Neuropsychological subtypes of incident mild cognitive impairment and mild neurocognitive disorder in a population-based cohort of older adults

Abstract

Background

Mild cognitive impairment (MCI) is a heterogenous condition which places individuals at higher risk for Alzheimer's disease, yet it is not well understood. Studies of primarily prevalent MCI have identified different subtypes characterized by different neuropsychological profiles, while a recent incident MCI study empirically identified four neuropsychological subtypes (amnestic, dysexecutive, dysnomic, and subtle cognitive impairment (SCI) subtypes).

Objective

We aimed to identify whether four distinct neuropsychological subtypes could be empirically derived in a sample of a) incident MCI and b) DSM5 mild neurocognitive disorder (mNCD).

Methods

We used data from the Personality and Total Health Through Life study. Participants were aged 72–78, with a diagnosis of incident MCI (n = 117), and/or mNCD (n = 161). We undertook a cross-sectional cluster analysis on neuropsychological data from participants from four domains: executive, memory, language, and visuospatial.

Results

For incident MCI, cluster analysis derived four subtypes, (dysexecutive, SCI, mixed dysnomic/visuospatial and mixed dysexecutive/visuospatial). For mNCD, the resulting four cluster solution included dysexecutive, SCI-amnestic/dysnomic, SCI-dysexecutive and mixed/global impairment. Discriminant function analysis revealed that 94% and 91% of MCI and mNCD participants respectively were correctly classified based on the cognitive domain scores, and further analysis confirmed the SCI groups showed reduced cognitive performance compared with matched cognitively unimpaired participants.

Conclusions

Neuropsychological subtypes were empirically derived in both incident MCI and mild NCD samples, with both SCI and dysexecutive clusters most reliably detected and consistent with previous studies. The early identification of these MCI/mNCD subtypes may help to identify patient groups for targeted early intervention in clinical settings.

Keywords

aging Alzheimer's disease cluster analysis cognitive dysfunction mild cognitive impairment

Introduction

Cognitive changes characterized as mild cognitive impairment (MCI)¹; affect 12–18% of adults aged over 50 years.^2,3 In the clinical context, diagnosis has typically been based on Petersen/Winblad International Working Group (IWG) criteria,^1,4 or expert consensus diagnosis using a combination of objective criteria and clinical judgement,⁵ although there is a potential for these approaches to result in false-positive errors.⁶ Despite recent progress in identifying promising biomarkers of Alzheimer's disease (AD) which could assist with prediction and diagnosis, including neuroimaging⁷ and blood biomarkers,⁸ clinical approaches to identifying MCI remain important for a number of reasons. First, in practice, many clinics do not have access to biomarkers, which can be costly, invasive, and time-consuming to collect, and therefore continue to rely on clinical diagnosis.⁹ Second, at the current time, biomarkers have largely been identified for AD,¹⁰ rather than other forms of dementia, limiting their usefulness in predicting progression to all types of dementia. Third, understanding a person's cognitive and functional change is an important component of treatment and management.¹¹

Despite the usefulness of identifying people with MCI, the diagnosis has a high level of heterogeneity, both the phenotypes and prognosis of people identified as MCI. Efforts have therefore been made to identify distinct neuropsychological subtypes which may improve our understanding and enable more accurate prediction of outcomes in this population. Traditional neuropsychological subtypes have been derived from the pattern of impairment, judged utilizing pre-defined cut-off scores within specific cognitive domains, resulting in classification into amnestic (single or multiple domain) or non-amnestic (single or multiple domain) subtypes.⁴ Alternatively, studies have classified subtypes of MCI using data-driven methodology such as cluster analysis or latent profile analysis of neuropsychological test scores. For example, Edmonds and colleagues⁶ used cluster analysis on neuropsychological data from previously classified prevalent MCI participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort and identified four subtypes: a dysnomic, dysexecutive, amnestic, and cluster-derived normal subtype. These subtypes were later found to be associated with distinct longitudinal patterns of cortical atrophy over 3 years.¹² Overall, the most robust subtypes of prevalent MCI identified via actuarial methods have been amnestic and dysexecutive MCI subtypes,^6,13–15 with a dysnomic subtype identified with the ADNI dataset,^12,13,16 and visuospatial subtypes also identified in some studies.^14,17 Mixed subtypes, typically with more impairment, have also been commonly identified.^13,14,17

Following these studies of prevalent MCI, Machulda and colleagues in the Mayo Clinic Study of Aging identified neuropsychological subtypes within a community sample with incident MCI, using an empirical approach.¹⁸ By studying a sample of incident MCI, rather than prevalent MCI, this study aimed to capture and characterize patterns of early, subtle cognitive decline, representing the earliest signs cognitive impairment. From a subset of participants 50 years or older and cognitively unimpaired at baseline, the authors identified participants with incident MCI using IWG and Petersen criteria,^1,4 requiring consensus agreement between study coordinator, examining physician and neuropsychologist (n = 506). Cluster analysis using a four-group solution revealed an amnestic subtype—isolated memory impairment; a dysexecutive subtype—mild impairment in memory, language, and visuospatial function, and prominent attention/executive impairment; a dysnomic subtype—mild to moderate impairment in memory, attention/executive, and visuospatial domains in addition to more prominent language impairment; and a subtle cognitive impairment (SCI) subtype—representing a subset of the amnestic cluster.¹⁸ The latter had distinctly lower cognitive and functional impairments, however, also showed poorer global cognitive impairment in comparison to a cognitively unimpaired matched control sample, validating the SCI label. These four subtypes were like those identified in the MCI ADNI cohort,⁶ with the exception that Edmonds et al. found the fourth group to have unimpaired cognition rather than SCI. These subtypes have subsequently been validated against cortical atrophy patterns¹⁹ and white matter hyperintensity volumes and degeneration.²⁰ These findings of neuropsychological subtypes represent a step forward in understanding the earliest stages of cognitive decline, and the SCI group may represent a promising target for early intervention. These subtypes, however, are yet to be replicated in an independent incident MCI sample.

In comparison to MCI, the category of mild neurocognitive disorder (mNCD) from the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5), may have some advantages, due to diagnostic criteria that better represent the heterogeneity of cognitive impairments present in the target population, and greater emphasis on objective measures.²¹ Specifically, DMS-5 mild NCD allows for maintained independence through greater effort, compensatory strategies or accommodation, and does not require subjective cognitive concerns from a patient or informant in addition to objective cognitive decline, resulting in a broader number of individuals meeting mild NCD criteria, in comparison to MCI. Additionally, social cognitive deficits are included within the possible cognitive impairments.²² In a population-based cohort of older adults from the Personality and Total Health (PATH) Through Life study,²³ our group previously showed that 83% of participants categorized as mild NCD were also categorized as MCI, showing considerable overlap, but also with subtle differences in cognitive profile between the two groups.²⁴

The current study had two key aims: a) evaluate, and if possible, replicate as closely as possible the identification of incident MCI neuropsychological subtypes in the Mayo Clinic Study of Aging cohort¹⁸ using the PATH sample; and b) investigate whether these same subtypes are identified in a DSM-5 mild NCD subsample.

Methods

All study information, description of dataset, variables included, authors’ prior knowledge of the data, and planned analyses were preregistered on Open Science Framework (https://osf.io/xqgwt).

Participants

Participants were from the PATH study, which has previously been described.²³ In brief, the PATH project is a large Australian, ongoing, population-based, longitudinal cohort study comprising approximately 7500 participants across three distinct age cohorts at baseline (20–24 years, 40–44 years, 60–64 years). Participants for the current study were aged 60–64 years at the Wave 1 data collection (2001–2002; n = 2551). These participants were initially recruited via invitation based on a random sample of the electoral roll within Canberra, Australian Capital Territory and Queanbeyan, New South Wales. Enrolment to vote is compulsory for all Australian citizens. Participants were reassessed every 4 years on a broad range of sociodemographic, health, lifestyle, and neuropsychological measures. For this study, we examined data collected from the 12-year follow-up, who were aged 72–78 years at Wave 4 (n = 1644; data collected 2014–2015). This study was conducted in accordance with the Declaration of Helsinki and approved by the Australian National University Human Research Ethics Committee. All participants provided written informed consent.

As reported in a previous study,²⁵ by Wave 4, 36% of the cohort had been lost to follow up, including 271 who were deceased, and the remainder withdrawn or lost for various reasons (refusal, left catchment area, etc.). Participants who completed all four waves of interviews were significantly more educated than those who had completed only the Wave 1 interview. Of the participants who completed Wave 4, we identified a subset of participants who had been previously identified as either MCI and/or mild NCD in a previous study (see Eramudugolla et al.²⁴). Briefly, participants were identified as meeting the criteria for either classification based on cognitive performance measured using MMSE and 15 cognitive tasks, self/informant reports of cognitive decline, informant interview, case file review by an experienced research neurologist and psychiatrist consensus for complex cases. Based on IWG criteria, 144 MCI participants were identified, and based on DSM-5 criteria, 169 mild NCD participants were identified. These participants were also grouped into clinical subtypes (Amnestic single domain, Amnestic multidomain, Non-amnestic single domain, Non-amnestic multidomain), according to established criteria.⁴ To identify participants with incident MCI (consistent with Machulda et al.¹⁸), those MCI participants who were also classified as MCI or Dementia at Wave 1, Wave 2 or Wave 3 were excluded, leaving a total sample of 118. Of the 118 participants with incident MCI, 100 (84.75%) were also categorized as mild NCD. See Table 1 for participant demographics. A significantly larger proportion of participants classified as incident MCI and/or mild NCD were of non-English speaking background (NESB) compared to the cognitively unimpaired sample. These participant groups were also had lower years of education, and higher depressive symptoms, than the unimpaired sample. In addition, the incident MCI sample also had a higher proportion of participants who were APOE ε4 carriers compared to the unimpaired sample.

Table 1.

Incident MCI and mNCD sample characteristics, with cognitively unimpaired PATH cohort participants for reference.

	Incident MCI (n = 118)	mNCD (n = 167)	CU (n = 1343)	Group differences*
Men (%)	64 (54.24%)	91 (54.49)	691 (51.45%)	N.S.^a
Age (years)	75.27 (1.62)	75.17 (1.63)	75.07 (1.43)	N.S.^b
Years of education	13.91 (2.77)	13.90 (2.92)	14.43 (2.48)	iMCI<CU; mNCD<CU^b
APOE ε4 carrier (%)	43 (36.44%)	51 (31.8%)	331 (25.83%)	iMCI>CU; mNCD=CU^a
NESB (%)	25 (22.7%)	40 (25%)	91 (7.16%)	iMCI>CU; mNCD>CU^a
Employed (%)	6 (5.10%)	10 (6.0%)	141 (10.52%)	N.S.^a
Depression (GDS)	2.11 (2.01)	1.79 (1.81)	1.47 (1.70)	iMCI>CU; mNCD>CU^b
Apathy (AES-7)	21.28 (2.86)	21.27 (3.02)	21.79 (3.04)	N.S.^b

Abbreviations: CU, cognitively unimpaired; iMCI, incident mild cognitive impairment; mNCD, mild neurocognitive disorder; NESB, non-English speaking background; N.S., not significant. *Group differences significant at p < 0.05.

Pearson's Chi-square test.

Independent samples t-test.

Materials

Of the 15 tasks included in the PATH cognitive battery, seven cognitive tasks were included across four domains: Memory, Language, Attention/Executive, and Visuospatial abilities (see Table 2). These tasks were selected from the full PATH battery used in Wave 4, in order to reflect the tests and domains used in Machulda and colleagues,¹⁸ for the purpose of replication. Participants were required to have at least six out of seven scores available, and at least one cognitive test score available for each cognitive domain. It was not possible to exactly replicate the Machulda et al. calculation of the neuropsychological domain z-scores method, as their method involved creating z-scores based on their cohort sample and US Local County population. Instead, we calculated z-scores based on the neuropsychological task means and standard deviations of the entire 60s Wave 4 cohort, stratified by gender and education (low: 5–10 years, medium: 10–15 years; high: 15+ years), as previously specified²⁴.

Table 2.

Neuropsychological measures included in the composite domain z-scores.

Cognitive Domain	Neuropsychological Measures
Learning and Memory	CVLT Delayed Recall; Benton Visual Retention Test Score
Language	Boston Naming Test Short Form Total Score; COWAT Letter Fluency (F & A)
Attention/Executive Function	Trailmaking Test B; SDMT Oral Version Total Score
Visuospatial	Benton Retention Test Copy Score

Abbreviations: CVLT, California Verbal Learning Test; COWAT, Controlled Oral Word Association Test; SDMT, Symbol Digit Modalities Test.

In addition to cognitive task scores from Wave 4, sociodemographic and health variables were also utilized to characterize the sample: including age, gender, years of education, and employment status. APOE ε4 carrier status was also included, and depression symptoms were measured using the Goldberg Depression Scale (GDS) Total Score, apathy symptoms were assessed using the Apathy Evaluation Scale 7 item version (AES-7) score, and global cognition was measured with the Mini-Mental State Examination (MMSE) score.

Statistical analyses

Statistical analyses were undertaken using the IBM SPSS Statistics for Windows (Version 29; Armonk, NY, USA), with alpha set to 0.05. Agglomerative hierarchical cluster analysis with Euclidean distance and Ward's linkage were conducted on the neuropsychological domain z-scores for each of the MCI and mNCD groups, with both three and four cluster solutions specified. We conducted a discriminant function analysis to quantitatively examine the ability of the cognitive domain scores to discriminate the cluster groups as mathematically intended, and examined the stability of the cluster solution using the leave-one-out cross-validation procedure, which minimizes the potential bias of using the same participants within the cluster and discriminant function analyses.²⁶ ANOVA or chi-square goodness of fit analyses were then used to assess any group differences between the subtype groups on relevant demographic, health, and cognitive variables (age, education, gender, APOE ε4 carrier, depression symptoms, Global z score, Memory z score, Language z score, Attention z score, Visuospatial z score). Where SCI groups were identified (classified as clusters where all cognitive domains had mean scores between −1 and 1), given previous literature identifying a group of MCI participants with unimpaired cognition and therefore potentially misclassified,^6,13 we then identified a cognitively unimpaired (CU) demographically matched comparison group with a 1:5 ratio, and then calculated the area under the receiver operating characteristic curve (AUROC) for each SCI cluster neuropsychological domain z-score, comparing these with the relevant CU group.

Results

Cluster analysis and discriminant function analysis results

Incident MCI group. The 3-cluster solution produced the following groups: (1) Dysexecutive (n = 44); (2) Subtle Cognitive Impairment (n = 49); and (3) Mixed (n = 24). The 4-cluster solution produced the same Dysexecutive and SCI Group, with the Mixed Group now split into two groups: dysnomic/visuospatial (n = 15); and dysexecutive/visuospatial (n = 9). Figure 1 shows the boxplots, and Table 3 the means and standard deviations, of neuropsychological domain z-scores for the 4-cluster solution. Linear discriminant analysis revealed that the 3-cluster solution model accurately classified 93.2% of participants, and the 4-cluster solution model accurately classified 94% of the participants. Regarding the leave-one-out cross-validation procedure, 93.2% were correctly classified in the 3-cluster solution, and 94% of participants were classified correctly in the 4-cluster solution.

Figure 1.

Standardized cognitive scores within neuropsychological domains, by incident MCI clusters (4 Cluster Solution). (a) Cluster 1: Dysexecutive Cluster; (b) Cluster 2: SCI Cluster; (c) Cluster 3: Mixed Dysnomic/Visuospatial Cluster; (d) Cluster 4: Mixed Dysexecutive/Visuospatial Cluster.

Table 3.

Means and standard deviations of standardized neuropsychological domain scores, by incident MCI and mNCD clusters (4 Cluster Solutions).

Neuropsychological Domain	Memory	Language	Executive	Visuospatial
Incident MCI
Cluster 1 (n = 44)	−0.55 (0.58)	−0.18 (0.67)	−1.34 (0.71)	0.11 (0.55)
Cluster 2 (n = 49)	−0.50 (0.70)	−0.58 (0.65)	−0.02 (0.56)	0.20 (0.57)
Cluster 3 (n = 15)	−0.84 (0.63)	−0.97 (1.44)	−0.68 (0.91)	−2.70 (0.57)
Cluster 4 (n = 9)	−0.73 (0.44)	−0.50 (0.49)	−2.63 (0.39)	−2.00 (0.37)
mNCD
Cluster 1 (n = 34)	−0.91 (0.41)	−0.68 (0.52)	−1.46 (0.71)	−0.20 (0.65)
Cluster 2 (n = 59)	−0.56 (0.72)	−0.66 (0.58)	−0.01 (0.56)	0.22 (0.53)
Cluster 3 (n = 40)	−0.16 (0.66)	0.31 (0.47)	−0.69 (0.56)	0.25 (0.44)
Cluster 4 (n = 28)	−1.10 (0.60)	−1.22 (1.28)	−1.43 (1.16)	−2.34 (0.78)

Mild NCD group. For mNCD, the 3-cluster solution produced the following groups: (1) Dysexecutive (n = 34); (2) Subtle Cognitive Impairment (n = 99) and (3) Mixed Cluster (n = 28). The 4-cluster solution resulted in the same dysexecutive and mixed clusters, but the SCI group was split into two groups: one with relatively poorer memory and language (SCI-Amnestic/Dysnomic; n = 59) and one with relatively poorer executive functioning (SCI-Dysexecutive; n = 40). Figure 2 shows the boxplots, and Table 3 the means and standard deviations, of the neuropsychological domain z-scores for the 4-cluster solution. Linear Discriminant Function Analysis showed that the 3-cluster solution correctly classified 90.7%, and the 4-cluster solution correctly classified 90.1% of participants. The leave-one-out cross-validation then correctly classified 89.4% of cases for the 3-cluster solution, and 87% of cases for the 4-cluster solution.

Figure 2.

Standardized cognitive scores within neuropsychological domains, by mild NCD clusters (4 Cluster Solution). (a) Cluster 1: Dysexecutive Cluster; (b) Cluster 2: SCI-Amnestic/Dysnomic Cluster; (c) Cluster 3: SCI-Dysexecutive Cluster; (d) Cluster 4: Mixed Dysnomic/Visuospatial Cluster.

Matching MCI/mNCD participants with controls

For each of the incident MCI and mNCD groups, we investigated whether participants classified in the SCI cluster differed from a group of matched cognitively unimpaired participants. To do this, we identified 5 cognitively unimpaired controls for each SCI case in each group, matching exactly on years of education and gender. Out of 49 MCI SCI participants, exact matches were identified for 47 participants. Out of the 59 mNCD SCI-Amnestic/Dysnomic participants, exact matches were identified for 57, and out of 40 mNCD SCI-Dysexecutive participants, exact matches were identified for 39 participants. Participants without exact matches were excluded from these analyses. Cognitively unimpaired participants had never been classified as MCI, mNCD, and did not have any diagnosis of dementia, either at the current timepoint or any previous timepoint. There were no significant differences in age between any of the SCI groups and their matched CU group (MCI SCI: M_diff = −0.07, p = 0.76; mNCD SCI-Amnestic/Dysnomic: M_diff = −0.24, p = 0.25; mNCD SCI-Dysexecutive: M_diff = −0.25, p = 0.34).

AUROC analyses

To confirm that SCI groups did in fact show subtle cognitive impairments, and were therefore not cognitively unimpaired but miscategorized as MCI or mNCD, we calculated the AUROC for the global domain as well as each neuropsychological domain z-score for the MCI SCI and mNCD SCI clusters and compared these with the relevant CU group as previously identified via the matching procedure. First, we calculated the AUROC for the MCI SCI cluster and compared this with the CU group (n = 209), testing whether it was significantly different from 0 at the p = 0.05 level. There were significant differences between groups for the memory (AUROC 0.77; p < 0.001; 95% CI [0.70–0.84]), language (AUROC 0.78; p < 0.001; 95% CI [0.71–0.85]) and executive domains (AUROC 0.66; p < 0.001; 95% CI [0.58–0.75]), as well as for the global domain (AUROC 0.79; p < 0.001; 95% CI [0.72–0.85]; Figure 3A). There was no difference between groups on visuospatial performance (AUROC 0.48; p = 0.70; 95% CI [0.39–0.57]). With regard to the mNCD SCI-Amnestic/Dysnomic group, in comparison to the matched CU group (n = 281), there were significant differences on the memory (AUROC 0.77; p < 0.001; 95% CI [0.70–0.83]), language (AUROC 0.79; p < 0.001; 95% CI [0.74–0.85]), attention (AUROC 0.63; p = 0.003; 95% CI [0.55–0.70]), and global domain (AUROC 0.80; p < 0.001; 95% CI [0.74–0.86]; Figure 3B). In contrast, no differences between groups were seen for the visuospatial domain (AUROC 0.51; p = 0.91, 95% CI [0.42–0.59]). Finally, the mNCD SCI-Dysexecutive group, in comparison to the matched CU group (n = 194), there were significant differences on the executive (AUROC 0.85; p < 0.001; 95% CI [0.79–0.91]) and memory domains (AUROC 0.66, p = 0.002; 95% CI [0.57–0.74]), as well as the global domain (AUROC 0.67; p < 0.001; 95% CI [0.59–0.75]; Figure 3C). In contrast, no differences between groups were seen for the language (AUROC 0.43; p = 0.19; 95% CI [0.34–0.52]) and visuospatial domains (AUROC 0.53; p = 0.53; 95% CI [0.43–0.63]).

Figure 3.

AUROC for Global Cognitive Performance for SCI cluster versus the matched cognitively unimpaired group for (a) incident MCI SCI cluster, (b) mild NCD SCI—Dysexecutive cluster, and (c) mild NCD SCI—Amnestic/Dysnomic cluster.

Demographic, cognitive and diagnostic cluster comparisons

There were no significant differences between clusters on any demographic, genetic, or self-reported mood or psychological variables for either the incident MCI or mNCD groups (see Supplemental Material for full results). Comparisons of clusters on global cognition and clinically categorized subtypes are also reported in the Supplemental Material. Interestingly, a higher proportion of participants from both the incident MCI Cluster 2 (SCI Group), and the mNCD Cluster 2 (SCI-Amnestic/Dysnomic) were also classified as Amnestic Single Domain MCI, compared to the other cluster groups.

Discussion

By undertaking a cluster analysis on an incident MCI group within the community-based PATH older cohort sample, the current study attempted to replicate previous findings of Machulda and colleagues,¹⁸ who identified four neuropsychological subtypes in their incident MCI sample. Similarly, we identified a four-cluster solution which provided a correct classification for a high proportion of the group with no indication of bias of over-fitting, with a distinct Dysexecutive Cluster. In addition, rather than two distinct SCI and amnestic groups, we identified only one SCI group, and two smaller Mixed groups—one with marked dysnomic and visuospatial impairments, and one with more severe dysexecutive and visuospatial impairments. There was no evidence of an Amnestic cluster. Interestingly, the initial three group solution of Machulda and colleagues included one group comprising both the SCI and amnestic groups, in addition to Dysexecutive and Dysnomic groups, and in the final four group solution the SCI group was a subgroup of the Amnestic group. In comparison, while the current study's three group solution also identified an SCI group, in addition to a Dysexecutive and Mixed group; however, the four cluster solution maintained the SCI cluster, and instead split the Mixed group into two distinct groups. Interestingly, Discriminant Function Analysis and cross-validation procedures showed that the 3- and 4-cluster solutions classified correctly a very similar percentage of participants (93.2 and 94% respectively). Given the very small numbers for Clusters 3 and 4, the 3-cluster solution may be more robust in this analysis. These findings extend previous studies to show that similar to prevalent MCI, within incident MCI samples, the Dysexecutive subtype is the most consistent, with an SCI subgroup also relatively robust. In comparison, other domain specific groups (such as dysnomic, visuospatial or mixed) are less consistent. One of the surprising findings in the current study was the absence of a clear Amnestic group. Those who had been clinically categorized as “Amnestic Single Domain MCI” more likely to be categorized in the SCI cluster, with 59.2% in this category. However, 40.8% of these participants were included within the other clusters, demonstrating a misalignment between the data-driven clustering and the clinical classification. The clinical classification required only one task to be impaired for classification into MCI, whereas the Memory Domain was a composite of two memory tasks. It might be that there is a discrepancy between performance on the two memory tasks, leading to less impaired z-scores for that domain, compared to single tasks. Yet the Machulda et al. Amnestic cluster participants did show a notably more impaired mean z-score (−1.8) than any of the incident MCI clusters in the current study (ranging from −0.50 to −0.84). The cognitive tasks included within the Memory Domain in the current study were similar to Machulda et al. (both contained a word list learning recall task and a visual recall task); however, Machulda et al. included one additional task, a story memory task. Given previous research has shown people with MCI or suspected dementia tend to be less impaired on story memory compared to word list tasks, this is unlikely to drive the lower impairments seen here.²⁷ One other possible explanation is that the z-scores in the current study were stratified by education and gender (in alignment with our previous research focused on MCI/mNCD in the PATH dataset²⁴), whereas this was not the case for the Machulda et al. study, which stratified based on age and sex. Given participants in the MCI group had significantly lower education, as well as a higher proportion of NESB, compared to the cognitively unimpaired participants, standardizing according to education level may have increased the overall z-scores within the incident MCI and mNCD groups. Education appeared to also be significantly lower in the MCI group compared to the cognitively unimpaired group in the Machulda et al. study; however, it is not clear what impact this may have had on the z-scores within the clusters. Further research should investigate the impact of education, NESB, or other demographic factors on cognitive performance and subsequent clustering in MCI subtype studies.

In addition to identifying a 4-cluster solution in MCI, we also identified a 4-cluster solution in our mild NCD subgroup, with distinct similarities and differences. Specifically, the initial three group solution was similar the incident MCI sample, identifying Dysexecutive, SCI, and Mixed subtypes. However, when the 4-cluster solution was identified, the SCI group was split into two groups, resulting in a Dysexecutive group, two SCI groups (one Amnestic/Dysnomic and one Dysexecutive), and again a Mixed subtype with more severe impairments. When comparing the 3- and 4-cluster solutions, slightly more participants were correctly classified using the 3-cluster solution, with less indication of a potential bias of overfitting, compared to the 4-cluster solution. This indicates that the 3-cluster solution may be the superior model in the mNCD sample. These findings extended those of previous studies in MCI samples, to show that in an mNCD sample, Dysexecutive and SCI subgroups are the most consistently identified, with more variability in the remaining cluster profiles^6,13–15. Overall, when comparing the 4-cluster solutions, more participants were correctly classified using the cluster model in incident MCI (94%), with no evidence of bias in over-fitting, compared to the mNCD (90.2%), which also had evidence of over-fitting (correct classification dropped to 87% during cross-validation). One potential consideration is that social cognitive deficits were included as part of the clinical criteria for mNCD, yet these were not captured in the cognitive domains utilized in the current study, resulting in a less robust model. Additionally, mNCD has a broader range of criteria compared to MCI, which could also result in more diverse neuropsychological profiles, less amenable to subtyping. Future research should investigate whether altering the potential cognitive domains to better reflect the mNCD criteria improves the clusters, or whether the model performance is influenced by other differences between mNCD and MCI. Regardless, in the current study, the neuropsychological subtypes were more robust in the incident MCI group.

Our study identified, in both incident MCI and mild NCD groups, one or more mixed clusters with more severe cognitive impairments, rather than the dysnomic cluster found by Machulda et al. One explanation for this pattern is the longer follow up period in the PATH study (4 years), in comparison to the Mayo Clinic Study of Aging (15 months). Although these participants were incident MCI, it is possible that some participants had developed MCI earlier in the four-year window and were therefore more impaired by the time of the assessment, whereas others may have recently developed signs of MCI, leading to their categorization in the SCI cluster. Future research should account for the onset of cognitive symptoms in MCI when studying subtypes, to disentangle the potential confounder of syndrome severity from neuropsychological subtype classification.

In the current study, similar to Machulda et al.,¹⁸ all the SCI clusters identified in the MCI and mNCD samples showed significantly poorer global cognitive function than closely matched cognitively unimpaired comparison groups, as well as domain-specific differences. This indicates that these SCI groups are likely to represent distinct neuropsychological subtypes of MCI, rather than cognitively normal participants who had been falsely categorized in the original clinical categorization. This SCI group differs from the category of Subjective Cognitive Impairment, which is commonly seen as an intermediate between cognitively normal and MCI, because objective cognitive deficits are present. Therefore, this group could represent the earliest signs of MCI and be a potential target for early intervention efforts. One possible confounder in interpreting the SCI findings, however, that must be acknowledged, is that a higher proportion of incident MCI and mild NCD participants were of NESB, and had higher self-reported depression symptoms, in comparison to those who were cognitively unimpaired. For at least some of these SCI participants, NESB and/or depressive symptoms may have impacted on their cognitive performance. Although participants were matched on years of education and gender, NESB could not be utilized as a matching criterion, due to the comparatively small proportion of participants who were of NESB. However, future research should consider NESB or other contributing factors, when identifying cognitive impairments within population datasets. Strengths of the current study include availability of a prospective cohort of incident MCI participants with results from a comprehensive battery of neuropsychological assessment, which allowed for a close replication of aspects of analyses undertaken in Machulda et al.¹⁸ Inclusion of a mNCD sample for comparison was an additional strength. One limitation of the current study was the available sample size for incident MCI was smaller than that of Machulda et al., leading to some small clusters, with potential implications for reliability. An additional limitation is that, for the visuospatial domain, only one neuropsychological test score was available, making this domain likely less reliable than the other three domains. Additionally, similar to all longitudinal cohort studies, the PATH Wave 4 cohort had higher levels of education than those who completed the Wave 1 assessment, indicating that the sample in the current study was less representative of the population than the baseline PATH sample. Finally, neuroimaging data was available for only a subset of PATH participants, resulting in insufficient power for an examination of the underlying neural substrates of the clusters in the current study, and data on cognitive trajectories or biomarkers beyond APOE ε4 carrier status, were also not available to better interpret the clusters.

In conclusion, these findings partially replicated and extended Machulda et al.,¹⁸ showing that empirically derived incident MCI and mNCD clusters can be reliably detected, with the Dysexecutive and SCI subtypes the most consistent. Interestingly, significantly more people clinically categorized as Single Amnestic MCI were also classified as SCI in the MCI group, and in the SCI-Amnestic/Dysnomic cluster in the mNCD group, showing evidence of concurrent validity of an amnestic subtype across the different subtyping approaches, albeit with potentially more subtle memory difficulties in the current study. In contrast, the cognitive impairment features of other subtypes, such as predominantly dysnomic, visuospatial, or mixed etiologies, are less consistent. With the move toward digital machine learning approaches for early screening and detection of MCI,²⁸ empirically derived and validated classifications of MCI/mNCD like these may assist with identifying and promptly addressing early the individualized needs of each patient, as well as better predicting progression to functional difficulties and dementia.

Supplemental Material

sj-docx-1-alz-10.1177_13872877251415023 - Supplemental material for Neuropsychological subtypes of incident mild cognitive impairment and mild neurocognitive disorder in a population-based cohort of older adults

Supplemental material, sj-docx-1-alz-10.1177_13872877251415023 for Neuropsychological subtypes of incident mild cognitive impairment and mild neurocognitive disorder in a population-based cohort of older adults by Sophie Claire Andrews, Ranmalee Eramudugolla, Craig Sinclair, Moyra Elizabeth Mortby, Nicolas Cherbuin and Kaarin Jane Anstey in Journal of Alzheimer's Disease

Footnotes

Acknowledgements

The authors thank the PATH interviewers, project team, participants, and Chief Investigators from the grants that funded these waves: Helen Christensen, Bryan Rodgers, Peter Butterworth, Simon Easteal, Andrew MacKinnon. The authors also thank Sacha Dubois and Bruce Weaver from Lakehead University, Ontario for providing guidance on conducting some of the statistical analyses in this manuscript.

ORCID iDs

Sophie Claire Andrews

Ranmalee Eramudugolla

Craig Sinclair

Moyra Elizabeth Mortby

Ethical considerations

This study was conducted in accordance with the Declaration of Helsinki, and approved by the Australian National University Human Research Ethics Committee.

Consent to participate

All participants provided written informed consent.

Consent for publication

Not applicable.

Author contribution(s)

Sophie Claire Andrews: Conceptualization; Data curation; Formal analysis; Methodology; Writing – original draft.

Ranmalee Eramudugolla: Conceptualization; Data curation; Methodology; Project administration; Writing – review & editing.

Craig Sinclair: Data curation; Formal analysis; Methodology; Writing – review & editing.

Moyra Elizabeth Mortby: Methodology; Writing – review & editing.

Nicolas Cherbuin: Funding acquisition; Methodology; Writing – review & editing.

Kaarin Jane Anstey: Conceptualization; Funding acquisition; Methodology; Project administration; Supervision; Writing – review & editing.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: PATH Waves 3 and 4 were funded by NHMRC Grants 418039 and 1002160. Kaarin Anstey is supported by an ARC Laureate Fellowship (FL190100011). Sophie Andrews is supported by an ARC DECRA Fellowship (DE210101138). Moyra Mortby is supported by a Dementia Australia Research Foundation – Dementia Centre for Research Collaboration Mid-Career Research Fellowship.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The data supporting the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Supplemental material

Supplemental material for this article is available online.

References

Winblad

Palmer

Kivipelto

, et al. Mild cognitive impairment – beyond controversies, towards a consensus: report of the international working group on mild cognitive impairment. J Intern Med 2004; 256: 240–246.

American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders. 5th ed. Washington D.C.: American Psychiatric Association, 2013.

Bai

Chen

Cai

, et al. Worldwide prevalence of mild cognitive impairment among community dwellers aged 50 years and older: A meta-analysis and systematic review of epidemiology studies. Age Ageing 2022; 51: afac173.

Petersen

. Mild cognitive impairment as a diagnostic entity. J Intern Med 2004; 256: 183–194.

Gabel

Foster

Heidebrink

, et al. Validation of consensus panel diagnosis in dementia. Arch Neurol 2010; 67: 1506–1512.

Edmonds

Delano-Wood

Clark

, et al. Susceptibility of the conventional criteria for mild cognitive impairment to false-positive diagnostic errors. Alzheimers Dement 2015; 11: 415–424.

Young

PNE

Estarellas

Coomans

, et al. Imaging biomarkers in neurodegeneration: Current and future practices. Alzheimers Res Ther 2020; 12: 49.

Khoury

Ghossoub

. Diagnostic biomarkers of Alzheimer’s disease: A state-of-the-art review. Biomark Neuropsychiatry 2019; 1: 100005.

Hansson

Edelmayer

Boxer

, et al. The Alzheimer’s Association appropriate use recommendations for blood biomarkers in Alzheimer’s disease. Alzheimers Dement 2022; 18: 2669–2686.

10.

Wang

Bakulski

Karvonen-Gutierrez

, et al. Blood-based biomarkers for Alzheimer’s disease and cognitive function from mid- to late life. Alzheimers Dement 2024; 20: 1807–1814.

11.

Weintraub

. Neuropsychological assessment in dementia diagnosis. Continuum (Minneap Minn) 2022; 28: 781–799.

12.

Edmonds

Weigand

Hatton

, et al. Patterns of longitudinal cortical atrophy over 3 years in empirically derived MCI subtypes. Neurology 2020; 94: e2532–e2544.

13.

Bondi

Edmonds

Jak

, et al. Neuropsychological criteria for mild cognitive impairment improves diagnostic precision, biomarker associations, and progression rates. J Alzheimers Dis 2014; 42: 275–289.

14.

Clark

Delano-Wood

Libon

, et al.

Are empirically-derived subtypes of mild cognitive impairment consistent with conventional subtypes?

J Int Neuropsychol Soc 2013; 19: 635–645.

15.

Libon

Xie

Eppig

, et al. The heterogeneity of mild cognitive impairment: A neuropsychological analysis. J Int Neuropsychol Soc 2010; 16: 84–93.

16.

Thomas

Edmonds

Delano-Wood

, et al. Longitudinal trajectories of informant-reported daily functioning in empirically defined subtypes of mild cognitive impairment. J Int Neuropsychol Soc 2017; 23: 521–527.

17.

Thomas

Bangen

Weigand

, et al. Cognitive heterogeneity and risk of progression in data-driven subtle cognitive decline phenotypes. J Alzheimers Dis 2022; 90: 323–331.

18.

Machulda

Lundt

Albertson

, et al. Neuropsychological subtypes of incident mild cognitive impairment in the Mayo Clinic Study of Aging. Alzheimers Dement 2019; 15: 878–887.

19.

Machulda

Lundt

Albertson

, et al. Cortical atrophy patterns of incident MCI subtypes in the Mayo Clinic Study of Aging. Alzheimers Dement 2020; 16: 1013–1022.

20.

Machulda

Lundt

Mester

, et al. White matter changes in empirically derived incident MCI subtypes in the Mayo Clinic Study of Aging. Alzheimers Dement (Amst) 2021; 13: e12269.

21.

Stokin

Krell-Roesch

Petersen

, et al. Mild neurocognitive disorder: An old wine in a new bottle. Harv Rev Psychiatry 2015; 23: 368–376.

22.

Sachs-Ericsson

Blazer

. The new DSM-5 diagnosis of mild neurocognitive disorder and its relation to research in mild cognitive impairment. Aging Ment Health 2015; 19: 2–12.

23.

Anstey

Christensen

Butterworth

, et al. Cohort profile: The PATH through life project. Int J Epidemiol 2012; 41: 951–960.

24.

Eramudugolla

Mortby

Sachdev

, et al. Evaluation of a research diagnostic algorithm for DSM-5 neurocognitive disorders in a population-based cohort of older adults. Alzheimers Res Ther 2017; 9: 15.

25.

Andrews

Eramudugolla

Velez

, et al. Validating the role of the Australian National University Alzheimer’s Disease Risk Index (ANU-ADRI) and a genetic risk score in progression to cognitive impairment in a population-based cohort of older adults followed for 12 years. Alzheimers Res Ther 2017; 9: 16.

26.

Tabachnick

Fidell

. Using multivariate statistics. Sydney, Australia: Pearson Education, 2018, pp. 299–345.

27.

Wong

Jeffers

Bell

, et al. Story memory impairment rates and association with hippocampal volumes in a memory clinic population. J Int Neuropsychol Soc 2022; 28: 611–619.

28.

Sabbagh

Boada

Borson

, et al. Rationale for early diagnosis of mild cognitive impairment (MCI) supported by emerging digital technologies. J Prev Alzheimers Dis 2020; 7: 158–164.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB