Facets of language performance in early‐onset and late‐onset Alzheimer's disease dementia

Abstract

INTRODUCTION

Early‐onset Alzheimer's disease dementia (EOAD) is characterized by more pronounced cognitive decline than late‐onset AD dementia (LOAD). Characteristic performance in spoken language remains undefined.

METHOD

A cross‐sectional analysis of 1189 people with EOAD and 4646 with LOAD from the National Alzheimer's Coordinating Center (NACC) was conducted.

RESULT

Based on data from their first NACC visit with AD, there was considerable heterogeneity in language performance across people with EOAD and LOAD. The distribution of naming ability was similar across these groups. On average, people with LOAD performed better than those with EOAD in category fluency, letter fluency, and spoken lexical retrieval, and had lower Clinical Dementia Rating (CDR) Language scores, although there was considerable overlap in the distributions for participants with EOAD and those with LOAD.

DISCUSSION

At diagnosis, the language profiles of EOAD and LOAD are distinct. There is substantial variability in both groups in multiple aspects of language.

Highlights

• Early‐onset Alzheimer's disease (EOAD) is associated with significantly poorer category and phonemic fluency and global spoken lexical retrieval compared to late‐onset Alzheimer's disease (LOAD) at time of diagnosis.

• Participants with EOAD dementia show greater severity and variability in clinician‐rated language functioning, as measured by Clinical Dementia Rating (CDR) Language scores.

• No significant group differences were observed in confrontation naming performance between EOAD and LOAD dementia.

• Findings support that there are distinct profiles of language performance in EOAD and LOAD at time of dementia diagnosis.

1. INTRODUCTION

Alzheimer's disease (AD) pathology is the most common cause of dementia, accounting for over half of cases worldwide. ¹ AD dementia is a clinical syndrome characterized by multi‐domain cognitive decline, including memory, accompanied by acquired functional impairment in a major life category. ² , ³ , ⁴ Two subtypes are identified by age at symptom onset: early‐onset (EOAD) and late‐onset (LOAD) AD dementia. While EOAD and LOAD are defined by objective impairments in the memory domain, changes to other cognitive domains are frequently present at diagnosis. The clinical presentation of LOAD is increasingly recognized as heterogeneous, ⁵ , ⁶ with emerging evidence for distinct cognitive subgroups demarcated by relative impairments across domains. ⁷ , ⁸ , ⁹ Similar heterogeneity at the time of dementia diagnosis is also appreciated in people with EOAD. ⁵ , ¹⁰ , ¹¹

The impact of AD dementia on the language domain has received growing attention. Subtle changes to language, such as to lexical retrieval (e.g., naming ⁶ and verbal fluency ¹² ), linguistic complexity (e.g., word frequency and grammatical structure ¹³ ), and linguistic understanding (e.g., written or auditory comprehension ¹⁴ ), may emerge early in AD dementia. Spoken language performance—evaluated through measures of naming, verbal fluency, and spontaneous speech—is indicative of cognitive decline and clinically differentiates between mild cognitive impairment (MCI) and dementia. ¹² , ¹⁵ , ¹⁶ , ¹⁷

Characterizations of language performance differences between EOAD and LOAD remain mixed. Certain studies report greater impairment for individuals with EOAD for elements of verbal learning, ¹⁸ comprehension, ¹⁹ writing, ²⁰ and verbal letter fluency, ²¹ with more rapid decline on average. ¹⁰ , ¹¹ , ¹⁹ , ²⁰ , ²² Other investigations support that individuals with LOAD present with greater decline across multiple language subdomains, ⁶ , ¹⁴ including semantic knowledge, ²³ confrontation naming ability, ¹⁹ , ²¹ , ²² , ²⁴ and verbal fluency. ²⁵ Others report inconsequential baseline differences in language performance on average between people with EOAD and people with LOAD. ¹¹ , ²⁶ These discrepancies may stem from methodological challenges, such as limited sample sizes, single‐site data, or the reliance on coarse screening tools such as the Mini‐Mental State Examination (MMSE) ²⁷ or the Montreal Cognitive Assessment (MoCA) ²⁸ , which capture language performance to a limited extent.

Consequently, considerable gaps remain in the scientific understanding of various aspects of language in AD dementia. This study addresses these gaps by characterizing facets of language performance in a large, well‐phenotyped sample of participants with EOAD or LOAD, specifically targeting individuals with a diagnosis of amnestic AD. We focus on granular elements of spoken lexical generation through core tasks of neuropsychological assessment (e.g., naming and verbal fluency). Naming and fluency are imperfect, but representative, tasks of spoken language, a central function of daily communication with direct implications for potential behavioral intervention in the absence of curative pharmacological treatment. ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ In this work, our aim is to characterize and compare facets of nuanced language performance in amnestic EOAD and LOAD. If EOAD and LOAD are phenotypically the same, differentiated only by age of symptom‐onset, then we would expect language performance to present correspondingly across groups. Conversely, if the established genetic and biological differences between EOAD and LOAD have direct implications for behavior, we would expect language performance to differ accordingly.

This study leverages the extraordinarily rich data resources of the National Alzheimer's Coordinating Center (NACC) dataset, ³⁶ representing a well‐characterized and multicenter sample of individuals living with AD. The dataset is uniquely suited for this analysis due to its breadth, rigorous diagnostic procedures, standardized collection by trained clinicians, and its large representation of EOAD—a population often underrepresented in large‐scale studies

These strengths position the current work to provide novel insight into characteristic language performance in EOAD and LOAD, aligning with ongoing efforts to advance dementia subtyping and characterization.

2. MATERIALS AND METHODS

2.1. Participants

Participant data were drawn from the NACC Uniform Data Set 3.0 Neuropsychological Battery (UDS3‐NB) ³⁶ . There were 26,157 participants in UDS3‐NB with any language items (66,112 records). Individuals who did not have AD as a primary or secondary etiology (variable: naccalzd) were excluded from the dataset. From the remaining 10,017 participants, 4148 for whom the variable naccudsd was not “Dementia” were removed. For these 5869 participants, only the record that corresponded to the first visit at which a diagnosis of amnestic AD was formulated were kept. In other words, participants with diagnoses of atypical clinical syndromes that may also arise due to AD pathology, such as primary progressive aphasia or posterior cortical atrophy, were excluded from this dataset. Finally, 34 participants with a global CDR ³⁷ score of 0 (a categorical score denoting normal functional performance across domains of memory, orientation, judgment and problem‐solving, community affairs, home and hobbies, and personal care) were removed from the dataset. We categorized participants as EOAD if they were younger than 65 at the time of their first visit with a diagnosis of AD in UDS1, UDS2, or UDS3. Our classification resulted in a participant sample of 1189 EOAD and 4646 LOAD (see Table 1).

TABLE 1.

NACC participant characteristics

Parameter	EOAD (n = 1189)	LOAD (n = 4646)	Total (n = 5835)	p‐Value *
Age (years)	58.8 (4.9)	78.1 (7.4)	74.2 (10.4)	< 0.001
Female	661 (55.6%)	2395 (51.5%)	3056 (52.4%)	0.013
Education				< 0.001
≤ High school	278 (23.6%)	1094 (23.7%)	1372 (23.7%)
Some/all college	582 (49.4%)	1843 (40.0%)	2425 (41.9%)
Post college	317 (26.9%)	1676 (36.3%)	1993 (34.4%)
Race †
American Indian or Alaska Native	9 (0.8%)	21 (0.5%)	30 (0.5%)	0.001
Asian	24 (2.0%)	117 (2.5%)	141 (2.4%)
Black or African American	65 (5.5%)	449 (9.8%)	514 (8.9%)
Native Hawaiian or Other Pacific Islander	5 (0.4%)	5 (0.1%)	10 (0.2%)
Other	14 (1.2%)	70 (1.5%)	84 (1.5%)
White	1055 (90.0%)	3942 (85.6%)	4997 (86.5%)
Any APOE ε4 alleles	711 (59.8%)	2515 (54.1%)	3226 (55.3%)	< 0.001
Age of initial cognitive decline, clinician assessment (years)				<0.001
< 50	165 (13.9%)	6 (0.1%)	171 (2.9%)
50–59	845 (71.4%)	113 (2.4%)	958 (16.4%)
60–64	173 (14.6%)	598 (12.9%)	771 (13.2%)
65–74	0 (0.0%)	2153 (46.4%)	2153 (37.0%)
75+	0 (0.0%)	1772 (38.2%)	1772 (30.4%)
CDR Sum of Boxes	5.7 (3.4)	5.5 (3.2)	5.6 (3.2)	0.201
CDR Language				< 0.001 ±
0	415 (34.9%)	2153 (46.3%)	2568 (44.0%)
0.5	369 (31.0%)	1674 (36.0%)	2043 (35.0%)
1	267 (22.5%)	620 (13.3%)	887 (15.2%)
2	129 (10.8%)	184 (4.0%)	313 (5.4%)
3	9 (0.8%)	15 (0.3%)	24 (0.4%)
GSLR a	−0.15 (1.10)	−0.00 (1.00)	−0.03 (1.02)	<0.001
Animals	10.7 (5.5)	11.2 (5.2)	11.1 (5.2)	0.002
Vegetables	6.3 (3.9)	6.8 (3.7)	6.7 (3.8)	<0.001
Category fluency a	−0.13 (1.09)	0.00 (1.00)	−0.03 (1.02)	<0.001
F fluency	8.8 (5.1)	9.9 (4.9)	9.7 (4.9)	<0.001
L fluency	8.2 (4.9)	9.2 (4.8)	9.0 (4.8)	<0.001
Letter fluency a	−0.24 (1.08)	−0.00 (1.00)	−0.05 (1.02)	<0.001
MINT	24.3 (6.9)	24.2 (6.9)	24.2 (6.9)	0.396
MoCA Naming	2.4 (0.9)	2.3 (0.9)	2.4 (0.9)	<0.001
Naming a	0.11 (1.01)	0.00 (1.00)	0.02 (1.00)	<0.001

Abbreviations: APOE, apolipoprotein E; CDR, Clinical Dementia Rating; EOAD, early‐onset Alzheimer's disease; GSLR, global spoken lexical retrieval; LOAD, late‐onset Alzheimer's disease; MINT, Multilingual Naming Test; MoCA, Montreal Cognitive Assessment; NACC, National Alzheimer's Coordinating Center

^a Standardized to mean 0, standard deviation (SD) 1 in first AD visit age 65 or later.

^* Wilcoxon rank sum tests for continuous variables, Fisher's exact test for categorical.

^† p‐Value is 0.001 for White versus all others.

^± p‐Value is for categories 2 and 3 combined.

RESEARCH IN CONTEXT

1. Systematic review: We reviewed the literature using traditional sources (e.g., PubMed). Many investigations characterize early‐onset Alzheimer's disease dementia (EOAD) as being associated with worse performance on neuropsychological assessment at diagnosis than the late‐onset (LOAD). Many large‐scale investigations of language in these populations treat language as a monolithic construct. Characteristic performance of specific aspects of language such as spoken language performance remains to be established in EOAD and LOAD.

2. Interpretation: We present a large‐scale comparison of multiple facets of language performance, including spoken language, between individuals with EOAD and LOAD. On average, performance of individuals with EOAD is distinctive from performance of those with LOAD at the first NACC visit on tests of verbal fluency and lexical retrieval but not naming, though there was considerable heterogeneity in both people with EOAD and LOAD. It may be useful to characterize multiple aspects of language for individuals with EOAD and LOAD.

3. Future directions: Characterization of multiple aspects of language performance at the time of AD diagnosis, including a specific focus on spoken communication, may be beneficial for people with EOAD and LOAD. Interventions to improve functional communication for individuals with EOAD and LOAD and the people who care for them may have important benefits in terms of quality of life, neuropsychiatric complications, and caregiver burden. Future research should center on individuals with language profiles indicative of potential problems with spoken language and functional communication.

2.2. Harmonized language score

We implemented the harmonization workflow as stated in Mukherjee et al. ³⁸ to harmonize the language domain in NACC. Briefly, qualified neuropsychologists and behavioral neurologists categorized NACC test items (UDS1, UDS2, and UDS3) into one of the following domains: memory, executive functioning, language, visuospatial, or none of these. The analytic team evaluated each NACC test item with the cognitive specialist panel to ensure administration and scoring are equivalent for anchor items. Items were treated as categorical and recoded to a maximum of 10 categories as needed. Overlapping test items across NACC and our item bank (data derived from additional previously harmonized and co‐calibrated data sets) were treated as anchor items. Test items served as indicators in a confirmatory factor analysis (CFA) model, with all anchor item parameters fixed and the non‐overlapping test items freely estimated. The CFA model was run on the most recent visit for each individual for a given data freeze (NACC freeze April 2024) to obtain item parameters (factor loadings and thresholds) for unique NACC items. These item parameters were applied to the longitudinal data set (e.g., all visits, not just the most recent visit) to obtain factor scores for global spoken lexical retrieval (GSLR) (Figure S1). The language items included in the tests implemented in UDS3 are listed in Table 2.

TABLE 2.

Language items in NACC UDS3‐NB

Item name	Item description	Domain subtype
animals	Tell me all the animals you can think of in 1 min	Category fluency
vegetables	Tell me all the vegetables you can think of in 1 min	Category fluency
udsverfc	Number of correct F‐words generated in 1 min	Letter fluency
udsverlc	Number of correct L‐words generated in 1 min	Letter fluency
minttots	Multilingual naming test—total score	Naming
mocanami	MoCA—naming (lion, rhino, camel)	Naming
mocarepe	MoCA—repetition	Repetition

Abbreviations: MoCA, Montreal Cognitive Assessment; NACC, National Alzheimer's Coordinating Center; UDS3‐NB, Uniform Data Set 3.0 Neuropsychological Battery.

2.3. Domain subtype‐specific scores

We constructed three subdomain specific scores: (1) category fluency (animals and vegetables); (2) letter fluency (F‐fluency, L‐fluency); and (3) naming (Multilingual Naming Test [MINT] ³⁹ and MoCA ²⁸ naming tasks). We excluded repetition from this step in the analyses, as there was a single MoCA item for this secondary domain. We employed a weighted average approach using standardized item responses. Prior to score computation, all items were standardized to ensure comparability across measures. The weights applied in the averaging process were derived from the standardized factor loadings obtained through CFA models calibrated on our item bank. This method allowed us to account for the relative contribution of each item to the underlying construct, thereby producing subdomain scores that more accurately reflect the latent language abilities being assessed (see Table 3).

TABLE 3.

Range and standardized loadings for language subdomain items

UDS3 items by subdomain	Original range	Standardized factor loading
Category fluency
Animals	0–54	0.895
Vegetables	0–60	0.879
Letter fluency
F	0–40	0.539
L	0–40	0.714
Naming
MINT total score	0–32	0.788
MoCA Naming	0–3	0.74

Abbreviations: MINT, Multilingual Naming Test; MoCA, Montreal Cognitive Assessment; UDS3, Uniform Data Set 3.0.

2.4. Statistical methods

Univariate comparisons of EOAD and LOAD were tested with Wilcoxon Rank Sum tests for continuous outcomes and Fisher's exact test for categorical outcomes. To describe the effect of age of onset, CDR Sum of Boxes (rated from 0 to 18 on a continuous scale, summing performance across the six domains evaluated in the global CDR), demographics, and apolipoprotein E (APOE) genotype (coded as ≥ 1 ε4 allele vs. 0 ε4 alleles) on language scores, we used linear regression, with robust standard errors because the residuals in the naming model were skewed.

In addition to the CDR Language score ⁴⁰ in the NACC dataset (clinician‐rated language ability based on a combination of informal and standardized assessment, scored on the same scale as the global CDR ³⁷ ), we have three subdomain scores for category fluency, letter fluency, and naming. We wanted to assess which of these measures were independently associated with EOAD when all four measures were in a single model. We used Poisson regression, which better estimates the relative risk than logistic regression does, modified as in Zou ⁴¹ because the EOAD versus LOAD outcome is binary. Models were adjusted for the CDR Sum of Boxes, sex, education, race, and the presence of any APOE ε4 alleles.

Regression assumptions were tenable in all models. We conducted sensitivity analyses using the same procedures by restricting the sample to participants biomarker‐confirmed AD, defined as having both abnormally low amyloid and abnormally elevated Tau or pTau in the cerebrospinal fluid (CSF) or abnormally elevated amyloid on positron emission tomography (PET) (defined as either variables AMYLCSF and CSFTAU or AMYLPET alone equating “yes” or “1”; EOAD = 401, LOAD = 738). Additional sensitivity analyses excluded the 45 individuals with any family evidence for a dominantly inherited AD mutation.

3. RESULTS

There were univariate differences between EOAD and LOAD for all the demographic characteristics, APOE ε4 genotype, and all the language measures except the MINT (Table 1). Education was fairly evenly matched; over 75% of both groups received a college degree or graduate education (see Table 1). The majority of participants with LOAD and those with EOAD were white, with somewhat more racial diversity in the LOAD sample. More participants with EOAD had ≥ 1 APOE ε4 allele. At their first NACC visit with AD dementia, 34% of the participants with EOAD received scores of mild to severe impairment (1–3) for CDR Language, almost double that of participants with LOAD (18%; see Figure 1). For both groups, close to one‐third of scores corresponded to questionable impairment on the CDR scale (0.5); this was slightly higher for participants with LOAD. Close to half of the participants with LOAD had normal language performance (CDR Language = 0).

FIGURE 1.

Global CDR and CDR Language scores in EOAD and LOAD. CDR, Clinical Dementia Rating; EOAD, early‐onset Alzheimer's disease; LOAD, late‐onset Alzheimer's disease

Linear regression models for our language scores found statistically significantly lower GSLR, category fluency, and letter fluency scores, and higher naming scores for participants with EOAD (Table 4), despite there being a nonsignificant difference in CDR Sum of Boxes between groups. The largest difference was observed for letter fluency, where those with EOAD were nearly a quarter SD lower. The models in Table 5 also include the effect of the CDR Language scores. Higher CDR Language scores (i.e., worse performance) were associated with lower scores (also worse performance) in GSLR, category fluency, and letter fluency in all models. In addition, when including CDR Sum of Boxes in the regression models, the differences between EOAD and LOAD were attenuated for GSLR, category fluency, and letter fluency; only letter fluency remained statistically significant. The difference in naming was stronger when including the CDR Sum of Boxes in the regression models, nearly a quarter of a SD higher in EOAD.

TABLE 4.

Linear regression models for GSLR, category fluency, letter fluency, and naming

Parameter	GSLR		Category fluency		Letter fluency		Naming
	Beta	95% CI	Beta	95% CI	Beta	95% CI	Beta	95% CI
AD before age 65	−0.13	(−0.19, −0.07)	−0.14	(−0.20, −0.07)	−0.23	(−0.30, −0.17)	0.11	(0.05–0.18)
CDR Sum of Boxes	−0.15	(−0.16, −0.14)	−0.16	(−0.17, −0.15)	−0.11	(−0.12, −0.10)	−0.12	(−0.13, −0.11)
Female	0.01	(−0.04,0.05)	0.12	(0.07–0.17)	0.11	(0.06–0.16)	−0.29	(−0.34, −0.24)
Education
Some/all college	0.14	(0.09–0.20)	0.01	(−0.05,0.07)	0.32	(0.25–0.38)	0.22	(0.15–0.29)
Post college	0.20	(0.13–0.26)	0.01	(−0.05,0.08)	0.48	(0.40,0.55)	0.24	(0.17–0.31)
White	0.20	(0.14–0.26)	0.07	(−0.00,0.13)	0.20	(0.12–0.27)	0.34	(0.26–0.42)
Any APOE ε4 alleles	0.08	(0.03–0.12)	0.04	(−0.01,0.09)	0.13	(0.08–0.18)	0.06	(0.01–0.11)
Observations	5790		5432		5147		4729

Abbreviations: AD, Alzheimer's disease; APOE, apolipoprotein E; CDR, Clinical Dementia Rating; CI, confidence interval; GSLR, global spoken lexical retrieval.

TABLE 5.

Linear regression models for GSLR, category fluency, letter fluency, and naming, additionally including the CDR Language score

Parameter	GSLR		Category fluency		Letter fluency		Naming
	Beta	95% CI	Beta	95% CI	Beta	95% CI	Beta	95% CI
AD before age 65	0.02	(−0.04,0.07)	−0.01	(−0.07,0.05)	−0.14	(−0.20,‐0.07)	0.24	(0.18–0.30)
CDR Sum of Boxes	−0.10	(−0.11,‐0.09)	−0.12	(−0.12,‐0.11)	−0.08	(−0.09,‐0.07)	−0.08	(−0.09,‐0.07)
CDR Language
0.5	−0.27	(−0.32,‐0.23)	−0.25	(−0.30,‐0.20)	−0.15	(−0.21,‐0.09)	−0.18	(−0.23,‐0.13)
1	−0.81	(−0.88,‐0.75)	−0.72	(−0.80,‐0.65)	−0.49	(−0.57,‐0.41)	−0.70	(−0.79,‐0.61)
2 or 3	−1.27	(‐1.39,‐1.15)	−1.26	(‐1.39,‐1.13)	−1.10	(‐1.24,‐0.96)	−1.30	(‐1.48,‐1.11)
Female	−0.03	(−0.07,0.02)	0.09	(0.05–0.14)	0.09	(0.04–0.14)	−0.32	(−0.37,‐0.27)
Education
Some/all college	0.15	(0.10–0.20)	0.01	(−0.04–0.07)	0.32	(0.25–0.38)	0.23	(0.17–0.30)
Post college	0.22	(0.16–0.27)	0.03	(−0.03,0.10)	0.48	(0.42–0.55)	0.27	(0.20–0.34)
Race (% White)	0.24	(0.18–0.29)	0.10	(0.03–0.16)	0.22	(0.15–0.29)	0.36	(0.28–0.44)
Any APOE ε4 alleles	0.05	(0.01–0.10)	0.02	(−0.03,0.06)	0.11	(0.06–0.16)	0.03	(−0.02,0.08)
Observations	5790		5432		5147		4729

Abbreviations: AD, Alzheimer's disease; APOE, apolipoprotein E; CDR, Clinical Dementia Rating; CI, confidence interval; GSLR, global spoken lexical retrieval.

We also examined which of the CDR Language ratings and domain‐specific language scores differentiated EOAD from LOAD when all four were in a single model. Three of the four retained statistical significance. Compared to those with LOAD, individuals with EOAD had worse CDR Language and letter fluency scores, but better naming scores (see Table 6). Of note, these differences were observed even though we adjusted for overall dementia severity by including a term for the CDR Sum of Boxes. Differences in category fluency were not significant.

TABLE 6.

Modified Poisson regression for AD before age 65 (IRR, and 95% CI), adjusted for CDR Sum of Boxes, sex, education, race, and any APOE ε4 alleles

Variable	IRR	95% CI
CDR Language
0.5	1.15	(1.00–1.32)
1	2.24	(1.91–2.61)
2 or 3	2.99	(2.34–3.83)
Category fluency	0.96	(0.88–1.04)
Letter fluency	0.85	(0.79–0.91)
Naming	1.36	(1.25–1.47)
Observations	4577

Abbreviations: AD, Alzheimer's disease; APOE, apolipoprotein E; CDR, Clinical Dementia Rating; CI, confidence interval; IRR, incidence rate ratio.

Sensitivity analyses restricted to participants with biomarker‐confirmed AD generally confirmed the primary analyses. Without adjustment for CDR Language, differences were attenuated for GLSR and category fluency (Table S1a) but of similar magnitude when adjusted for CDR Language (Table S1b). In the examination of the relative effects of the language measures, most estimates were similar; only the CDR Language effects were reduced (Table S1c). Sensitivity analyses excluding the 45 individuals with any family evidence for a dominantly inherited AD mutation were quite similar to the primary analyses (Table S2a–c).

4. DISCUSSION

This study examined characteristic performance in several aspects of language, including spoken language, in individuals with EOAD and LOAD. We measured language through the use of standardized assessments and the CDR Language score, derived from interviews with informants and provider ratings. We found that participants with EOAD had lower language scores on average than participants with LOAD for category fluency, letter fluency, and GSLR, and higher scores for confrontation naming ability. Furthermore, lower letter fluency, higher confrontation naming ability, and higher CDR Language scores were independently associated with EOAD and observed when adjusting for overall dementia severity. Attenuation was only observed for category fluency when the CDR Language scores and domain‐specific language scores were combined in a single model. The finding that participants with EOAD performed better on letter than category fluency is a departure from previous findings. ⁴² , ⁴³ , ⁴⁴ This outcome further underscores the greater atypicality of language performance in people living with EOAD and its clinical relevance in diagnostic formulation.

Several previous studies have shown differences in speech, language, and communication in early‐onset dementias, characterizing rarer forms of language‐led syndromes (e.g., the logopenic variant of primary progressive aphasia) as early‐onset rather than late‐onset. Notably, prior detailed investigations of cognitive subdomains, including language, in individuals with EOAD and those with LOAD are of a smaller scale and restricted to single study sites. ⁶ Hammers et al. ⁶ provided a larger multi‐site comparison of cognitive profiles in EOAD and LOAD through the Alzheimer's Disease Neuroimaging Initiative (ADNI) ⁴⁵ and Longitudinal Early‐Onset Alzheimer's Disease Study cohorts (LEADS) ⁴⁶ . Their findings suggest that, despite previous findings indicating worse performance in non‐amnestic domains in EOAD, language may be more impaired in LOAD, with substantial variability in both groups. Based on the averaged raw scores alone, participants with EOAD had worse scores on average than those with LOAD for category fluency, similar scores on average compared to those with LOAD for confrontation naming and story recall, and better scores on average compared to those with LOAD for word recognition. For the composite language score, however, participants with LOAD had lower scores on average than people with EOAD overall. In contrast, an independent meta‐analysis of 42 studies on the cognitive profiles of EOAD and LOAD by Seath et al. ⁴⁷ found that participants with EOAD differ significantly from those with LOAD at baseline: a greater proportion of participants with early‐onset dementia exhibited atypical presentations with additional impairments to non‐amnestic domains such as language and a greater overall severity of symptoms. While data from 21,856 patients were synthesized for Seath et al.’s ⁴⁷ work (n = 5544 for EOAD), all cognitive domains were evaluated through the MMSE. Of note, while six MMSE items assess language directly, only two (confrontation naming of three pictures and repetition of the phrase “No ifs, ands or buts”) specifically address spoken language.

Nuanced characterization and documentation of language performance in EOAD and LOAD has direct clinical implications beyond diagnostic differentiation. Targeted behavioral intervention is paramount to enhance, maintain, or compensate for communication in people living with AD dementia, particularly spoken language. ²⁹ , ³⁰ , ³¹ Speech and language therapy is the primary intervention shown to slow communication decline ³² , ³³ , ³⁴ and to enhance life participation and well‐being for people living with dementia. ³⁰ , ³¹ , ³³ , ³⁵ Therefore, accurate characterization of language performance is critical. We used tasks of naming and verbal fluency to address this issue, assessments that are widely used clinically and in research. ¹² , ¹⁵ , ¹⁶ , ¹⁷ Confrontation naming involves retrieving and producing words to identify objects of concepts, typically elicited through picture‐naming prompts. Verbal fluency requires generating a collection of words that meet phonological or categorical rules. Both tasks draw upon cognitive functions beyond language, such as executive function. Despite this limitation, these tasks remain representative of an important part of linguistic function.

Our analyses build on prior work and examine multiple features of language, including spoken lexical retrieval, to further evaluate symptomatic features of amnestic EOAD and LOAD. Individuals with EOAD demonstrated more severe language impairment at diagnosis and greater heterogeneity than those with LOAD. Our results support that, on average, confrontation naming was higher in EOAD, but letter and category fluency were lower. GLSR scores and the CDR Language scores were also lower on average, in individuals with EOAD. Our work thus suggests that tests of naming alone are insufficient to capture the divergent linguistic profiles of EOAD and LOAD. Considering language as a monolithic construct, or even the subdomain of spoken lexical retrieval as such, may obscure important differences.

The CDR Language score, ascertained by specialist clinicians, integrates patient and care partner reports, observation, and clinical evaluation ⁴⁰ of discourse, speech mechanics, auditory comprehension, repetition, semantics, reading, and writing. We leveraged the global CDR to further classify phenotypic presentation. While the time of diagnosis is a proxy for identifying substantial changes in cognition, the time between symptom onset and diagnosis is highly variable. ⁴⁸ This could lead to reduced confidence in the EOAD and LOAD classifications. In our sample, the average age of symptom onset in participants classified as EOAD was 54.4 years (SD: 5.14), approximately 4 years before diagnosis (Table 1), with 99% reporting symptoms before age 63. For participants with LOAD, the average age at symptom onset was 72.4 years (SD: 7.63), preceding diagnosis by about six years, where only 15% reported symptoms before age 65. As such, we are confident in our age of onset group classifications. Comparing individuals at comparable stages in global clinical progression reduced variability in disease severity, and controlling for overall CDR Sum of Boxes allowed us to evaluate the specific role language in isolation from global cognition. This focus helped isolate differences attributable specifically to age of onset (i.e., EOAD vs. LOAD), rather than disease stage. One limitation of this approach is that functional impairment is evaluated in the context of a person's baseline performance, environment, lifestyle demands, and responsibilities—all of which are a function of age.

Our study leveraged a robust sample of individuals with EOAD and LOAD from a large, multi‐site cohort of incidental and prevalent AD from NACC. The standardized data collection across Alzheimer's Disease Research Centers (ADRCs), conducted by highly trained clinicians and research personnel with deep expertise in dementia diagnosis and assessment, ensures the clinical rigor of EOAD and LOAD classification, and ascertainment of language performance in a standardized fashion across all participant encounters. This methodological rigor allows us to move beyond the limitations of single‐clinician or single‐site case series, providing a much broader and more reliable foundation for examining language subdomains across AD onset types. Our study protocol in NACC allows for a more comprehensive evaluation of spoken language, with multiple items per subdomain. One limitation of this work is the lack of participant data prior to NACC enrollment: it is possible that some participants were misclassified as LOAD if they were diagnosed with AD elsewhere before age 65 but joined NACC after age 65. There was likely participant overlap with the LEADS and ADNI cohorts described in Hammers et al. ⁶ LEADS and ADNI study participants agree to additional research procedures, and these samples may be even less generalizable than those in the NACC dataset.

Our work shows that a substantial proportion of individuals with primary amnestic presentations of AD, regardless of age of onset, exhibit decline in spoken lexical retrieval function at the time of diagnosis. This decline is more pronounced and varied on average in individuals with EOAD than in individuals with LOAD. Both tasks of confrontation naming and verbal fluency—skills central to daily effective communication—showed variability within and across groups, where EOAD was characterized by greater heterogeneity and symptom impairment. The distinction goes beyond phenotype: clear differences in atrophy patterns between EOAD and LOAD have been presented in the literature, ⁶ with predominant posterior involvement and relative sparing of the medial temporal regions in EOAD and predominant hippocampal and medial temporal atrophy in LOAD. ⁶ , ⁴⁹ , ⁵⁰ EOAD has also been associated with greater amyloid buildup, tau deposition in cortical regions, and hypometabolism. ⁶ , ⁴⁹ , ⁵⁰ Each of these findings, including our own, indicate that EOAD and LOAD are differentiated by far more than age of symptom onset. Our work focused on behavioral aspects and revealed heterogeneity in both group classifications; thus, we cannot attribute differences in language performance to distinct regional distributions of atrophy, protein depositions, or metabolic activity. Substantial clinical implications remain, where an earlier onset can have profound consequences for lifestyle, occupation, dependents, and general responsibilities. Thorough characterization of distinct linguistic profiles in AD may therefore support early detection and effective intervention planning.

Our work here sheds light on how AD dementia may differentially affect communication in EOAD versus LOAD. Diminished language performance was common at the time of AD diagnosis and somewhat more prevalent in participants with EOAD, supporting that measures of language are crucial to characterize EOAD and LOAD populations, alongside memory. Our findings highlight the importance of communication‐focused assessment and intervention and underscore the need for ongoing research in these areas.

CONFLICT OF INTEREST STATEMENT

The authors have no relevant conflicts of interest or financial or other nonprofessional benefits to disclose that could bias the authors in the conduct of the reported work. Author disclosures are available in the Supporting Information.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Gallée J, Gibbons LE, Choi S‐E, et al. Facets of language performance in early‐onset and late‐onset Alzheimer's disease dementia. Alzheimer's Dement. 2025;21:e70705. 10.1002/alz.70705