Visual memory correlates with AD biomarkers in cognitively unimpaired individuals

Abstract

INTRODUCTION

Alzheimer's disease (AD) pathology evolves silently for many years prior to disease, and early identification of individuals at greatest risk is critical for improving outcomes.

METHODS

We conducted exploratory factor analysis (EFA) using neuropsychological testing of 1697 healthy adults. Factor scores were submitted as predictors for structural magnetic resonance imaging, functional connectivity, and cerebrospinal fluid (CSF) biomarkers.

RESULTS

The EFA model identified five latent factors. Factor 1, Verbal Memory, included high loadings from Rey Auditory Verbal Learning Test, and Factor 2, Visual Memory, included high loadings from Rey Complex Figure Test (RCFT). Factor 2 positively correlated with hippocampal volume, precuneus and posterior salience network connectivity, and negatively with CSF tau. Factor 1 correlated with hippocampal volume but not with network connectivity or CSF biomarkers.

DISCUSSION

Results demonstrate a significant relationship of visual memory with AD biomarkers. RCFT may be an important assessment for evaluating and tracking individuals during the earliest stages of AD pathology.

Highlights

• Exploratory factor analysis identified five latent cognitive domains.

• Both verbal memory and visual memory were correlated with hippocampal volume.

• Visual memory correlated with cerebrospinal fluid (CSF) tau levels and resting state functional magnetic resonance imaging connectivity.

• Verbal memory did not correlate with CSF biomarkers or network connectivity.

1. INTRODUCTION

Alzheimer's disease (AD) is a progressive neurodegenerative disorder, characterized by gradual cognitive decline, leading to loss of independence and ultimately death. The disease progresses along a continuum, beginning with an asymptomatic phase lasting up to two decades, during which pathophysiological changes are detectable but cognitive symptoms are not. Eventually, these changes can give rise to the symptomatic phases manifesting as mild cognitive impairment (MCI) and progressing to increasingly severe stages of dementia. The transition from asymptomatic to symptomatic stages represents a critical period for intervention; by the time symptoms are clinically observable, significant neuronal damage has already occurred. ¹ While current treatments show promise for slowing disease progression in individuals at early stages of disease, ² , ³ they are ineffective if administered after significant neurodegeneration has taken place. ⁴ Recognizing the need for earlier interventions, research has increasingly focused on identifying biomarkers and imaging techniques capable of detecting AD pathology in the asymptomatic phase, allowing for interventions that could delay or prevent clinical disease.

The progression of AD pathology has been well documented, with amyloid beta (Aβ) plaques and neurofibrillary tau tangles being identified as core hallmarks of the disease. Aβ deposition precedes neocortical tau pathology, with amyloid accumulation beginning many years before the onset of cognitive impairment. ⁵ In the asymptomatic phase, elevated amyloid burden is detectable through positron emission tomography (PET) imaging, cerebrospinal fluid (CSF), or blood‐based biomarkers, despite an absence of clinical symptoms. ⁶ After amyloid deposition, hyperphosphorylated tau proteins begin to form tangles, leading to neuronal damage, brain atrophy, and eventually cognitive impairment. ⁷ , ⁸ The presence of these pathological changes in asymptomatic individuals suggests a lengthy window during which early intervention could, in theory, delay or prevent progression to symptomatic AD. ¹

Amyloid PET imaging and CSF assays for Aβ and tau are the two best‐established tools for identifying amyloid pathology in non‐demented individuals. ⁹ Additionally, presence of the apolipoprotein E (APOE) ε4 allele can identify those at heightened genetic risk. ¹⁰ Neuroimaging methods assessing brain structure and connectivity, such as structural magnetic resonance imaging (MRI) and functional MRI (fMRI), have also been explored as predictors of AD progression. Structural MRI can detect early atrophy in regions like the hippocampus and entorhinal cortex, key sites of early neurodegeneration in AD, ¹¹ , ¹² , ¹³ while fMRI can reveal disruptions in brain network connectivity that precede cognitive impairment. ⁶ , ¹⁴

In parallel with the development of biomarker‐based frameworks, considerable effort has been directed toward identifying sensitive cognitive measures that can detect early signs of AD‐related decline in asymptomatic individuals. One prominent example is the Preclinical Alzheimer Cognitive Composite (PACC), ¹⁵ which combines episodic memory, executive function, and global cognition to track subtle cognitive decline prior to symptom onset. The PACC has since been widely adopted in preclinical AD trials and shown to be sensitive to amyloid‐related changes in cognitively normal individuals. ¹⁶ , ¹⁷ In addition to composite measures, longitudinal studies have explored the absence of learning curves or diminished practice effects as potential early indicators of progression to MCI or AD. ¹⁸ , ¹⁹ These efforts underscore the potential value of neuropsychological testing for detecting incipient AD‐related decline.

The goal of the current study is to determine whether neuropsychological performance in different cognitive domains among cognitively unimpaired individuals can predict early markers of AD (i.e., reduced brain volume in key regions, disruptions in functional connectivity, and CSF levels of Aβ and tau). To accomplish this goal, we conducted an exploratory factor analysis (EFA) on baseline neuropsychological testing data collected on 1697 Emory Healthy Brain Study (EHBS) participants. ²⁰ The factor analysis allowed for the grouping of tests into cognitive domains. The factors representing these cognitive domains were then used as predictors in median regression models to evaluate their relationship with structural MRI, resting state functional connectivity, and CSF Aβ and tau.

RESEARCH IN CONTEXT

1. Systematic review: We searched the literature for studies assessing neuropsychological performance and Alzheimer's disease (AD) biomarkers in cognitively unimpaired adults. Few studies have assessed the sensitivity of visuospatial memory measures, such as the Rey Complex Figure Test (RCFT), to early AD‐related changes. Moreover, most prior work has focused on individuals with cognitive impairment, leaving a gap in knowledge about these relationships during the preclinical stage.

2. Interpretation: Our findings suggest that visuospatial episodic memory may serve as an early cognitive indicator of AD‐related changes in hippocampal volume, resting‐state functional connectivity, and cerebrospinal fluid tau. These results are in line with the existing literature cited in the paper.

3. Future directions: Longitudinal analyses should determine whether visuospatial episodic memory performance predicts subsequent biomarker changes and clinical progression. Broader use of the RCFT, which may be facilitated through simplified scoring mechanisms, can further test these relationships in larger diverse samples of asymptomatic AD.

2. METHOD

2.1. Participants

For a participant to qualify for EHBS, ²⁰ they must be between the ages of 50 and 75 with no preexisting diagnosis of MCI, dementia, or other cognitive impairment; be willing to travel to Emory University; and be able and willing to complete a lumbar puncture, brain MRI, and other study procedures. Exclusionary criteria for enrollment include being unable to speak or understand English; current use of anticoagulant medication; history of bleeding disorders, stroke or transient ischemic attack, congestive heart failure, multiple sclerosis, amyotrophic lateral sclerosis, Parkinson's disease, epilepsy, active human immunodeficiency virus, tuberculosis, hepatitis B or C virus not fully treated; schizophrenia; kidney disease requiring dialysis; solid organ transplant; and life‐threatening illnesses. The protocol was approved by the Emory Institutional Review Board (approval number: 2025P011608), and all participants provided written informed consent. The study was performed in accordance with the ethical standards in the 1964 Declaration of Helsinki.

Data used in the current analyses are from 1697 EHBS participants’ baseline visits, completed prior to March 17, 2025. Mean age of participants at their baseline visit was 63.0 ± 6.8 years and mean education was 16.6 ± 2.1 years, indicating most participants had at least a 4‐year degree. Other demographic characteristics (sex, race, and ethnicity) are presented in Table 1 along with APOE genotypes.

TABLE 1.

Participant demographics and APOE genotypes.

	N (%)
Sex
Female	1101 (64.9%)
Male	596 (35.1%)
Primary race
White	1366 (80.5%)
Black	280 (16.5%)
Asian	25 (1.5%)
American Indian or Alaska Native	9 (0.5%)
Native Hawaiian or Pacific Islander	3 (0.2%)
Other	13 (0.7%)
Unknown	1 (0.06%)
Ethnicity
Non‐Hispanic	1640 (96.6%)
Hispanic	53 (3.1%)
No Information	4 (0.2%)
APOE
0 ε4	1163 (68.5%)
1 ε4	482 (28.4%)
2 ε4	52 (3.1%)

Abbreviation: APOE, apolipoprotein E.

2.2. Procedures

Participants provided details on their medical history including current medications and completed several surveys including the Patient Health Questionnaire (PHQ‐8), ²¹ Generalized Anxiety Disorder scale (GAD‐7), ²² Cognitive Function Instrument (CFI), ²³ and Quick Dementia Rating System (QDRS). ²⁴ The EHBS neuropsychological battery consisted of the Montreal Cognitive Assessment (MoCA), ²⁵ Rey Auditory Verbal Learning Test (RAVLT), ²⁶ Rey Complex Figure Test (RCFT), ²⁷ , ²⁸ Number Span Test Forward (NSF) and Backward (NSB), ²⁹ Judgement of Line Orientation (JOLO), ³⁰ Controlled Oral Word Association Test (COWA), ³¹ Category Fluency for animals, ³² Multilingual Naming Test (MiNT), ³³ Trail Making Test (TMT) Parts A and B, ³⁴ and the National Institutes of Health (NIH) Toolbox Picture Vocabulary Test. ³⁵

Several changes in the neuropsychological battery impacted data availability. From April 2017 to August 2018, the Free and Cued Selective Reminding Test ³⁶ , ³⁷ was used in place of RAVLT. For COWA, F fluency was always collected as part of the MoCA; FAS fluency was used until March 2021, when it was replaced by FL fluency. TMT and the NIH Toolbox Picture Vocabulary Test were administered only during in‐person visits, and restrictions on in‐person activities during the COVID‐19 pandemic limited data collection.

MoCA was not used in the factor analysis, as it measures a variety of cognitive processes. NIH Toolbox Picture Vocabulary Test was not used, as 23% of participants were missing data, and MiNT was not used, as performance neared ceiling for most participants. See Table 2 for summary of neuropsychological tests.

TABLE 2.

Neuropsychological test scores.

	Mean	SD	Total possible score	% Missing
RAVLT A1–5	47.0	9.9	75	5.2
RAVLT A6	9.2	3.4	15	5.4
RAVLT A7	9.0	3.8	15	5.3
RAVLT Rec	12.2	2.4	15	5.3
NSF	6.9	1.3	9	0
NSB	5.3	1.3	8	0
JOLO	25.1	4.0	30	0.1
F Fluency	14.2	4.4	–	0.1
L Fluency	14.8	4.2	–	17.3
Category fluency	21.6	5.2	–	0
RCFT Copy	31.7	4.4	36	0.1
RCFT IR	17.8	6.6	36	0.4
RCFT DR	17.2	6.4	36	0.8
RCFT Rec	8.5	1.9	12	0.4
TMT A	33.9	11.8	300	0.7
TMT B	72.2	32.5	300	0.7
Tests not used in factor analysis
MoCA	26.5	2.5	30	0
MiNT	30.6	1.9	32	0
NIH Toolbox Vocab	116.2	9.1	–	23.0

Notes: % Missing indicates the percentage of participants missing data for each test. RAVLT 1–5, Total words across five learning trials; RAVLT A6, immediate recall (after distractor list); RAVLT A7, delayed recall (after 20 minute delay); RAVLT Rec, recognition discrimination (hits–false alarms); F Fluency, phonemic fluency for words starting with “F” (from the Montreal Cognitive Assessment); L Fluency, phonemic fluency for words starting with “L”; Category Fluency, semantic fluency for animals; RCFT Copy, copy trial; RCFT IR, immediate recall trial; RCFT DR, delayed recall trial (after 30 minute delay); RCFT Rec, recognition discrimination (hits–false alarms); TMT A (in seconds); TMT B (in seconds); NIH Toolbox Vocab, NIH toolbox picture vocabulary test uncorrected score.

Abbreviations: JOLO, Judgement of Line Orientation; MiNT, Multilingual Naming Test; MoCA, Montreal Cognitive Assessment; NIH, National Institutes of Health; NSB, Number Span Backward; NSF, Number Span Forward; RAVLT, Rey Auditory Verbal Learning Test; RCFT, Rey Complex Figure Test; SD, standard deviation; TMT A, Trail Making Test Part A; TMT B, Trail Making Test Part B.

2.3. Analyses

2.3.1. Statistical analysis

We analyzed the EHBS baseline data to explore the relationship of neuropsychological performance with MRI regional brain volumes, resting‐state functional connectivity, and with CSF biomarkers using R software (v4.3.1). Correlations of cognitive domain factor scores with outcome variables were evaluated with median regression models using the quantreg package (v5.96), ³⁸ while controlling for demographic variables (age, education, sex, and race) and presence or absence of an APOE ε4 allele. To correct for multiple comparisons, the Benjamini–Hochberg (B‐H) procedure was applied, and a B‐H critical value for a false discovery rate (FDR) of 0.20 was chosen. The raw p values were sorted and compared to their corresponding B‐H critical values for each factor score. The critical value was calculated as: (i/m)Q, where i represents the rank, m represents the number of comparisons, and Q represents the FDR.

2.3.2. Missing data mechanism

Because data were missing on some neuropsychological assessments, the missing data mechanism was diagnosed. There was no systematically missing data based on the data collection procedure and clinical judgement. The non‐parametric test of homoscedasticity across different patterns of missing data in the MissMech package (v1.0.3) ³⁹ was used to check if data were missing completely at random (MCAR). The hypothesis of MCAR was not rejected at 0.05 significance level (p = 0.443). We also checked whether missing data were related to demographic and cognitive variables. Compared to the group with no missing data, those missing data had lower education (16.2 ± 2.2 vs. 16.8 ± 2.0, p < 0.001). No other demographic variables (Ps > 0.08) nor any cognitive variables (Ps > 0.06) significantly differed between the groups.

To increase statistical power and obtain robust estimates accounting for uncertainty introduced by missing data, we performed multiple imputation on the raw data to predict missing values in each incomplete cognitive score based on observed data and demographic variables. The multiple imputation was conducted with multivariate imputation with chained equations (MICE) from the mice package (v3.16.0), ⁴⁰ then the covariance matrices of the imputed data sets were combined into a single covariance matrix using Rubin rules in the mifa package (v0.2.0). ⁴¹ Non‐parametric bootstrap confidence intervals were generated for the variance explained by different numbers of principal components. Thirty imputed datasets were created with age, sex, race, and education as predictors in the imputation models. The resulting covariance matrix was converted to a correlation matrix for the EFA.

2.3.3. EFA

We conducted EFA using the EFATools package (v0.4.4). ⁴² First, we used the Kaiser–Meyer–Olkin (KMO) test as a measure of sampling adequacy and the Bartlett test of sphericity to determine the factorability of the data. Next, the determinant score was calculated to examine multicollinearity among the variables. To determine the number of factors to retain, we used three methods: the Kaiser–Guttman rule (or Kaiser criterion; retain all factors with associated eigenvalues > 1), examination of the scree plot along with Raiche non‐graphical solution ⁴³ to identify the acceleration factor and optimal coordinate solutions, and parallel analysis using EFA and means decision rule. Finally, we performed EFA using principal axis factoring with a promax rotation to facilitate interpretation of results. An oblique rotation was selected, as factors were expected to correlate with one another. Variables were assigned to factors if their loadings exceeded 0.30 and loaded on a single factor. Any variable not meeting these criteria would be dropped from the model.

Bartlett factor scores were created to evaluate the relationships between the factors and MRI volumetric data, resting state functional connectivity, and CSF biomarkers (Aβ42, phosphorylated tau [p‐tau] 181, and total tau [t‐tau]). Linear regression was used for MRI volumetric and resting state functional connectivity models. Median regression was used in CSF models, as data were not normally distributed. CSF samples were assayed on the Roche Elecsys platform ⁴⁴ with standard pre‐processing procedures as described. ⁴⁵ Details on MRI acquisition, data preprocessing, and identification of functional networks are reported elsewhere. ⁴⁶ We used 14 functional networks from templates identified by Shirer et al. ⁴⁷ : auditory, basal ganglia, left (LECN) and right (RECN) executive control, language, precuneus, sensorimotor, visuospatial, anterior salience, posterior salience, dorsal and ventral default mode (DMN), higher visual, and primary visual. For the structural MRI data, we analyzed bilateral Automated Anatomical Labeling regions for the hippocampus and entorhinal cortex. The structural MRI data were corrected for the total intracranial volume (proportion of total brain volume).

3. RESULTS

3.1. Participant surveys

The mean PHQ‐8 score was 1.86 ± 2.30 and the mean GAD‐7 score was 1.41 ± 2.33, indicating participants had minimal self‐reported depressive and anxiety symptoms. The mean CFI score was 1.91 ± 2.0, the mean QDRS Clinical Dementia Rating Scale Sum of Boxes score was 0.29 ± 0.63, and the mean QDRS Global Score was 0.10 ± 0.20, indicating participants had minimal subjective cognitive complaints at their baseline visit.

3.2. EFA

Correlations between variables used in the factor analysis are presented in Table 3. While some strong correlations exist between variables (e.g., RAVLT immediate and delayed recall, r = 0.84, RCFT immediate and delayed recall r = 0.91), the determinant of the correlation matrix was 0.0006, suggesting multicollinearity was not a significant concern for these data. The overall KMO value was 0.831, and the Bartlett test of sphericity was significant, 𝜒²(120) = 12,535.09, p < 0.001, suggesting the data are suitable for factor analysis.

TABLE 3.

Correlations between neuropsychological test scores.

	RAVLT A1–5	RAVLT A6	RAVLT A7	RAVLT Rec	NSF	NSB	JOLO	F Flu	L Flu	Cat Flu	RCFT Copy	RCFT IR	RCFT DR	RCFT Rec	TMT A	TMT B
RAVLT A1–5	1	0.79	0.78	0.58	0.18	0.29	0.15	0.23	0.27	0.34	0.24	0.28	0.29	0.18	−0.23	−0.29
RAVLT A6		1	0.84	0.61	0.12	0.24	0.14	0.18	0.21	0.33	0.21	0.30	0.30	0.16	−0.22	−0.26
RAVLT A7			1	0.64	0.09	0.20	0.14	0.21	0.23	0.31	0.21	0.30	0.29	0.16	−0.22	−0.24
RAVLT Rec				1	0.04	0.16	0.09	0.14	0.17	0.22	0.18	0.23	0.24	0.14	−0.17	−0.18
NSF					1	0.45	0.16	0.17	0.25	0.13	0.08	0.11	0.12	0.06	−0.16	−0.21
NSB						1	0.23	0.19	0.23	0.23	0.17	0.20	0.19	0.12	−0.17	−0.27
JOLO							1	0.10	0.18	0.23	0.35	0.32	0.31	0.19	−0.27	−0.33
F Flu								1	0.57	0.35	0.10	0.12	0.13	0.07	−0.20	−0.25
L Flu									1	0.39	0.18	0.16	0.16	0.09	−0.24	−0.32
Cat Flu										1	0.20	0.24	0.26	0.16	−0.30	−0.33
RCFT Copy											1	0.48	0.47	0.24	−0.21	−0.31
RCFT IR												1	0.91	0.41	−0.25	−0.32
RCFT DR													1	0.38	−0.24	−0.33
RCFT Rec														1	−0.13	−0.22
TMT A															1	0.52
TMT B																1

Notes: RAVLT 1–5, Total words across five learning trials; RAVLT A6, immediate recall (after distractor list); RAVLT A7, delayed recall (after 20 minutes delay); RAVLT Rec, recognition discrimination (hits–false alarms); F Flu, phonemic fluency for words starting with “F” (from the Montreal Cognitive Assessment); L Flu, phonemic fluency for words starting with “L”; Cat Flu, category/semantic fluency for animals; RCFT Copy, copy trial; RCFT IR, immediate recall trial; RCFT DR, delayed recall trial (after 30 minute delay); RCFT Rec, recognition discrimination (hits–false alarms).

Abbreviations: JOLO, Judgement of Line Orientation; NSB, Number Span Backward; NSF, Number Span Forward; RAVLT, Rey Auditory Verbal Learning Test; RCFT, Rey Complex Figure Test; TMT A, Trail Making Test Part A; TMT B, Trail Making Test Part B.

Eigenvalues are presented in the scree plot in Figure 1. Using the Kaiser criterion (eigenvalues > 1), a five‐factor solution was indicated. We additionally conducted parallel analysis using 1000 simulated datasets, which likewise supported retention of five factors. Scree test decision rules based on Raiche methods yielded mixed results: the acceleration factor suggested a two‐factor solution, whereas the optimal coordinates method supported a five‐factor solution. Given convergence across the Kaiser criterion, parallel analysis, and the optimal coordinates approach, we retained five factors for our EFA model.

FIGURE 1.

Scree plot and cumulative variance explained. In the scree plot (A), the dotted line depicts the cutoff for the Kaiser rule (> 1) and the dashed line depicts the scree plot's optimal coordinates index. Five eigenvalues meet the criteria for the Kaiser rule, scree plot, and parallel analysis. B, Bar graph showing cumulative variance explained by each of the five factors.

We found the five‐factor model met our criteria of having every variable load onto a single factor with a loading > 0.30. The five‐factor model explained 56% of the total variance. Table 4 presents the factor loadings for the five‐factor model. The factors can be characterized by the underlying cognitive domains: Factor 1 = Verbal Memory (RAVLT learning, immediate recall, delayed recall, recognition discrimination), Factor 2 = Visual Memory (RCFT copy, immediate recall, delayed recall, recognition discrimination), Factor 3 = Fluency (F and L letter fluency, animal category fluency), Factor 4 = Working Memory (NSF, NSB), and Factor 5 = Visual Processing (JOLO, TMT A & B). Several factors were moderately to strongly correlated (see Table 5), suggesting an oblique rotation was appropriate for these data.

TABLE 4.

Factor loadings.

Variable	F1	F2	F3	F4	F5
RAVLT A1–5	0.82	−0.01	0.02	0.10	0.01
RAVLT A6	0.92	0.00	−0.04	0.01	0.01
RAVLT A7	0.94	0.00	0.01	−0.05	−0.02
RAVLT Rec	0.69	0.03	−0.01	−0.05	−0.01
RCFT Copy	0.02	0.43	−0.02	−0.01	0.21
RCFT IR	0.00	0.99	0.01	−0.01	−0.07
RCFT DR	0.00	0.96	0.02	−0.01	−0.05
RCFT Rec	0.03	0.38	−0.03	0.01	0.08
F Fluency	−0.02	−0.02	0.80	−0.03	−0.06
L Fluency	−0.03	−0.01	0.76	0.05	0.02
Category fluency	0.14	0.06	0.33	−0.02	0.22
NSF	−0.06	−0.03	0.04	0.68	−0.03
NSB	0.06	0.03	−0.03	0.69	0.00
JOLO	−0.07	0.22	−0.05	0.11	0.36
TMT A	−0.01	0.02	0.00	0.07	−0.88
TMT B	0.02	−0.04	0.00	−0.01	−0.76

Notes: RAVLT 1–5, Total words across five learning trials; RAVLT A6, immediate recall (after distractor list); RAVLT A7, delayed recall (after 20 minute delay); RAVLT Rec, recognition discrimination (hits–false alarms); RCFT Copy, copy trial; RCFT IR, immediate recall trial; RCFT DR, delayed recall trial (after 30 minute delay); RCFT Rec, recognition discrimination (hits–false alarms); F1, Factor 1 (Verbal Memory); F2, Factor 2 (Visual Memory); F3, Factor 3 (Fluency); F4, Factor 4 (Working Memory); F5, Factor 5 (Visual Processing).

Abbreviations: JOLO, Judgement of Line Orientation; NSB, Number Span Backward; NSF, Number Span Forward; RAVLT, Rey Auditory Verbal Learning Test; RCFT, Rey Complex Figure Test; TMT A, Trail Making Test Part A; TMT B, Trail Making Test Part B.

TABLE 5.

Factor intercorrelations.

Factors
	1	2	3	4	5
1	1	0.36	0.36	−0.39	−0.29
2		1	0.25	−0.46	−0.27
3			1	−0.51	−0.41
4				1	0.46
5					1

Note. Factor 1, verbal memory; Factor 2, visual memory; Factor 3, fluency; Factor 4, working memory; Factor 5, visual processing.

To verify that our imputed data did not influence the factor structure, we completed an EFA using only the 1369 participants who had scores on all cognitive variables. Table S1 in supporting information presents the factor loadings, and Table S2 in supporting information presents the factor correlations. The factor structure was virtually identical between the datasets with and without imputed values.

3.3. Relationship between factor scores and structural MRI, functional networks, and CSF biomarkers

Of the 1697 participants with factor scores and APOE data, 1257 (74.1%) had CSF data, 1208 (71.2%) had structural MRI data, and 825 (48.6%) had functional connectivity MRI data. All results are presented controlling for demographics and APOE genotype.

We found that Factor 1 and Factor 2 were positively associated with hippocampal volume (β = 0.12, 95% confidence interval [CI: 0.06, 0.17], p = 0.0001 and β = 0.11, [0.06, 0.17], p = 0.0001, respectively; Figure 2A). No other factors were associated with hippocampal volume (p > 0.321), and no factors were associated with entorhinal cortex volume (p > 0.238).

FIGURE 2.

Associations between factor scores and hippocampal volume (A), functional connectivity (B–D), and CSF p‐tau and t‐tau (F, G). Factor 1 (F1) = verbal memory, Factor 2 (F2) = visual memory, Factor 3 (F3) = fluency, Factor 4 (F4) = working memory, Factor 5 (F5) = visual processing. CI, confidence interval; CSF, cerebrospinal fluid; p‐tau, phosphorylated tau; t‐tau, total tau.

We found significant relationships between Factor 1 and connectivity in the basal ganglia network (β = 0.12, [0.04, 0.20], p = 0.003); between Factor 2 and connectivity in the precuneus (β = 0.08, [0.01, 0.15], p = 0.026) and posterior salience (β = 0.10, [0.03, 0.17], p = 0.006) networks; and between Factor 4 and connectivity in the posterior salience network (β = 0.11, [0.04, 0.18], p = 0.003). No other relationships were significant (all p > 0.056). These results are shown in Figure 2B–D.

Only Factor 2 was significantly associated with CSF levels of p‐tau (β = −0.07, [−0.86, −0.08], p = 0.014; Figure 2E) and t‐tau (β = −0.10, [−11.9, −2.02], p = 0.005; Figure 2F; all other factors: p > 0.089 for p‐tau, p > 0.062 for t‐tau). No factors were associated with Aβ42 levels (p > 0.091).

After multiple comparison adjustment with the B‐H procedure to control the FDR, significant associations presented survived B‐H correction at FDR = 0.20.

To determine which cognitive test scores from each factor made the strongest contributions in the significant relationships detailed above, we conducted follow‐up median regression tests using the cognitive scores as predictor variables, demographics and APOE genotype as covariates, and the hippocampal volume, functional connectivity, and CSF data as outcome variables. Median regression was used because cognitive scores were not normally distributed. The same FDR corrections were applied. Results are presented in Table 6. The relationship of Factor 2 to each significant outcome variable was driven primarily by RCFT immediate and delayed recall, with less significant contributions from copy and recognition.

TABLE 6.

Results of follow‐up tests.

	Test	β	95% CI	p	B‐H critical value
Hippocampal volume
Factor 1 (verbal memory)	RAVLT A1–5	0.061	[−0.024, 0.145]	0.158	0.20
	RAVLT A6	0.075	[0.003, 0.147]	0.043	0.15
	RAVLT A7	0.085	[0.014, 0.155]	0.019	0.10
	RAVLT Rec	0.118	[0.030, 0.207]	0.009	0.05
Factor 2 (visual memory)	RCFT Copy	−0.017	[−0.096, 0.063]	0.682	0.20
	RCFT IR	0.107	[0.036, 0.178]	0.003	0.10
	RCFT DR	0.107	[0.036, 0.179]	0.003	0.05
	RCFT Rec	0.054	[−0.003, 0.111]	0.066	0.15
Precuneus network functional connectivity
Factor 2 (visual memory)	RCFT Copy	0.036	[−0.037, 0.110]	0.334	0.20
	RCFT IR	0.096	[0.018, 0.174]	0.016	0.05
	RCFT DR	0.090	[0.005, 0.175]	0.039	0.15
	RCFT Rec	0.082	[0.013, 0.151]	0.020	0.10
Posterior salience network functional connectivity
Factor 2 (visual memory)	RCFT Copy	−0.002	[−0.101, 0.097]	0.971	0.20
	RCFT IR	0.107	[0.007, 0.208]	0.037	0.05
	RCFT DR	0.045	[−0.063, 0.153]	0.412	0.15
	RCFT Rec	−0.039	[−0.131, 0.052]	0.400	0.10
Factor 4 (working memory)	NSF	0.110	[0.027, 0.194]	0.010	0.10
	NSB	0.017	[−0.081, 0.115]	0.735	0.20
Basal ganglia network functional connectivity
Factor 1 (verbal memory)	RAVLT A1–5	0.125	[0.013, 0.237]	0.029	0.15
	RAVLT A6	0.118	[0.012, 0.223]	0.029	0.10
	RAVLT A7	0.134	[0.017, 0.250]	0.025	0.05
	RAVLT Rec	0.036	[−0.065, 0.137]	0.482	0.20
CSF p‐tau
Factor 2 (visual memory)	RCFT Copy	−0.001	[−0.073, 0.070]	0.968	0.20
	RCFT IR	−0.069	[−0.126, −0.012]	0.018	0.10
	RCFT DR	−0.067	[−0.120, −0.014]	0.014	0.05
	RCFT Rec	−0.026	[−0.089, 0.037]	0.418	0.15
CSF t‐tau
Factor 2 (visual memory)	RCFT Copy	−0.015	[−0.082, 0.051]	0.652	0.15
	RCFT IR	−0.094	[−0.155, −0.033]	0.003	0.10
	RCFT DR	−0.092	[−0.150, −0.034]	0.002	0.05
	RCFT Rec	0.0005	[−0.064, 0.065]	0.989	0.20

Notes: P values in bold text indicate those that are less than their respective Benjamini‐Hochberg (B‐H) critical values. RAVLT 1–5, Total words across five learning trials; RAVLT A6, immediate recall (after distractor list); RAVLT A7, delayed recall (after 20 minute delay); RAVLT Rec, recognition discrimination (hits–false alarms); RCFT Copy, copy trial; RCFT IR, immediate recall trial; RCFT DR, delayed recall trial (after 30 minute delay); RCFT Rec, recognition discrimination (hits–false alarms).

Abbreviations: CI, confidence interval; CSF, cerebrospinal fluid; NSB, Number Span Backward; NSF, Number Span Forward; p‐tau, phosphorylated tau; RAVLT, Rey Auditory Verbal Learning Test; RCFT, Rey Complex Figure Test; SD, standard deviation; TMT A, Trail Making Test Part A; TMT B, Trail Making Test Part B; t‐tau, total tau.

4. DISCUSSION

The objective of the current study was to determine whether performance on neuropsychological tests could predict early changes in AD biomarkers in cognitively normal individuals. Using EFA, we grouped neuropsychological test scores into distinct cognitive domains and examined their relationship with structural MRI, resting‐state functional connectivity, and CSF Aβ42, t‐tau, and p‐tau. Our major findings involved Factor 2, which consisted of scores on the RCFT. While the RCFT assesses visuospatial construction and episodic memory, performance can also be impacted by other abilities including motor execution and executive function involving planning and organization. Higher scores on this factor were significantly associated with larger hippocampal volume, greater functional connectivity in the precuneus and posterior salience networks, and lower CSF t‐tau and p‐tau levels. Follow‐up tests showed that RCFT immediate and delayed recall scores were the main drivers of these relationships. These findings suggest that RCFT, particularly the recall elements, may be sensitive to certain biomarkers that have been associated with disease development years before cognitive decline is manifest.

4.1. Structural MRI

We found that both Factor 1 (Verbal Memory) and Factor 2 (Visual Memory) were positively associated with larger hippocampal volume. The hippocampus is one of the earliest sites of neurodegeneration in AD, with volumetric reductions often preceding clinical diagnosis by years. ¹³ Our findings align with prior research showing that hippocampal atrophy is associated with verbal ⁴⁸ , ⁴⁹ and visual ⁵⁰ memory impairments in individuals at risk for AD. Both the RAVLT and RCFT require encoding, storage, and retrieval of detailed information—processes that are critically dependent on hippocampal integrity. Identification of a significant relationship of cognition with a neurodegeneration marker (i.e., hippocampal volume) in a cognitively unimpaired cohort was unexpected and may be driven by presence of individuals experiencing the very earliest stages of cognitive decline.

4.2. Resting state functional connectivity

Our findings showed that Factor 2 (Visual Memory) was associated with greater resting state functional connectivity in the precuneus and posterior salience networks. The precuneus plays a central role in episodic memory retrieval and visuospatial imagery, ⁵¹ and is among the earliest regions to accumulate amyloid and exhibit metabolic and connectivity disruptions in AD. ⁵² , ⁵³ Reduced precuneus network connectivity has been linked to CSF biomarkers of AD, even in asymptomatic individuals. ⁵⁴ Similarly, the posterior salience network, which supports attentional reorienting and integration of salient environmental stimuli, ⁵⁵ is also vulnerable in aging and AD progression. ⁵⁶ Our findings suggest that individuals with preserved visuospatial episodic memory exhibit stronger functional connectivity within these networks, potentially indicating resilience to early neurodegenerative changes.

We also observed significant associations between Factor 1 (Verbal Memory) and stronger connectivity in the basal ganglia network, and between Factor 4 (Working Memory) and stronger connectivity in the posterior salience network. Although the RAVLT is commonly conceptualized as a medial temporal lobe–dependent episodic memory task, performance also depends on learning efficiency, strategy use, and response selection processes supported by fronto‐striatal circuits. ⁵⁷ The observed association between Factor 1 and basal ganglia connectivity may therefore reflect the contribution of procedural and incremental learning mechanisms that may support verbal learning in cognitively unimpaired individuals. ⁵⁸ , ⁵⁹

The relationship between Factor 4 and posterior salience network connectivity is consistent with the attentional and executive demands of working memory tasks, which require the dynamic allocation of cognitive resources, maintenance of task‐relevant information, and suppression of interference. ⁶⁰ The posterior salience network is thought to mediate switching between large‐scale brain networks and support sustained attention during cognitively demanding tasks. Stronger connectivity within this network may therefore support efficient working memory performance in asymptomatic populations and may reflect compensatory or scaffolding mechanisms that help maintain cognitive function prior to overt AD‐related decline. ⁶¹ , ⁶²

4.3. CSF

An important aspect of our study was the relationship between cognitive performance and CSF biomarkers. We found that higher Factor 2 (Visual Memory) scores were associated with lower CSF t‐tau and p‐tau levels but not with Aβ42. Although Aβ42, t‐tau, and p‐tau are all considered early biomarkers of AD, prior studies suggest amyloid accumulation alone is insufficient to cause cognitive impairment. ⁶³ In contrast, abnormal tau accumulation appears to be more strongly associated with clinical progression in asymptomatic individuals at risk for AD. ⁶⁴ , ⁶⁵ , ⁶⁶ Similarly, our findings suggest that CSF t‐tau and p‐tau levels are more strongly associated with cognitive performance among unimpaired individuals than levels of Aβ42.

4.4. APOE genotype

APOE ε4 is the strongest genetic risk factor for late‐onset AD and strongly influences levels of Aβ42, t‐tau, and p‐tau in cognitively normal individuals across the lifespan. ⁶⁷ While APOE genotype was a significant covariate for all CSF biomarkers, the presence or absence of an ε4 allele was not associated with significant differences in any factor scores.

4.5. Conclusions and future directions

Overall, our results suggest that visual episodic memory, as evaluated by the RCFT, may be particularly sensitive to early AD biomarker changes within a cohort of cognitively unimpaired individuals. These findings are in line with previous studies. ⁶⁸ , ⁶⁹ It is noteworthy that identified associations of Factor 2 (Visual Memory) with CSF t‐tau and p‐tau levels and precuneus and posterior salience network connectivity was not found with Factor 1 (Verbal Memory). While episodic memory is broadly recognized and incorporated as an important measure in AD‐related studies and trials, this is most commonly assessed with verbal episodic memory tasks. Lack of use of the RCFT is likely due in part to the test being difficult and time consuming to score. Loring et al. ⁷⁰ developed a 10‐element recall scoring system and recognition task showing high correlation with the original scoring methodology while significantly reducing time burden. This simplified approach may lead to wider adoption of the RCFT, which may be more effective at identifying individuals earlier in the evolution of AD pathology.

While our study provides valuable insights into the relationship of specific cognitive domains with biomarkers of AD, several limitations should be noted. First, effect sizes were modest in magnitude, limiting explanatory power at the individual level at this early stage. Second, our analyses were cross‐sectional, limiting our ability to infer causality. Third, the sample was predominantly White, female, and highly educated, thereby limiting generalizability of the findings to the population at large. As EHBS is a longitudinal study that is still enrolling participants with a focus on increasing demographic diversity, our next step will be to incorporate data from subsequent visits to track cognitive and neurobiological changes over time. We predict effects will strengthen as impairments begin to emerge in this cohort. Fourth, AD biomarkers were limited to CSF and MRI features, and amyloid and tau PET results were not included. Finally, our analysis lacks well‐designed task‐based fMRI, which may be more sensitive to early changes in AD.

We also note some strengths and limitations of EFA. EFA does not, by itself, guarantee that extracted factors represent discrete cognitive domains, and shared variance may reflect non‐cognitive influences such as age, subclinical pathology, or other unobserved processes. Our goal in applying EFA was not to claim the discovery of “pure” cognitive domains, but rather to derive empirically informed composite dimensions of performance that summarize shared variance across tests. While our factors did appear to cluster partially by test battery, battery‐specific clustering likely reflects a combination of overlapping cognitive demands, shared task characteristics, and method variance. This observation reinforces the need for an empirical approach to dimensionality reduction rather than assuming a priori domain assignments based solely on test labels. Using EFA allowed us to derive cognitive domains in an unbiased, cohort‐specific manner, reducing the risk of misclassification and enhancing construct validity for downstream regression analyses.

Outstanding questions include: (1) Do other visuospatial episodic memory tasks show similar sensitivity as RCFT to AD biomarkers in asymptomatic individuals? (2) Does the factor structure change in subsequent testing with the influence of practice effects? (3) How does the factor structure hold up in clinical populations? (4) Do other factors become more relevant further into disease progression? Future research should continue to explore the role of visuospatial episodic memory in identifying individuals at risk for cognitive decline, with an emphasis on longitudinal assessment in diverse populations.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the supporting information.

CONSENT STATEMENT

All human research participants provided informed consent.

Supporting information

Supporting Information

Supporting Information