did not consistently remain significant after correction for multiple comparisons. This overlap underscores the need for molecular biomarkers to distinguish AD biology from other common age-associated neurodegenerative pathologies, an increasingly important challenge in the era of disease-modifying therapies. Although AD biology was associated with a predominantly amnestic profile consistent with early AD with milder executive dysfunction,^37,38 AD-negative participants also exhibited substantial memory impairment. This pattern is compatible with non-AD etiologies common in advanced age, including limbic-predominant age-related TDP-43 encephalopathy (LATE),^12,39 which may account for a meaningful proportion of memory-predominant presentations in the absence of AD biomarkers.^40,41

Another notable observation was the higher prevalence of TBI among participants with MCI without AD biology. Although effect sizes were modest, this finding aligns with prior work identifying TBI as a risk factor for dementia.^42,43 It raises the possibility that alternative mechanisms, such as chronic traumatic encephalopathy, may contribute to cognitive decline in the very old. In the absence of validated in vivo biomarkers of TBI-related neurodegeneration, further research is needed, particularly given the high frequency of falls in very old adults.⁴⁴

The second aim of our study was to evaluate the clinical consequences of AD biology in this age group. Although AD-negative participants also showed deterioration, likely reflecting comorbid pathologies and frailty,⁴⁵ AD-positive participants declined at nearly double the rate on MMSE and had shorter dementia-free survival. Although the annual difference in MMSE points may seem modest, its cumulative impact is clinically meaningful and may determine conversion from MCI to dementia. These findings extend previous observations in younger cohorts and align with studies in octogenarians and nonagenarians,^18,46 supporting the idea that AD biology confers an independent risk of cognitive deterioration even in very advanced age. In addition, plasma p-Tau217 retained prognostic value, with higher levels associated with poorer cognition and increased dementia risk, consistent with observations in younger populations.³⁴ However, in very old individuals, the mere presence of amyloid or CSF AD biology may not discriminate fast from slow progressors as effectively as p-Tau217, which may better identify active, clinically symptomatic disease. Overall, our results support the view that AD biology retains strong prognostic value, particularly after excluding non-AD neurodegenerative diseases. One important implication of our findings is that the results of AD biomarkers in the very old should be carefully interpreted, considering the clinical phenotype to exclude neurodegenerative diseases other than AD.

Although recent studies have proposed the use of double predefined thresholds of p-Tau217 and even a 2-step strategy to classify patients into low, intermediate, and high probability of amyloid pathology,^2,35,47 those cutoffs have been extrapolated from younger cohorts and may not perform well in very old individuals. When applied to our sample, the proposed and previously validated in younger ages p-Tau217 cutoffs (<0.19 pg/mL, >0.39 pg/mL)² correctly identified AD pathology with excellent specificity and PPV (97%), but shown reduced sensitivity (47%) and placed half of the sample into the intermediate zone, limiting their clinical application. Our findings suggest that p-Tau217 cut-off values in very old patients may be lower than those proposed for younger populations. This may reflect not only age-related differences in the cerebral distribution of tau pathology⁴⁸ but also the high burden of copathologies^16,49 and lower cognitive reserve, which may contribute to cognitive decline. To date, no validated, age-adjusted p-Tau217 threshold exists, underscoring the need for additional studies before a biomarker-based strategy can be applied in clinical practice.

Collectively, these results carry important clinical implications. They challenge the assumption that biomarker testing is of limited value in very old adults,⁵⁰ demonstrating that identifying AD pathology provides meaningful prognostic information. This may support more informed care planning, guide the use of disease-modifying therapies, and help allocate resources efficiently. Nonetheless, biomarkers should be interpreted within a broader geriatric context. Non-AD pathologies, frailty, gait impairment, medical comorbidities, and psychosocial factors remain highly prevalent among the very and the oldest-old^44,45,49 and are likely to interact with AD biology.

Renal impairment is common in very old adults and may influence plasma biomarker concentrations. Consistent with this, plasma p-Tau217 levels increased with worsening renal function in our cohort (eFigure 1). Importantly, adjusting for eGFR did not materially alter the discrimination of AD biology or the associations with longitudinal cognitive decline and progression to dementia, suggesting that our main findings are unlikely to be driven solely by renal comorbidity. A multimodal approach, integrating biomarker testing with comprehensive geriatric assessment, may be the most effective strategy for determining who benefits most from intensive diagnostic evaluation or targeted interventions.⁸

Our study also has some limitations. Our results are based on a monocentric study of a unique cohort with up to 10 years of follow-up, in which participants aged 80 years or older were not systematically excluded from CSF biomarker testing. Although longitudinal follow-up was available, the sample size was modest, particularly for participants aged >85 years, who comprised 12/167 (7%) of the cohort (3 AD-negative and 9 AD-positive). This small subgroup precluded separate analyses with adequate statistical power and warrants caution when extrapolating our findings specifically to individuals aged >85 years. Postmortem confirmation was not available, and biomarkers of non-AD neurodegenerative pathologies (e.g., LATE, Lewy body disease) were not assessed. Future research should focus on larger multicenter cohorts with neuropathologic validation and should investigate resilience and protective factors in biomarker positive older adults who remain clinically stable.

In summary, AD biomarkers retain significant diagnostic and prognostic value in very old individuals with cognitive complaints. CSF and plasma measures of AD biology predict faster cognitive decline and increased dementia risk, and plasma p-Tau217 offers a scalable, noninvasive option for detecting AD pathology in this age group. In an era of emerging therapies and personalized medicine, excluding very old adults from biomarker-based diagnostic pathways risks creating inequities in access to care. Our findings help fill the critical evidence gap surrounding this often-overlooked population.

Glossary

Aβ42

amyloid-β 42

AD

Alzheimer disease

ADNC

Alzheimer disease neuropathologic change

AUC

area under the curve

CDR-SB

Clinical Dementia Rating Sum of Boxes

eGFR

estimated glomerular filtration rate

HR

hazard ratio

IDDD

Interview for Deterioration in Daily Living in Dementia

LATE

limbic-predominant age-related TDP-43 encephalopathy

LME

linear mixed-effect

MCI

mild cognitive impairment

MMSE

Mini-Mental State Examination

NPV

negative predicted value

PPV

positive predicted value

p-Tau181

phosphorylated tau at threonine 181

p-Tau217

phosphorylated tau at threonine 217

ROC

receiver operating characteristics

SPIN

Sant Pau Initiative on Neurodegeneration

TBI

traumatic brain injury

Author Contributions

C. Ceriello: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. O.H. Torres: drafting/revision of the manuscript for content, including medical writing for content. S. Rubio-Guerra: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. J. García-Castro: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. J. Selma-Gonzalez: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. I. Sala: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. M.B. Sánchez-Saudinós: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. L. Videla: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. E. Vera-Campuzano: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. J. Arranz: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. Í. Rodríguez-Baz: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. L. Maure-Blesa: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. O. Dols-Icardo: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. S. Valldeneu: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. I. Barroeta: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. M. Santos-Santos: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. M. Carmona-Iragui: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. L. Vaqué-Alcázar: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. E. Alvarez-Sanchez: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. O. Lorente: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. M. Carreras: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. A. Bejanin: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. O. Belbin: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. A. Lleó: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. J. Fortea: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. D. Alcolea: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. I. Illán-Gala: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data.

Study Funding

We acknowledge the Support for Research Groups funding from the Department of Research and Universities of the Generalitat de Catalunya (2021 SGR 00979). The SPIN cohort was funded by Instituto de Salud Carlos III (ISCIII) through the projects PI14/01126, PI17/01019, PI18/00335, PI19/00882, PI18/00435, PI22/00611, INT19/00016, INT23/00048, PI17/01896, PI22/00307, AC19/00103, PI22/00758, ICI23/00032, INT21/00073, PI20/01473, PI24/01087, PI23/01786, PI21/00791, PI2400598 and co-funded by the European Union. The SPIN cohort received additional support from NIH grants (R01 AG056850-01A1, R21 AG056974, R01 AG061566, and R01AG080470), Generalitat de Catalunya (2017-SGR-547, SLT006/17/125, SLT006/17/119, SLT002/16/408, SLT042/25/000032), “La Marató de TV3” foundation grants (20141210, 044412, and 20142610), Fundació Bancaria “La Caixa” (DABNI project), Fundació Catalana Síndrome de Down, and Fundació Víctor Grífols i Lucas. This study also received funding from the Centro de Investigación Biomédica en Red sobre Enfermedades Neurodegenerativas (CIBERNED) Program 1, jointly funded by the Fondo Europeo de Desarrollo Regional (FEDER) and the European Union under the framework “Una manera de hacer Europa”/“A way to make Europe.” S. Rubio-Guerra received contract funding from the Alzheimer's Association (AACSF-21-850193). J. García-Castro, J.E. Arriola-Infante, I. Rodríguez-Baz, and L. Maure Blesa are supported by a Río Hortega grant (CM23/00176, CM22/00219, CM22/00052, CM23/00291) from the Carlos III National Institute of Health of Spain, partly funded by the European Social Fund. M. Carmona-Iragui is a senior Atlantic Fellow for Equity in Brain Health at the Global Brain Health Institute (GBHI) and reports grants from the Alzheimer's Association (AARG-22-973966), Global Brain Health Institute (GBHI) (GBHI_ALZ-18-543740), Jérome Lejeune Foundation (1913 cycle 2019B), and Societat Catalana de Neurologia (SCN2020). L. Vaqué-Alcázar is supported by a Sara Borrell grant (CD23/00235) from the Carlos III National Institute of Health of Spain. Miguel Santos-Santos acknowledges support from the Spanish Institute of Health Carlos III co-funded by the European Union (Juan Rodés research grant JR18-00018; Fondo de investigación sanitaria grant PI19/00882), the Department of Research and Universities from the Generalitat de Catalunya (2021 SGR 00979), the Alzheimer's Association clinician scientist fellowship (AACSF-22-972945), and the NIH (R01AG080470). A. Bejanin acknowledges support from the Instituto de Salud Carlos III through the Miguel Servet grant “CP20/00038,” which is co-funded by the European Union, the Alzheimer's Association (AARG-22-923680), and the Ajuntament de Barcelona, in collaboration with Fundació La Caixa (23S06157-001). O. Dols-Icardo receives funding from the Alzheimer's Association (AARF-22-924456) and the Fondation Jérôme Lejeune (PDC-2023-51; 202307). I. Illán-Gala is a senior Atlantic Fellow for Equity in Brain Health at the Global Brain Health Institute (GBHI) and receives funding from the Alzheimer's Association and the Alzheimer Society (GBHI ALZ UK-21-720973 and AACSF-21-850193). I. Illán-Gala was also supported by the Juan Rodés Contract (JR20/0018) from the Carlos III National Institute of Health of Spain, partly funded by the European Social Fund.

Disclosure

S. Rubio-Guerra serves on the advisory board for educational events of Esteve Pharmaceuticals, S.A. J. Arranz reports receiving personal fees for service on advisory boards, speaker honoraria, or educational activities from Lilly, Roche Diagnostics, and Esteve, outside the submitted work. I. Barroeta reports receiving personal fees for educational activities from Adium. M. Carmona-Iragui reports receiving personal fees for service on advisory boards, speaker honoraria, or educational activities from Lilly, Roche, Adium Pharma, Esteve, and Neuraxpharm, outside the submitted work. O. Belbin reports holding a patent for markers of synaptopathy in neurodegenerative disease (licensed to ADx NeuroSciences N.V., WO2019175379 Markers of synaptopathy in neurodegenerative diseases). A. Lleó reported receiving personal fees for service on advisory boards or speaker honoraria from Almirall, Beckman-Coulter, Biogen, Eisai, Esteve, Fujirebio-Europe, Grifols, KRKA, Lilly, Novartis, NovoNordisk, Nutricia, Otsuka Pharmaceutical, Roche, and Zambón. AL is co-author of a patent for markers of synaptopathy in neurodegenerative disease (licensed to ADx, EPI8382175.0) and on antibodies for amyloid precursor, methods and uses thereof, European priority (EP25382226). J. Fortea reported receiving personal fees for service on the advisory boards, adjudication committees, or speaker honoraria from AC Immune, Adamed, Alzheon, Biogen, Eisai, Esteve, Fujirebio, Ionis, Laboratorios Carnot, Life Molecular Imaging, Lilly, Novo Nordisk, Perha, Roche, Zambón, and outside the submitted work. He reports holding a patent for markers of synaptopathy in neurodegenerative disease (licensed to ADx, EPI8382175.0). D. Alcolea reported receiving personal fees for service on advisory boards of Fujirebio-Europe, Roche Diagnostics, Grifols S.A., and Lilly, and received speaker honoraria from Fujirebio-Europe, Roche Diagnostics, Nutricia, Krka Farmacéutica S.L., Zambon S.A.U., Neuraxphar, and Esteve Pharmaceuticals S.A. D. Alcolea declares a filed patent application (WO2019175379 A1 Markers of synaptopathy in neurodegenerative disease). I. Illán-Gala reported receiving personal fees for service on advisory boards of UCB and Nutricia, and received speaker honoraria from Almirall, Esteve Pharmaceuticals S.A., Kern Pharma, Krka Farmacéutica S.L., Lilly, Nutricia, and Zambon S.A.U. All other authors report no disclosures relevant to the manuscript. Go to Neurology.org/N for full disclosures.