Introduction
There is an important need to identify early cognitive changes in individuals at risk for dementia prior to the development of significant cognitive and functional decline. Cognitive function measured by comprehensive neuropsychological evaluation is typically expressed as mean level of performance within domains such as memory, attention, and executive function. However, there has been growing recognition that intraindividual variability in neuropsychological performance within a single testing session (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019; Gleason et al., Reference Gleason, Norton, Anderson, Wahoske, Washington, Umucu, Koscik, Dowling, Johnson, Carlsson and Asthana2018; Koscik et al., Reference Koscik, Berman, Clark, Mueller, Okonkwo, Gleason, Hermann, Sager and Johnson2016; Malek-Ahmadi et al., Reference Malek-Ahmadi, Lu, Chan, Perez, Chen and Mufson2017), may be sensitive to early cognitive changes and may reflect subtle decline in cognition that can be detected before traditional neuropsychological thresholds for cognitive impairment are met.
Intraindividual variability has two main operationalizations including dispersion and inconsistency. In contrast to dispersion (which examines within-person variability across tasks), inconsistency measures within-person variability on a single task. In the present study, we focus on dispersion. Although some variability across domains is seen in normal cognitive profiles, increased variability has been found to be associated with decreased neurological integrity (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019; Malek-Ahmadi et al., Reference Malek-Ahmadi, Lu, Chan, Perez, Chen and Mufson2017). Indeed, greater cognitive dispersion has been associated with an increased likelihood of being classified as having AD (Halliday et al., Reference Halliday, Stawski, Cerino, Decarlo, Grewal and Macdonald2018) and greater dementia incidence at follow-up (Watermeyer et al., Reference Watermeyer, Goerdten, Johansson and Muniz-Terrera2021). Studies examining brain changes underlying greater cognitive dispersion in aging, dementia risk, and neurodegenerative disease have shown that elevated cognitive dispersion is associated with faster rates of cerebral atrophy in the medial temporal lobes (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019), disruptions in functional connectivity networks (Meeker et al., Reference Meeker, Ances, Gordon, Rudolph, Luckett, Balota, Morris, Fagan, Benzinger and Waring2021), and reduced integrity of white matter pathways interconnecting cortical regions mediating executive function and attention (Sorg et al., Reference Sorg, Merritt, Clark, Werhane, Holiday, Schiehser, Bondi and Delano-Wood2021). In addition, an autopsy study showed that greater cognitive dispersion was significantly associated with more severe neurofibrillary tangle pathology and trended toward an association with more severe neuritic plaques (Malek-Ahmadi et al., Reference Malek-Ahmadi, Lu, Chan, Perez, Chen and Mufson2017). However, findings from a study examining intraindividual variability and cerebrospinal fluid (CSF) based markers of amyloid-beta (Aβ) found that cognitive dispersion was not significantly associated with CSF Aβ (Watermeyer et al., Reference Watermeyer, Marroig, Ritchie, Ritchie, Blennow and Muniz-Terrera2020). We know of no published studies that have examined the association of cognitive dispersion and in vivo brain measures of Aβ using positron emission tomography (PET) imaging. In addition, to the best of our knowledge, no published studies have examined the association between cognitive dispersion and subtle cerebrovascular changes including cerebral blood flow (CBF) measured with arterial spin labeling (ASL) magnetic resonance imaging (MRI).
In the present study, we sought to determine whether cognitive dispersion is associated with cerebral amyloidosis in vivo by examining whether older adults without clinical dementia who are Aβ-positive (Aβ+) on PET imaging demonstrate higher cognitive dispersion relative to those who are Aβ negative (Aβ-). In the present study, we focused on dispersion given that data from a single task with various intervals suitable to calculate inconsistency (e.g., a reaction time task with several trials) was not available for analysis. In addition, to further elucidate mechanisms underlying increased cognitive dispersion, we examined associations between cognitive dispersion and CBF across our entire sample of older adults as well as the subsample who was Aβ+. We hypothesized that higher cognitive dispersion would be associated with: (1) PET Aβ accumulation; and (2) reduced CBF in AD-vulnerable regions, particularly among those individuals who are Aβ+.
Method
The ADNI dataset
Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). The ADNI was launched in 2003 as a public–private partnership, led by Principal Investigator Michael W. Weiner, MD. This study was approved by the Institutional Review Board at the ADNI study sites. Treatment of human participants during this study was in full accordance with ethical standards set forth by the Helsinki Declaration.
The primary goal of ADNI has been to test whether serial MRI, PET, other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer’s disease (AD). For up-to-date information, see www.adni-info.org.
Participants
Specific enrollment criteria for ADNI have been previously described in detail elsewhere (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019). Briefly, participants from ADNI were 55–90 years old, had ≥6 years of education or work history equivalent, were fluent in English or Spanish, had a Geriatric Depression Scale <6, had a Hachinski Ischemia Scale <5, adequate vision and hearing to perform neuropsychological tests, were in generally good health and without significant head trauma or neurologic disease, were stable on permitted medications, and had a reliable study partner. ADNI includes participants with normal cognition, MCI, and dementia. The current study included 148 participants from ADNI GO/ADNI 2 cohorts when ASL MRI was collected. Participants were included if they had ASL data collected within 12 months of their baseline visit, did not have dementia at their baseline study visit, and had available baseline neuropsychological testing and florbetapir PET imaging data. Although ADNI is a longitudinal study, the present analyses examined cross-sectional associations of baseline data.
Cognitive dispersion index
The cognitive dispersion index reflects variability across cognitive measures at a single time point. We calculated the index of dispersion using procedures previously described (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019). Briefly, standard summary measures from tests designed to assess multiple different cognitive abilities were included in the cognitive dispersion index. Six neuropsychological measures were selected given their routine use in assessing early cognitive changes in AD, administration across all ADNI waves, and sampling of three different domains of cognition (i.e., language, processing speed/executive function, and episodic memory). These six measures were as follows: (1) Animal Fluency, total score; (2) 30-item Boston Naming Test (BNT) total score; (3) Trail Making Test (TMT), Part A; time to completion; (4) TMT, Part B; time to completion; (5) Rey Auditory Verbal Learning Test (AVLT) 30-min delayed free recall; number of words recalled; and (6) AVLT recognition; number of words correctly recognized.
Prior to calculating the cognitive dispersion index, individual raw scores for each measure were converted into age-, gender-, and education-adjusted Z scores with a mean of 0 and standard deviation of 1 using regression coefficients derived from robust cognitively normal participants (n = 385) who had at least 1 year of follow-up and remained cognitively normal throughout their participation in the ADNI study (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019; Edmonds et al., Reference Edmonds, Delano-Wood, Clark, Jak, Nation, McDonald, Libon, Au, Galasko, Salmon and Bondi2015). The two TMT Z scores were multiplied by −1 so that higher Z scores represent better performance for all scores. The intraindividual standard deviation across the 6 Z scores was calculated to create the cognitive dispersion index. A high score on the cognitive dispersion index reflects greater variability across neuropsychological measures whereas a low score on the cognitive dispersion index indicates more consistency across measures (regardless of scores on the individual neuropsychological measures included in the cognitive dispersion index). In addition, mean level of cognitive performance was calculated as the average of the 6 Z scores that were included in the cognitive dispersion index.
Cognitive status
Participants diagnosed with dementia by ADNI were excluded from the current study. To determine cognitive status (MCI vs. normal cognition), actuarial neuropsychological MCI criteria were applied to all participants in this sample (Edmonds et al., Reference Edmonds, Delano-Wood, Clark, Jak, Nation, McDonald, Libon, Au, Galasko, Salmon and Bondi2015). Participants were considered MCI if they performed >1 SD below the age-/education-/sex-adjusted mean on: (1) 2 neuropsychological measures within the same cognitive domain; or (2) at least 1 measure across all 3 sampled cognitive domains. The six neuropsychological test scores included in the cognitive dispersion index were considered in the diagnostic criteria for MCI.
T1-weighted anatomical and ASL MRI data acquisition and processing
Detailed information describing the imaging data acquisition and processing is available online at www.loni.usc.edu. MR imaging was performed on a 3.0 Tesla scanner and structural MRI and ASL scans were collected during the same session.
A T1-weighted 3D MPRAGE sequence was collected using the following parameters: field of view = 256 mm, repetition time = 2300 ms, echo time = 2.98 ms, flip angle = 9°, and resolution = 1.1 x 1.1 x 1.2 mm3. Structural scans were motion corrected, skull stripped, segmented, and parcellated using FreeSurfer Version 5.1 (surfer.nmr.mgh.harvard.edu; Fischl et al., Reference Fischl, Salat, Busa, Albert, Dieterich, Haselgrove, van der Kouwe, Killiany, Kennedy, Klaveness, Montillo, Makris, Rosen and Dale2002, Reference Fischl, van der Kouwe, Destrieux, Halgren, Ségonne, Salat, Busa, Seidman, Goldstein, Kennedy, Caviness, Makris, Rosen and Dale2004).
Pulsed ASL scans were collected using QUIPS II with thin-slice TI1 periodic saturation with echo-planar imaging (Luh et al., Reference Luh, Wong, Bandettini and Hyde1999). Scan parameters include the following: inversion time of arterial spins (TI1) = 700 ms, total transit time of spins (TI2) = 1900 ms, tag thickness = 100 mm, tag to proximal slice gap = 25.4 mm, repetition time = 3400 ms, echo time = 12 ms, field of view = 256 mm, matrix = 64 x 64, 24 4-mm thick axial slices [52 tag + control image pairs], time lag between slices = 22.5 ms.
As previously described (Bangen et al., Reference Bangen, Thomas, Sanchez, Edmonds, Weigand, Delano-Wood and Bondi2021; Sanchez et al., Reference Sanchez, Thomas, Edmonds, Bondi and Bangen2020; Thomas et al., Reference Thomas, Osuna, Weigand, Edmonds, Clark, Holmqvist, Cota, Wierenga, Bondi and Bangen2021), ASL data processing was largely automated and involved motion correction, aligning each ASL frame to the first frame using a rigid body transformation, and least squares fitting using SPM 8 (http://www.fil.ion.ucl.ac.uk/spm/). Perfusion-weighted images were computed as the difference of the mean-tagged and mean-untagged ASL images and were intensity scaled to account for signal decay during acquisition and to generate intensities in meaningful physiological units. After geometric distortion correction, ASL images were aligned to structural T1 images using FSL. In order to minimize the effects of lower perfusion in white matter on CBF estimates, a partial volume correction was performed that assumed that CBF in gray matter is 2.5 times greater than in white matter. The partial volume corrected perfusion-weighted images were normalized by the reference image (i.e., an estimate of blood water magnetization) to convert the signal into physical units (ml/100 g tissue/min). ADNI quality control procedures to determine a global pass/fail rating were based on visual inspection of signal uniformity, geometrical distortions, gray matter contrast, and presence of large artifacts. A rating of “unusable” in any of these categories resulted in a global “fail” and that participant was excluded from the present study.
FreeSurfer-derived anatomical regions of interest (ROIs) were applied to CBF maps to extract regional CBF estimates for each participant. Our primary analyses examined the following five a priori ROIs: (1) hippocampus; (2) inferior parietal lobe (IPL); (3) inferior temporal gyrus (ITG); (4) medial orbitofrontal cortex (mOFC); and (5) rostral middle frontal gyrus (rMFG). These regions were selected given prior work showing these regions are vulnerable to early AD-related change (Dickerson et al., Reference Dickerson, Stoub, Shah, Sperling, Killiany, Albert, Hyman and Detoledo-Morrell2011) as well as to be consistent with our previous studies examining CBF in ADNI (e.g., Sanchez et al., Reference Sanchez, Thomas, Edmonds, Bondi and Bangen2020; Thomas et al., Reference Thomas, Osuna, Weigand, Edmonds, Clark, Holmqvist, Cota, Wierenga, Bondi and Bangen2021). CBF ROI values were residualized by precentral CBF, which was selected to serve as a reference region as it is not thought to be impacted in early AD (allowing for adjustment of individual variation in CBF that is likely not due to AD pathologies) as well as to be consistent with previous ADNI ASL studies that used this approach (Mattson et al., Reference Mattsson, Tosun, Insel, Simonson, Jack, Beckett, Donohue, Jagust, Schuff and Weiner2014; Yew & Nation., Reference Yew and Nation2017). Mean CBF corrected for partial volume effects was extracted for each of the ROIs and reference region for each hemisphere separately. To reduce the number of statistical comparisons, averaged bilateral CBF estimates for each ROI were used as the dependent variable in analyses. Bilateral CBF estimates were calculated by averaging the mean CBF of each hemisphere. If participants were missing baseline ASL but had ASL within the first year of their baseline visit, the first occasion of ASL data was used in analyses. In addition, for use in secondary analyses examining brain morphometry, the mean bilateral cortical thickness of IPL, ITF, mOFC, and rMFG was computed by averaging thickness estimates for the left and right hemispheres for each ROI. Total hippocampal volume was computed by summing the volume of left and right hippocampi and was normalized by estimated total intracranial volume.
Florbetapir PET data acquisition and processing
PET scanning with an 18F-florbetapir tracer was used to measure amyloid burden. A detailed description of ADNI florbetapir PET imaging data acquisition and processing can be found online (www.loni.usc.edu). Briefly, florbetapir scans were co-registered, averaged, reoriented into a standard 160 × 160 × 96 voxel image grid with 1.5 mm3 voxels and smoothed to a uniform isotropic resolution of 8 mm full width at half maximum. Each participant’s first florbetapir image was co-registered with the T1-weighted image.
A cortical summary standardized uptake value ratio (SUVR) was calculated by dividing the mean florbetapir uptake across four main cortical regions (i.e., frontal, anterior/posterior cingulate, lateral parietal, and lateral temporal cortices) by whole cerebellar (white and gray matter) florbetapir uptake. Greater cortical Aβ load is thought to increase retention of florbetapir. Aβ positivity was established using the recommended threshold for cross-sectional florbetapir analyses of 1.11 using the whole cerebellum as the reference region (Landau et al., Reference Landau, Thomas, Thurfjell, Schmidt, Margolin, Mintun, Pontecorvo, Baker and Jagust2014).
Statistical analyses
Demographic and clinical characteristics were examined with descriptive statistics. T tests for continuous variables and chi-square (X 2) tests for categorical variables were used to compare Aβ+ versus Aβ- groups on demographics and clinical variables. Analysis of covariance (ANCOVA) was used to compare Aβ+ and Aβ- groups in terms of cognitive dispersion after adjusting for age, gender, education, and mean cognitive performance. In addition, hierarchical linear regression, adjusting for age, gender, education, and mean cognitive performance was used to examine the associations between the cognitive dispersion index and CBF in the five a priori ROIs: hippocampus, IPL, ITG, mOFC, and rMFG. In addition, we performed linear regressions substituting regional volume or cortical thickness of the ROI as the dependent variable in place of CBF to determine whether dispersion was significantly associated with morphometry. For all models, covariates were entered on Step 1 and cognitive dispersion was entered on Step 2. In a series of sensitivity analyses, in an attempt to further clarify the potential role of cognitive dispersion, we re-ran our primary models reversing the position of mean cognitive performance and cognitive dispersion. That is, we entered age, gender, education, and cognitive dispersion on Step 1 and mean cognitive performance on Step 2. Finally, to explore whether the pattern of findings was driven by those participants who were Aβ+ (on the AD continuum as evidenced by significant cerebral amyloidosis), secondary analyses stratified by Aβ status (Aβ+ and Aβ-) were performed. To address potential inflation of type I error resulting from multiple comparisons, we applied the Benjamini–Hochberg procedure (Benjamini & Hochberg, Reference Benjamini and Hochberg1995) to our results. We assessed results when the false discovery rate (FDR) was controlled at 0.05 and 0.10.
All analyses were performed using Statistical Package for the Social Sciences (SPSS) version 26 (SPSS IBM, New York, USA). Figures were made with R version 4.5.0 (https://cran.r-project.org/) and SPSS. An alpha = 0.05 was set for statistical significance; all tests were two-tailed.
Results
Participant characteristics
Participants’ demographic and clinical data are presented in Table 1. One hundred and forty-eight older adults ranging in age from 55 to 85 (mean ± SD = 70.98 ± 6.3) comprised the present sample. There were 72 women (48.6%), the sample was 93.9% white, and the average number of years of formal education was approximately 17 (SD = 2.51). Compared to Aβ- participants, the Aβ+ group were significantly older, more likely to be characterized as MCI, and exhibited poorer cognitive scores (i.e., lower mean level of cognitive performance and poorer performance on each of the individual cognitive measures included in the cognitive dispersion index with the exception of animal fluency on which the groups did not differ). There were no significant group differences in terms of education, gender, or race (p-values > 0.05).
Table 1. Demographics for overall sample and by Aβ PET imaging status
Note. Results from t tests for continuous variables and chi-square tests for dichotomous variables. Data are summarized as mean (standard deviation), unless otherwise indicated. Effects sizes (Cohen’s d for t tests and phi for chi-square tests) are reported as absolute values. Significant group differences (p < .05) appear in bold font. CI = confidence interval; Aβ = amyloid beta; MCI = mild cognitive impairment; AVLT = Rey Auditory Verbal Learning Test.
a Amyloid negativity versus positivity was based on the recommended threshold for cross-sectional florbetapir analyses of 1.11 using the whole cerebellum as the reference region (Landau et al., Reference Landau, Thomas, Thurfjell, Schmidt, Margolin, Mintun, Pontecorvo, Baker and Jagust2014).
b Mean cognitive performance is the mean of the six age-, sex-, and education-adjusted neuropsychological Z scores included in the cognitive dispersion index. The six scores were Animal Fluency, total score; 30-item Boston Naming Test (BNT) total score; Trail Making Test (TMT), Part A; time to completion; TMT, Part B; time to completion; Rey Auditory Verbal Learning Test (AVLT) 30-min delayed free recall; number of words recalled; and AVLT recognition; number of words correctly recognized.
Cognitive dispersion by Aβ status
After adjusting for age, gender, education, and mean cognitive performance, there was a significant main effect of Aβ status on cognitive dispersion such that participants who were Aβ+ showed higher cognitive dispersion relative to those who were Aβ- (F(1,142) = 9.132, p = .003) (See Figure 1).
Figure 1. Cognitive dispersion by amyloid-beta (Aβ) positivity versus negativity. The lines represent group medians and the boxes represent the interquartile range; the y axis represents the model predicted cognitive dispersion values after controlling for age, gender, education, and mean cognitive performance.
Associations of cognitive dispersion and regional CBF
Hierarchical linear regression models showed that, across the entire sample, after adjusting for age, gender, education, and mean cognitive performance, higher cognitive dispersion was significantly associated with reduced CBF in IPL (β = −.183, p = .027; Overall model: R 2 = .117, F(5,142) = 3.748, p = .003) and ITG (β = −.214, p = .011; Overall model: R 2 = .091, F(5,142) = 2.858, p = .017). There were no significant associations between cognitive dispersion and CBF in mOFC, rMFG, or hippocampal regions (all p’s > .05) (See Table 2 and Figure 2). When we performed sensitivity analyses in which we re-ran our primary models reversing the position of mean cognitive performance and cognitive dispersion, results remained similar to the primary models. That is, mean level of cognitive performance related to hippocampal CBF whereas cognitive dispersion was associated with CBF in IPL and ITG. There were no other significant associations between mean level of cognitive performance or cognitive dispersion and CBF. See Supplemental Material file for results of these sensitivity analyses.
Table 2. Hierarchical linear regression models for association of cognitive dispersion and regional CBF adjusting for demographics and mean level of cognitive performance
SE = standard error; CBF = cerebral blood flow; IPL = inferior parietal lobe; ITG=inferior temporal gyrus; mOFC = medial orbitofrontal cortex; rMFG = rostral middle frontal gyrus; sr = semi partial correlation coefficient. For gender, women are the reference group. Bold values are statistically significant (p < .05). Cognitive dispersion was calculated by using the intraindividual standard deviation across 6 Z scores.
Figure 2. Scatterplot depicting the association between cognitive dispersion and CBF across the entire sample for inferior parietal (A) and inferior temporal (B) cortices by amyloid status. The x axes depict model predicted cognitive dispersion values adjusting for age, gender, education, and mean level of cognitive performance. The y axes depict regional CBF residualized by precentral CBF. Shaded area represents 95% confidence intervals.
Associations of cognitive dispersion and brain morphometry
After adjusting for age, gender, education, and mean cognitive performance, cognitive dispersion was not significantly associated with hippocampal volume or cortical thickness of IPL, ITG, mOFC, or rMFG regions (all p-values > 0.05; See Table 3).
Table 3. Hierarchical linear regression models for association of cognitive dispersion and regional volume/cortical thickness adjusting for demographics and mean level of cognitive performance
SE = standard error; CBF = cerebral blood flow; IPL = inferior parietal lobe; ITG=inferior temporal gyrus; mOFC = medial orbitofrontal cortex; rMFG = rostral middle frontal gyrus; sr = semi partial correlation coefficient. For gender, women are the reference group. Bold values are statistically significant (p < .05). Cognitive dispersion was calculated by using the intraindividual standard deviation across 6 baseline Z scores.
Secondary analyses stratified by Aβ status
Analyses were conducted to examine whether the associations among cognitive dispersion and regional CBF and volume/cortical thickness were driven by those participants who were considered Aβ+. Among Aβ+ individuals (n = 62), adjusting for age, gender, education, and mean cognitive performance, higher cognitive dispersion was significantly associated with reduced IPL CBF (β = −.341, p = .011; Overall model: R 2 = .113, F(5,57) = 1.457, p = .218) and reduced ITG CBF (β = −.295, p = .028; Overall model: R 2 = .100, F(5,57)=1.265, p = .292). See Figure 2. Cognitive dispersion was not significantly associated with CBF in hippocampal, mOFC, or rMFG regions. When examining volume/cortical thickness, cognitive dispersion was not significantly associated with hippocampal volume or thickness of IPL, OTG, mOFC, or rMFG, ITG. Among Aβ- individuals (n = 84), there were no significant associations between cognitive dispersion and CBF across ROIs or between cognitive dispersion and volume/cortical thickness (all p-values > 0.05) (See Tables 4 and 5 and Figure 2).
Table 4. Hierarchical linear regression models for association of cognitive dispersion and regional cerebral blood flow adjusting for demographics and mean level of cognitive performance in Aβ+ individuals
Aβ = amyloid beta; SE = standard error; CBF = cerebral blood flow; IPL = inferior parietal lobe; ITG = inferior temporal gyrus; mOFC = medial orbitofrontal cortex; rMFG = rostral middle frontal gyrus; sr = semi partial correlation coefficient. For gender, women are the reference group. Bold values are statistically significant (p < .05). Cognitive dispersion was calculated by using the intraindividual standard deviation across 6 baseline Z scores.
Table 5. Hierarchical linear regression models for association of cognitive dispersion and regional cerebral blood flow adjusting for demographics and mean level of cognitive performance in Aβ- individuals
Aβ = amyloid beta; SE = standard error; CBF = cerebral blood flow; IPL = inferior parietal lobe; ITG = inferior temporal gyrus; mOFC = medial orbitofrontal cortex; rMFG = rostral middle frontal gyrus; sr = semi partial correlation coefficient. For gender, women are the reference group. Bold values are statistically significant (p < .05). Cognitive dispersion was calculated by using the intraindividual standard deviation across 6 baseline Z scores.
False discovery rate
Statistical significance of all reported findings was retained under a 0.10 FDR but not maintained under a 0.05 FDR.
Discussion
In a sample of well-characterized older adults free of clinical dementia, we found that those individuals who were Aβ+ on PET imaging showed greater cognitive dispersion than their counterparts who were Aβ-. In addition, greater cognitive dispersion was significantly associated with reduced CBF in IPL and ITG regions after adjusting for age, gender, education, and mean cognitive performance. Secondary analyses stratified by Aβ status revelated that these associations were driven primarily by Aβ+ individuals rather than Aβ- individuals. Across the entire sample and within the Aβ+ and Aβ- subgroups, there were no significant associations between cognitive dispersion and brain morphometry (i.e., cortical thickness and volume).
The study of cognitive dispersion has a long history in the field of psychology (Vance et al., Reference Vance, Del Bene, Frank, Billings, Triebel, Buchholz, Rubin, Woods, Li and Fazeli2021). It has been long believed that a large degree of test scatter or variation characterizes some types of psychological disorders (Plake et al., Reference Plake, Reynolds and Gutkin1981) and many commonly administered neuropsychological measures include indices of scatter or discrepancy as standardized variables (e.g., Wechsler Adult Intelligence Scale beginning with the Revised (WAIS-R) version, Delis-Kaplan Executive Function Scale, California Verbal Learning Test), although base rate information is not always available (Jacobson et al., Reference Jacobson, Delis, Peavy, Wetter, Bigler, Abildskov, Bondi and Salmon2009). Nonetheless, there is recent growing interest in using cognitive dispersion indices to predict future decline and cognitive outcomes in various neurological disorders including AD, human immunodeficiency virus, and traumatic brain injury. Increased cognitive dispersion may manifest in a reduced ability to integrate cognitive processes, which could then lead to reduced cognitive control and functional inefficiency (Fellows & Schmitter-Edgecombe, Reference Fellows and Schmitter-Edgecombe2015). It has been theorized that increasing cognitive dispersion may reflect a disruption of neural networks (Jacobson et al., Reference Jacobson, Delis, Peavy, Wetter, Bigler, Abildskov, Bondi and Salmon2009; Parasuraman & Martin, Reference Parasuraman and Martin1994), and several neuroimaging studies have shown that increasing intraindividual variability is associated with reduced functional connectivity (Lin & McDonough, Reference Lin and McDonough2022). AD has long been conceptualized as involving a disconnection syndrome (Delbeuck et al., Reference Delbeuck, Van der Linden and Collette2003) given that early AD is characterized by the loss of cortico-cortical projections that promote interactions of multiple brain regions. Recent neuroimaging studies have shown disruption of multiple networks including the frontoparietal and default mode networks in AD (Contreras et al., Reference Contreras, Aslanyan, Sweeney, Sanders, Sagare, Zlokovic, Toga, Han, Morris, Fagan, Massoumzadeh, Benzinger and Pa2020; Meeker et al., Reference Meeker, Ances, Gordon, Rudolph, Luckett, Balota, Morris, Fagan, Benzinger and Waring2021). The current findings suggest that cognitive dispersion is elevated in individuals who are Aβ+ and that reduced CBF may play a role in increasing dispersion, although future longitudinal studies are needed to confirm this.
Previous studies have shown that greater cognitive dispersion at baseline predicts progression to MCI (Gleason et al., Reference Gleason, Norton, Anderson, Wahoske, Washington, Umucu, Koscik, Dowling, Johnson, Carlsson and Asthana2018), faster rates of medial temporal atrophy (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019), and increased risk of incident MCI (Holtzer et al., Reference Holtzer, Jacobs and Demetriou2020). Our current findings suggest that higher cognitive dispersion is also associated with cerebral amyloidosis, which dovetails with previously published studies including an autopsy study showing that greater cognitive dispersion was significantly associated with more severe neurofibrillary tangle pathology and trended toward an association with more severe neuritic plaques (Malek-Ahmadi et al., Reference Malek-Ahmadi, Lu, Chan, Perez, Chen and Mufson2017) and a second study what showed that CSF measured Aβ moderated the relationship between cognitive dispersion and resting-state functional connectivity (Meeker et al., Reference Meeker, Ances, Gordon, Rudolph, Luckett, Balota, Morris, Fagan, Benzinger and Waring2021). However, our findings differ from another recent study which reported that cognitive dispersion was not associated with CSF markers of AD pathology (Watermeyer et al., Reference Watermeyer, Marroig, Ritchie, Ritchie, Blennow and Muniz-Terrera2020). Differences between the current study and that by Watermeyer and colleagues (2020) may account for the discrepant findings including the method of measuring amyloid status (CSF vs. PET) and the type and number of cognitive tests used to calculate the cognitive dispersion index (Watermeyer and colleagues used more measures than the 6 included in the present study and they also included experimental measures).
In the present study we found greater cognitive dispersion was associated with hypoperfusion in posterior but not anterior regions, which is in line with previous MRI studies showing brain changes in posterior regions in AD risk (Brickman et al., Reference Brickman, Zahodne, Guzman, Narkhede, Meier, Griffith, Provenzano, Schupf, Manly, Stern and Mayeux2015; Yew & Nation, Reference Yew and Nation2017). Previous research examining small vessel cerebral vascular disease as measured by white matter hyperintensities (WMH) has shown normal aging-related increases in WMH volume in anterior regions but AD-specific increases in WMH in posterior regions (Brickman et al., Reference Brickman, Zahodne, Guzman, Narkhede, Meier, Griffith, Provenzano, Schupf, Manly, Stern and Mayeux2015). Our current finding is also in line with a previous ADNI ASL study that examined the same 5 a priori regions we studied in the present study and found that the only region that showed CBF differences between Aβ+ older adults free of dementia versus participants with AD dementia was the inferior parietal region. In addition, individuals with AD showed reduced CBF in hippocampus, inferior parietal, and inferior temporal regions relative to Aβ- older adults but there were no CBF differences among the cognitive groups in frontal regions (Yew & Nation., Reference Yew and Nation2017). Our finding of significant associations between higher cognitive dispersion and reduced CBF in posterior but not frontal regions is in line with previous research within the ADNI sample that focused on group differences in CBF based on cognitive status (i.e., unimpaired cognition, objectively defined subtle cognitive decline [Obj-SCD], and MCI). This previous study (which had some overlap with the current study’s sample) showed that those with Obj-SCD had altered CBF in the IPL and hippocampus relative to a cognitively normal group, suggesting early neurovascular dysfunction in these key regions may precede later cognitive impairment (Thomas et al., Reference Thomas, Osuna, Weigand, Edmonds, Clark, Holmqvist, Cota, Wierenga, Bondi and Bangen2021). Similar to the present study, Thomas and colleagues did not find significant differences between cognitive groups in CBF in frontal regions.
Cognitive dispersion was not associated with morphometry including regional gray matter volume (hippocampus) or cortical thickness (IPL, ITG, mOFC, rMFG), consistent with our previous findings showing no cross-sectional associations between baseline cognitive dispersion and morphometry (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019). However, in our previous study we found that higher cognitive dispersion predicted faster rates of medial temporal lobe atrophy at 24-month follow-up in the ADNI cohort (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019). This pattern of findings suggests that cognitive dispersion is a sensitive marker of future neurodegeneration. This previously published paper (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019) included 736 participants some of which overlapped with the present study, although did not examine the majority of the regions included in the current paper. The present findings that cognitive dispersion is associated with reduced CBF but not morphometry at baseline suggests that ASL CBF is a useful marker of early and subtle brain changes that may be observed prior to significant atrophy, and dovetails with our previous research showing hypoperfusion predicts later neurodegeneration (Bangen et al., Reference Bangen, Thomas, Sanchez, Edmonds, Weigand, Delano-Wood and Bondi2021).
Cognitive dispersion indices may be influenced by relative differences in the difficulty, sensitivity, and score distributions of the component tasks as well as floor or ceiling effects (Cherry et al., Reference Cherry, Buckwalter and Henderson2002; Jacobson et al., Reference Jacobson, Delis, Peavy, Wetter, Bigler, Abildskov, Bondi and Salmon2009). However, intraindividual variability indices have been used as a means of identifying subtle decline in cognitive skills relative to those cognitive abilities that may be more resilient to neurodegenerative processes (Jacobson et al., Reference Jacobson, Delis, Peavy, Wetter, Bigler, Abildskov, Bondi and Salmon2009). In the early phases of neurodegeneration, an individual may show mild declines in one or two cognitive abilities while other abilities may be less affected. Given that some individuals in a preclinical stage of AD may not show a significant memory impairment and may perform within the intact or normal range on individual cognitive tests, it may be that intraindividual variability metrics are more sensitive than individual tests scores, particularly in identifying individuals who are experiencing very subtle decline and/or who are high functioning (Jacobson et al., Reference Jacobson, Delis, Peavy, Wetter, Bigler, Abildskov, Bondi and Salmon2009; Storandt et al., Reference Storandt, Grant, Miller and Morris2006).
Although comparing different dispersion metrics was not a primary purpose for the current study, in an effort to determine whether our findings may relate to differences in sensitivity to AD across tasks (mean level of performance) rather than pure measures of dispersion, we calculated dispersion two additional ways and re-ran our primary models with these alternative dispersion metrics. Given that episodic memory is typically affected early in AD together with our results suggesting that the two AVLT measures may be more sensitive to Aβ status relative to other measures included in the dispersion index (see Table 1), we re-ran our models with alternative dispersion indices that varied based on whether or how the AVLT measures were included. First, we calculated a dispersion variable not including the two AVLT measures. That is, we calculated a dispersion variable with the following four variables: BNT, Animals Fluency, Trails A, and Trails B. Results from the model with inferior parietal CBF as the dependent variable remained significant (p = .006) although the results from the model with inferior temporal CBF as the dependent variable was somewhat attenuated and was a trend (p = .080). As with our primary models, the models with hippocampal CBF and frontal CBF as the dependent variables were not significant. Next, we created a dispersion variable using the same 6 measures including in our original variable but partialed out mean AVLT performance. The pattern of results remained similar to our primary analyses although results were attenuated with cognitive dispersion relating to CBF at a trend level (p = .052 for inferior parietal CBF and p = .072 for inferior temporal lobe). Given that differences in sensitivity to AD across different measures may contribute to dispersion effects, future research should more directly compare the predictive utility of different dispersion metrics as well of intraindividual variability indices relative to individual test scores.
Cognitive dispersion has been measured using different approaches across studies. Consistent with several previous studies, we assessed cognitive dispersion as within-person variability across different neuropsychological measures (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019; Gleason et al., Reference Gleason, Norton, Anderson, Wahoske, Washington, Umucu, Koscik, Dowling, Johnson, Carlsson and Asthana2018; Watermeyer et al., Reference Watermeyer, Marroig, Ritchie, Ritchie, Blennow and Muniz-Terrera2020) rather than the inconsistency of trial performance across one task. Although few studies have directly compared dispersion and inconsistency, one previous study have found that these two methods of measuring intraindividual variability are moderately correlated (r = .38 on a choice reaction time task; r = .31 on a 1-back task) and are both associated with increasing age and cognitive decline (Hilborn et al., Reference Hilborn, Strauss, Hultsch and Hunter2009). Future studies comparing dispersion and inconsistency will help determine how these two metrics may complement each other. To improve generalizability and to consider multiple cognitive domains, we selected neuropsychological tests that are commonly used in research and clinical settings to be included in our dispersion metric. In addition, the approach we used to calculate cognitive dispersion has been found to be particularly advantageous as it can be calculated from one testing session, showing cognitive dispersion has potential for clinical utility due to its ease of implementation without change to standardized testing procedures (Holtzer et al., Reference Holtzer, Verghese, Wang, Hall and Lipton2008). Notably, however, there is not yet consensus on how to best operationalize cognitive dispersion. Additional studies providing empirical support for which measures and how many measures form the optimal dispersion index, establishment of a universally accepted method for calculating dispersion, and development of normative databases will increase the utility of dispersion in both research and clinical settings.
Strengths of the study include the well-characterized sample of older adults; integration of multimodal imaging data including PET Aβ, ASL perfusion, and structural MRI; and use of a sensitive measure of cognitive dispersion (i.e., an intraindividual standard deviation across multiple domains of cognitive functioning). Cognitive dispersion has several advantages relative to other proposed markers to detect early brain changes (e.g., PET imaging or lumbar puncture to measure CSF) including being low-cost and noninvasive. ASL MRI has advantages over other imaging techniques designed to measure CBF due to its noninvasive nature (i.e., does not require injection of contrast agent). Reduced CBF is also a well-established marker of subtle vascular change and has been associated with poorer everyday functioning (Sanchez et al., Reference Sanchez, Thomas, Edmonds, Bondi and Bangen2020), faster rater rates of memory decline, neurodegeneration, and progression of small vessel disease (Bangen et al., Reference Bangen, Thomas, Sanchez, Edmonds, Weigand, Delano-Wood and Bondi2021). This study expands on previous research on cognitive functioning and CBF by linking hypoperfusion to increased cognitive dispersion.
It should be noted that the effects of cognitive dispersion in our models are modest and accounted for 3–4% and 8–11% of the variance in regional CBF across the entire sample and within the Aβ+ subgroup, respectively (entire sample: sr = −.176 for IPL and sr = −.205 for ITG; Aβ+ subgroup: sr = −.327 for IPL and sr = −.283 for ITG). Although these are modest effects, these models adjust for important demographic variables that influence CBF (i.e., age, gender) and mean-level cognitive performance and findings remained similar in sensitivity analyses in which the position of cognitive dispersion and mean cognitive performance were reversed (i.e., cognitive dispersion served as a covariate and mean cognitive performance served as the independent variable). Taken together, these findings provide support for the notion that cognitive dispersion may have incremental utility in assessing dementia risk although future studies comparing multiple cognitive metrics (e.g., dispersion, inconsistency, episodic memory indices) and additional risk factors are needed to clarify this. Limitations of the study include a homogeneous racial/ethnic distribution and a highly educated sample so results may not be generalizable to groups with differing demographic characteristics. Future research in more diverse samples is needed. Additional limitations include the cross-sectional design. Future studies using larger samples should examine data longitudinally to determine whether cognitive dispersion predicts future changes in CBF levels, interacts with biomarkers including Aβ levels to predict clinical outcomes, and adds incremental value in predicting outcomes relative to other cognitive indices and risk factors. Future research is also needed to further evaluate the specificity and sensitivity of dispersion metrics in preclinical and prodromal dementia. In particular, it will be important to examine the associations between dispersion and biomarkers beyond Aβ including those related to cerebrovascular disease burden.
In summary, in conjunction with previous evidence linking cognitive dispersion to faster rates of cerebral atrophy in AD-vulnerable brain regions (Bangen et al., Reference Bangen, Weigand, Thomas, Delano-Wood, Clark, Eppig, Werhane, Edmonds and Bondi2019) and increased AD neuropathology (Malek-Ahmadi et al., Reference Malek-Ahmadi, Lu, Chan, Perez, Chen and Mufson2017), our findings suggest that cognitive dispersion may be a useful noninvasive marker of early cognitive and brain changes especially in the context of those who are Aβ+. Since greater cognitive dispersion was not associated with brain morphometry, but was associated with reduced CBF, this indicates that cognitive dispersion may be a marker of early vascular changes in the brain and may be useful in identifying participants for clinical trials that target vascular risk or amyloid although future longitudinal studies are needed to confirm this.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1355617722000649
Acknowledgments
Data used in this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). ADNI investigators contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this manuscript. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wpcontent/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf
Funding statement
The analyses in this manuscript were funded by the US Department of Veterans Affairs Clinical Sciences Research and Development Service (Career Development Award-2 1IK2CX001865 to K.R.T. and 1IK2CX001415 to E.C.E.; and Merit Award 1I01CX001842 to K.J.B), NIH grants (R01 AG063782 to K.J.B.), and the Alzheimer’s Association (AARF-17-528918 to K.R.T., AARG-18-566254 to K.J.B., AARG-17-500358 to E.C.E.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Data collection and sharing for this project was funded by the Alzheimer’s Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer’s Association; Alzheimer’s Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer’s Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California.
Conflicts of interest
None.
