Exploring heterogeneity in mild cognitive impairment

Abstract

Background

Mild cognitive impairment (MCI) is a heterogeneous diagnostic entity, without a clear prognosis, often accompanied by psychiatric symptomatology and physical frailty.

Objective

Understanding the heterogeneity within MCI is a critical step in improving the early detection of cognitive decline and developing effective interventions.

Methods

Cross-sectional multivariate latent mixture analyses of data from patients evaluated between 2015 and 2019, who were routinely entered into a multidisciplinary database for research purposes. A sample of 538 community-dwelling older adults drawn from a large academic medical center, referred from within the Department of Neurology (63.7% Female, M_age= 67.8, SD_age= 10.6). Participants completed comprehensive neuropsychological assessments, psychiatric symptom measures, and frailty evaluations.

Results

Latent profile analyses supported five profiles of cognitive impairment: At-Risk, Pre-MCI, Amnestic MCI, Multiple Domain MCI, and Major Cognitive Impairment. The inclusion of concomitant psychiatric symptoms and frailty criteria revealed two additional profiles: Psychiatric/Frail, without cognitive impairments, and Multiple Cognitive Domains/Psychiatric/Frail. Critically, 55% of those classified as Healthy based on cognitive data alone were reclassified. Significant profile-wise differences emerged across auxiliary variables of brief cognitive screening, sociodemographics, and medical and psychosocial risk.

Conclusions

Results highlight heterogeneity represented by neurologic patients referred for neuropsychological evaluation that include key physical and emotional symptoms known to increase the risk of cognitive decline. Findings are in alignment with more recent research suggesting that the traditional paradigm cognitive impairment may need to be expanded to improve diagnostic accuracy and to develop more tailored, precision-driven interventions.

Keywords

Alzheimer's disease dementia latent profile analysis mild cognitive impairment physical frailty psychiatric symptoms

Introduction

The diagnosis of mild cognitive impairment (MCI) does not carry the same meaning for all patients. For some, MCI may indicate a diminished state of cognition that remains unchanged or improved by follow-up. For others, the diagnosis represents the intermediary stage between normal cognition and advancing cognitive decline that progresses over time.¹ Indeed, prospective studies reveal a striking variability in prognosis among those identified with MCI,^2,3 with as many as 50% remaining unchanged or reverting to normal after five years. Part of this heterogeneity is due to methodological differences across studies with regard to follow up period, participant inclusion criteria, and sample characteristics including age, gender, vascular risk and medical co-morbidities, ethnicity, and genetic risk.⁴

Efforts to characterize the heterogeneity of MCI have focused on the type and degree of cognitive impairment,⁵ most typically by classifying patients based on the presence or absence of memory impairments (i.e., amnestic versus non-amnestic), or by the number of cognitive domains impaired (i.e., single versus multiple domain). Yet, more detailed testing that extends beyond conventional screening batteries has shown variability within these subtypes,⁶ calling into question their clinical utility⁷ and raising the possibility of false-positive errors. For example, cluster analyses of the Alzheimer's Disease Neuroimaging Initiative cohort revealed one-third of adults who had been formally diagnosed with MCI based on a brief screening, clinician rating, and subjective memory complaints, performed within normal limits on a more comprehensive neuropsychological assessment and showed no evidence of decline when reexamined five years later.⁵ This study and others^5,6,8–12 demonstrate that the likelihood of misclassification and false positive errors is increased when diagnosing MCI from brief assessments and conventional diagnostic criteria compared to more comprehensive, quantitatively driven approaches. As a result, the prognostic utility of the MCI diagnosis is limited, clouding decision making for clinical trials, and producing a potentially false narrative regarding what lies ahead for patients and their caregivers.

While these studies have improved our understanding of the spectrum of mild cognitive impairments, research in this area remains constrained by additional methodological limitations. The diagnostic criteria of MCI have been criticized as insensitive to subtypes of impairment and prone to false-positive errors.¹³ Further, the number and breadth of cognitive measures widely differs between studies seeking to understand heterogeneity in cognitive impairment.^14,15 Another limitation is that study samples are often prescreened for MCI,¹¹ a procedure that can introduce selection bias.

A critical shortcoming that has received less attention is a tendency to view age-related cognitive changes as isolated processes. As individuals age cognitively, they are also likely to experience concomitant physical and emotional changes that are well-known to impact cognition but are not part of a formal assessment. For example, data from the UK Biobank, a population-based study of middle-aged and older adults, found nearly half of those sampled demonstrate at least one indicator of frailty.¹⁶ Moreover, frail adults demonstrated a two-fold increase in the risk of developing dementia over a five-year period.¹⁶ Both frailty and subsequent disengagement in physical activities have been shown to be prodromal indicators of MCI and dementia.^16,17 Further, psychiatric symptoms and conditions exacerbate cognitive deficits, are more likely to arise in those with MCI, and increase the risk of conversion from MCI to dementia.¹⁸

Given cognition in older age is best understood in the context of cognitive and non-cognitive changes that take place over the life course, then a framework that considers an in depth and refined study of individual contributors may be better able to characterize the heterogeneity of MCI. Further, this approach offers potential to inform diagnostic decision making and further clarify the clinical heterogeneity within and between MCI subtypes.^13,15,19,20

The present study sought to characterize the heterogeneity of age-related cognitive impairment using a latent profile analysis (LPA), an innovative analytic technique rarely used in aging research. Emergent profiles are not based on preconceived hypotheses or examiner biases as to how to best conceptualize cognitive subgroups. Rather, they are data driven and able to reveal subpopulations hidden in multivariate data that may otherwise go undiscovered. Using this analytic approach, cognitive profiles may be identified from a range of metrics, including important non-cognitive risk factors linked to decline, such as physical frailty and psychiatric distress. To this end, we leveraged a diverse sample of older adults referred for evaluation at a neuropsychological clinic embedded within a large academic medical center. The aims of this study were to 1) identify latent profiles of neuropsychological impairment; 2) further characterize these profiles by including physical, psychiatric, and psychosocial variables; and 3) examine other noncognitive correlates of the latent profiles. These aims should advance the goal of understanding heterogeneity within those suspected to be experiencing cognitive decline.

Methods

Participants

The final sample was comprised of 538 adults, without histories of a diagnosed neurocognitive disorder, referred for neuropsychological evaluation to the University of Miami's Department of Neurology. Patients were referred based on the discretion of the referring neurologist, often due to subjective cognitive complaints and/or to guide differential diagnosis. Few participants were removed from inclusion in this study due to stroke (n = 5), movement disorders (n = 4), or dementia (n = 3). Sample characteristics are reported in Table 1. Participants ranged from 22 to 92 years (M = 68, SD = 11), were largely White (89%), female (64%), and English-speaking (67%). Half were Hispanic/Latino (49%).

Table 1.

Sample characteristics and descriptive statistics of demographics and categorical frailty criteria.

Continuous variables	M	(SD)
Age	67.8	(10.6)
Years of Education	14.9	(3.3)
Body Mass Index	27.8	(5.6)
Mini Mental Status Exam	26.7	(3.1)
Categorical Variables	N	(%)
Gender (Female)	230	(63.7)
Race
Caucasian	320	(88.6)
African American	21	(5.8)
Asian American	8	(2.2)
Other/Unknown	10	(2.8)
Ethnicity (Hispanic/Latino)	177	(49.0)
Language (English)	241	(66.8)
Frailty Criteria	N	(%)
Exhaustion	97	28.0
Weight Loss	44	12.4
Inactivity	26	7.7
Slowness	59	17.0
Weakness	209	60.4

Measures

Cognition

Participants completed a battery of 16 neuropsychological measures with well-established validity, assessing five cognitive domains: Language, Visuospatial Reasoning, Learning & Memory, Processing Speed, and Executive Functioning (Table 2). Raw scores were converted to z-scores relative to published normative data and coded such that higher z-scores corresponds to better performance. As it is not feasible to adequately describe the development, format, and validity of each measure herein, we refer interested readers to a comprehensive compendium of neuropsychological measures²¹ and the respective test manuals.

Table 2.

Descriptive statistics for continuous indicators of latent profile analyses in the full sample.

Measure	Label	M	SD	Skew	Kurtosis
Language
Boston Naming Test	BNT	−0.66	1.23	0.23	−0.35
National Adult Reading Test/Word Accentuation Test	NART/WAT	0.49	0.73	−0.89	1.85
DKEFS – Letter Fluency	LF	−0.53	1.20	0.01	−0.35
DKEFS – Category Fluency	CF	−0.69	1.27	−0.17	−0.29
Visuospatial Reasoning
WAIS-IV Block Design	BD	−0.19	0.94	−0.02	0.15
Judgement of Line Orientation	JLO	−0.34	1.16	−0.42	−0.48
Rey-Osterrieth Complex Figure – Copy	ROCF – C	−1.25	1.28	0.40	−0.40
Learning & Memory
Logical Memory – Immediate Recall	LM I	−0.21	1.16	−0.05	−0.28
Logical Memory – Delayed Recall	LM II	−0.35	1.37	0.25	0.67
California Verbal Learning Test – Trials 1–5	CVLT – T1–5	−0.77	1.18	−0.04	−0.22
California Verbal Learning Test – List B	CVLT – B	−0.57	1.10	0.31	0.69
California Verbal Learning Test – Short Delay	CVLT – SD	−0.86	1.17	0.09	−0.42
California Verbal Learning Test – Long Delay	CVLT – LD	−1.01	1.27	−0.06	−0.74
Rey-Osterrieth Complex Figure – Recall	ROCF – D	−1.35	1.34	0.53	−0.46
Rey-Osterrieth Complex Figure – Recognition	ROCF – R	−1.19	1.38	−0.21	−0.43
Brief Visual Memory Test – Immediate Recall	BVMT – IR	−1.61	1.22	0.74	−0.28
Brief Visual Memory Test – Delayed Recall	BVMT – DR	−1.38	1.41	0.50	−0.88
Processing Speed
WAIS-IV Coding	Coding	−0.23	1.05	0.07	0.01
Grooved Pegboard (Dominant Hand)	GPB (DH)	−0.99	1.10	−0.02	0.13
Grooved Pegboard (Dominant Hand)	GPB (NDH)	−0.96	0.99	0.09	0.77
Trail Making Test: Part A	TMT-A	−0.69	1.31	−0.13	−0.35
Executive Functions
Wisconsin Card Sorting Test – Correct	WCST – C	−1.02	1.60	−0.52	−1.14
Wisconsin Card Sorting Test – Errors	WCST – E	−0.17	1.46	0.51	0.25
Digit Span – Forward	DS – F	−0.31	1.03	0.29	−0.04
Digit Span – Backward	DS – B	−0.12	1.03	0.30	0.15
Digit Span – Sequencing	DS – S	−0.29	0.98	−0.07	0.17
Trail Making Test: Part B	TMT-B	−0.99	1.33	−0.25	−0.83
Psychiatric Symptoms
Beck Depression Inventory – II	BDI-II	11.42	9.90	1.13	0.79
Beck Anxiety Inventory	BAI	8.34	9.34	1.68	3.27
Apathy Evaluation Scale	AES	33.05	11.13	0.85	0.56
Fatigue Symptom Index	FSI	16.09	17.18	1.01	0.03

All neuropsychological measures were standardized to z-scores based on available normative data.

Language was assessed via five tasks. Word reading was assessed with the National Adult Reading Test (NART)²² in English-speaking participants or the Word Accentuation Test (WAT)²³ in Spanish-speaking participants. Semantic retrieval and confrontation naming was measured via the 60-item version of the Boston Naming Test (BNT).²⁴ Phonologically and semantically guided verbal fluency were assessed by the Letter Fluency and Category Fluency tests of the Delis-Kaplan Executive Function System (D-KEFS),^25,26 respectively.

Visuospatial Reasoning was assessed with three tasks. The Block Design subtest of the Wechsler Adult Intelligence Scales – 4^th Edition (WAIS-IV)^21,27 measured visuospatial manipulation and reasoning. The Judgment of Line Orientation²⁸ task was included to measure visuospatial perception and orientation. Constructional praxis and visual organization and planning was measured by the copy trial of the Rey–Osterrieth Complex Figure (ROCF)^29,30 test.

Learning & Memory was represented by ten scores derived from four tests. Verbal learning and memory was measured by immediate and delayed recall of the Logical Memory test of the Wechsler Memory Scales,³¹ a brief story memory task, and total learning of a word list over five repetitions (i.e., Trials 1–5), immediate memory of a distractor list (List B) and short and long delayed recall scores of the California Verbal Learning Test- 2^nd Edition (CVLT).³² Visuospatial learning and memory was measured by the immediate and delayed recall trials of the Brief Visual Memory Test – Revised (BVMT),³³ and the delayed recall and recognition trials of the ROCF.

Processing Speed was measured by four timed tasks of visual scanning and psychomotor speed: completion time in seconds on the letter sequencing subtest of the Trail Making Test (TMT-A),²⁵ the number of items transcribed on the Coding subtest of the WAIS-IV,²⁷ and completion time of the dominant hand (DH) and non-dominant hand (NDH) trials of the Grooved Pegboard (GPB)³⁴ test.

Executive Functioning was represented by six scores derived from three tasks. Working memory, simple attention, and mental control was assessed via the Digit Span Forward, Backward, and Sequencing trials of the WAIS-IV,²⁷ which required participants to mentally maintain and manipulate strings of digits. Identifying, maintaining, and updating mental sets was measured by the correct number of categories sorted and total perseverative errors of the Wisconsin Card Sorting Test (WCST).³⁵ Set-shifting and self-monitoring was assessed by completion time on Part B of the Trail Making Test (TMT-B).²⁵

Psychiatric symptoms

Psychiatric symptoms were assessed with four validated self-report measures spanning depressive symptoms, anxiety, apathy, and chronic fatigue. Depressive symptoms and anxiety were assessed via the 21-tem Beck Depression Inventory – 2^nd edition (BDI)³⁶ and the 21-item Beck Anxiety Inventory (BAI),³⁷ respectively. Apathy, a diminished motivation to engage in goal-directed behavior, was evaluated by the 18-item self-report version of the Apathy Evaluation Scale (AES).³⁸ The presence, severity, and interference resulting from fatigue-related symptoms was measured via the 14-item Fatigue Symptom Inventory (FSI).³⁹ Clinical thresholds were determined from extant literature when available: BDI-II scores above 13,⁴⁰ BAI scores above 7,³⁷ and AES scores above 35.³⁸

Frailty

Frailty was assessed in line with the Fried Frailty Phenotype,⁴¹ a widely used and validated set of five criteria: Weakness, Slowness, Physical Inactivity, Exhaustion, and Weight Loss. Functional dependence was rated by the referring physician prior to neuropsychological workups. Weakness was measured using a Jamar Hydraulic Hand Dynamometer and calculated as the average score across three trials in the dominant hand. Slowness was operationalized as the average time (seconds) to walk a 15-foot straight line at routine pace across three trials. Physical Inactivity was measured using the Community Healthy Activities Model Program for Seniors (CHAMPS) Questionnaire for Older Adults,⁴² a self-report inventory of frequency and duration of engagement in a variety of listed physical activities (e.g., walking, swimming, weightlifting, stretching, gardening) per week, over the last four weeks. Average kilocalories expended (Kcals) per week were estimated using a weight-adjusted algorithm. Exhaustion was assessed by two questions drawn from the Center for Epidemiological Studies Depression Scale:⁴³ frequency in the last week that 1) “I could not get going”, and 2) “I felt that everything I did was an effort”. Weight Loss was quantified as the K-coefficient, representing the percentage of weight loss from the past year.

Analyses

Latent profile analysis (LPA), a combination of latent variable and mixture modeling techniques, was utilized to identify subgroups underlying neuropsychological data. Broadly, mixture analyses are those which aim to reveal groups, often referred to as clusters, classes, or profiles, from a multivariate set of data where a group classification variable is not known.⁴⁴ It was expected that this approach would reveal neuropsychological profiles not observed through either diagnostic or domain-wise classification.

LPA is a subset of latent variable models, such as confirmatory factor analysis, in which relationships between observed variables (i.e., indicators) are modeled as the effect of an “latent” underlying construct.⁴⁵ In LPA, the latent construct is the categorical classes, which are posited to explain the relationship among observed variables. Latent variable models offer the benefit of being “model-based” techniques, which typically require several assumptions to be met. In the case of LPA, assumptions include the distribution of indicators is approximately normal, that the within-group covariances between indicators are zero (i.e., local independence), and that indicator variances are invariant across profiles, while indicator means are necessarily non-invariant. The benefit of these requirements is that LPA facilitates testing the degree of fit between observed data and model-implied data, as well as comparisons of model fit across solutions.

Models were estimated with robust maximum likelihood using the accelerated EM algorithm,⁴⁶ which allowed for implementation of full information maximum likelihood to account for missing data assumed to be missing at random, conditional on covariates. This method is accommodating of non-normality in indicators, and allows for the simultaneous estimation of continuous and categorical indicators. Model convergence was dependent on minimal change in loglikelihood across iterations. Analyses were conducted in Mplus Version 8.8. The final number of estimated profiles was determined based on several considerations: parametric and bootstrapped likelihood ratio tests (LRT), model fit metrics, entropy, local independence, class representation, and model replicability.^45,47,48 When deciding on the “optimal” number of profiles to extract, recommendations are to evaluate model fit based on multiple indices: the Bayesian information criteria (BIC), the sample-adjusted BIC (SABIC), the classification likelihood information criteria (CLC), and the integrated classification likelihood (ICL-BIC).⁴⁵ Model entropy reflects the precision of model classification, with values closer to 1 indicating greater within group homogeneity and thus better model reliability.⁴⁹ Class representation was deemed reasonable if all classes were estimated to represent at least 25 participants (∼5% of the sample).

Within LPA, the joint density of indicators is represented as the mixture of the specified number of profile densities.⁴⁴ Model parameters thus reflect the properties of these densities (i.e., means/thresholds, variances, & covariances) as well as the proportion of participants assigned to each profile. For each participant, profile assignment is probabilistic based on similarity to profile parameters.⁵⁰ Indicator means are estimated within profiles, allowing for the comparison of specific indicators across classes. For continuous indicators, estimates represent the statistical average (i.e., means). For dichotomous indicators, estimates represent thresholds, the natural logarithm of the odds of indicator endorsement. To aid interpretation, thresholds were converted to probabilities. Indicator variances are constrained to be equal across profiles, analogous to the assumptions of homogeneity of variance.⁵¹ Covariances, both indicator-to-indicator and indicator-to-profile, are constrained to zero, analogous to the assumption of local independence.^44,51 Taken together, within the unmodified LPA framework, latent profiles may differ in indicator means while holding indicator variance and covariances constant across profiles.⁵²

Following identification of the optimal model(s), parameters of neuropsychological scores, psychiatric symptom measures, and frailty criteria were compared across profiles. Cognitive impairments at the profile-level were identified by the Jak/Bondi criteria¹⁴ to facilitate interpretation and contextualization within the extant literature. Specifically, profiles were considered impaired when: 1) performance on two or more neuropsychological tasks falling one-and-a-half standard deviations or more below normative expectancy within the same domain, or 2) performance on one neuropsychological task one-and-a-half standard deviations or more below normative expectancy, in three different domains. Extant research suggests that classification determined from these criteria may be more indicative of underlying neuropathology.⁵³ Psychiatric symptoms were considered elevated for a profile when the estimated mean was above clinical cut-offs. Frailty criteria were considered elevated when the estimated proportion of participants meeting the criteria exceeded 50%.

Derived classes of the best fitting solution then were compared across sociodemographic variables and medical covariates (e.g., body mass index (BMI), blood pressure, sleep quality, etc.). Such variables were treated as auxiliary covariates, as inclusion of covariates in class derivation can inadvertently change class composition, classification, and interpretation. Covariates were examined with the modified BCH procedure to compare distal outcomes (i.e., auxiliary variables) across extracted profiles.^54,55 This approach estimates profile means and standard errors for auxiliary covariates, which also facilitates significance testing, without influencing profile parameters.

Results

Neuropsychological profiles

The first set of latent profile analyses were conducted with only cognitive indicators. Fit metrics are reported in Table 3. Attempts to estimate more than six profiles consistently resulted in unstable models, including difficulty replicating model parameters and very low within-class representation (n < 5). Consequently, solutions indicating two to six profiles were examined. Entropy was high across solutions (≥ .90), indicating consistent classification in all considered models. The six-profile solution was selected based on the bootstrapped LRT, BIC, SABIC, CLC, and ICL-BIC.

Table 3.

Comparisons of latent profile analyses (LPA) model fit statistics based on the number of classes derived.

Classes	LMR LRT	p	BLRT	p	BIC	SABIC	Entropy	CLC	ICL-BIC
Cognitive Indicators
Two	2822	<0.001	2838	<0.001	40,579	40,310	0.92	40,107	80,686
Three	951	0.012	956	<0.001	39,805	39,444	0.91	39,196	79,001
Four	489	0.265	492	<0.001	39,495	39,041	0.90	38,741	78,237
Five	342	0.543	334	<0.001	39,333	38,787	0.90	38,430	77,763
Six	283	0.369	285	<0.001	39,230	38,592	0.90	38,168	77,398
Cognitive, Psychiatric, and Frailty Indicators
Two	2903	<0.001	2915	<0.001	53,899	53,559	0.92	53,291	107,190
Three	952	0.104	956	<0.001	53,182	52,721	0.91	52,379	105,560
Four	670	0.243	672	<0.001	52,748	52,166	0.92	51,716	104,464
Five	473	0.115	475	<0.001	52,511	51,809	0.93	51,252	103,763
Six	319	0.061	321	<0.001	52.429	51.607	0.94	50,924	103,353
Seven	253	0.402	254	<0.001	52,413	51,471	0.93	50,695	103,108
Eight	221	0.712	222	<0.001	52,429	51,366	0.92	50,493	102,923

Bolded values indicate the best fit within each criterion.

Profile means are reported in Supplemental Table 1. The first profile represented a Healthy group without impairments on any cognitive task. The second profile demonstrated select weaknesses, but without clinical impairments, and thus labeled At-Risk. The third profile demonstrated an isolated impairment on one visuospatial memory test, with preserved scores on other memory measures, and thus labeled Pre-MCI. The fourth and fifth profiles demonstrated more consistent cognitive impairments. The fourth profile was impaired on all learning & memory tasks except LM I and CVLT List B, and was labeled Mildly Impaired, Amnestic. Profile five, Mildly Impaired, Multiple Domains, demonstrated fewer stark impairments within the memory domain but included impairments in processing speed and executive functions. The sixth profile demonstrated impairments across all cognitive domains and nearly all tasks, representing a Major Cognitive Impairment profile.

Integrating frailty and psychiatric symptoms

With the inclusion of psychiatric and frailty indicators, the optimal number of classes increased to eight. Attempting to extract nine or more profiles produced non-replicable and unstable solutions, thus two-to-eight profiles were evaluated (Table 3). Entropy remained high across all solutions. The LMR LRT supported a two-profile solution, and the BIC supported a seven-profile solution; however, an eight-profile solution was selected based on the bootstrapped LRT, SABIC, CLC, and ICL-BIC as well as the interpretability of the additional profiles.

The eight-profile solution replicated the prior six-profile solution while establishing two additional profiles not revealed by neuropsychological indicators alone (Table 4). The Healthy, At-Risk, Pre-MCI, Mildly Impaired, and Major Cognitive Impairment profiles demonstrated the same patterns of neuropsychological performance as in the six- profile solution, supporting consistency in the cognitive profiles extracted across the six-profile and eight-profile solutions.

Table 4.

Class-wise comparisons of latent classes with inclusion of psychiatric and frailty indicators.

	Healthy(N = 73)	Psych(N = 47)	At-Risk(N = 74)	Pre-MCI(N = 87)	aMCI(N = 82)	MD MCI(N = 44)	MCI+(N = 64)	Major(N = 59)
Language
NART/WAT	0.95	0.78	0.38	0.49	0.69	0.01	0.26	−0.11
BNT	0.35	−0.09	−0.84	−0.23	−0.70	−1.34	−1.12	−1.70
LF	0.22	0.09	−0.80	0.08	−0.52	−1.36	−0.85	−1.53
CF	0.24	−0.01	−0.70	−0.05	−1.18	−1.32	−0.96	−1.92
Visuospatial Reasoning
BD	0.50	0.21	−0.17	−0.03	−0.03	−0.80	−0.71	−1.20
JLO	0.24	0.04	−0.10	−0.22	−0.41	−0.86	−0.79	−1.09
ROCF – C	0.11	0.09	−1.08	−0.83	−2.27	−1.14	−1.63	−2.63
Learning & Memory
LM I	0.82	0.85	−0.18	0.00	−0.72	−0.43	−0.36	−1.78
LM II	0.70	1.10	−0.05	−0.11	−1.45	−0.40	−0.38	−2.14
CVLT – T1–5	0.22	0.30	−0.16	−0.49	−1.74	−1.00	−0.96	−2.43
CVLT – B	0.38	0.05	−0.27	−0.47	−1.04	−0.79	−0.65	−1.79
CVLT – SD	0.23	0.27	−0.30	−0.71	−2.13	−0.86	−0.99	−2.24
CVLT – LD	0.23	0.27	−0.31	−0.90	−2.41	−1.04	−1.15	−2.61
ROCF – D	0.07	0.10	−1.33	−0.93	−2.31	−1.39	−1.73	−2.76
ROCF – R	−0.18	−0.25	−0.90	−0.91	−1.89	−1.27	−1.48	−2.29
BVMT – IR	0.28	−0.91	−0.85	−2.04	−2.50	−2.30	−2.01	−2.85
BVMT – DR	0.80	−0.35	−0.36	−1.91	−2.71	−2.06	−1.78	−2.81
Processing Speed
Coding	0.54	0.13	−0.17	0.32	−0.03	−1.03	−0.87	−1.53
GPB (DH)	−0.23	−0.96	−0.62	−0.67	−1.07	−1.64	−1.49	−1.83
GPB (NDH)	−0.24	−0.89	−0.71	−0.69	−0.88	−1.71	−1.64	−1.56
TMT – Part A	0.06	−0.07	−0.62	−0.27	−0.57	−1.80	−1.06	−1.82
Executive Functioning
DS – F	0.30	0.03	−0.61	−0.04	0.10	−0.98	−0.82	−1.07
DS – B	0.39	0.55	−0.55	0.30	0.37	−0.96	−0.56	−1.10
DS – S	0.27	0.13	−0.55	0.08	−0.19	−1.12	−0.71	−1.68
WCST – C	−0.67	−0.30	−0.45	−0.82	−1.68	−1.71	−1.27	−2.31
WCST – E	0.78	−0.16	−0.27	0.55	−0.15	−1.24	−0.74	−1.47
TMT – Part B	−0.04	0.00	−0.77	−0.27	−1.13	−2.51	−1.43	−2.73
Psychiatric Symptomatology
BDI-II	5.34	21.35	9.22	7.35	7.12	7.74	28.16	10.62
BAI	4.11	14.38	5.86	4.93	4.57	4.35	23.09	10.00
AES	29.13	35.57	30.60	27.43	32.26	33.99	48.24	34.22
FSI	6.44	30.53	16.22	7.85	7.30	11.94	41.01	19.13
Frailty Criteria
Exhaustion	5%	63%	33%	0%	26%	19%	79%	19%
Weight Loss	8%	16%	10%	12%	18%	0%	19%	11%
Inactivity	0%	10%	8%	7%	12%	12%	7%	12%
Slowness	0%	15%	5%	11%	29%	5%	43%	37%
Weakness	34%	78%	50%	66%	61%	63%	79%	71%

Bolded values indicate clinical significance.

Table 4.

Class-wise comparisons of latent classes with inclusion of psychiatric and frailty indicators.

	Healthy(N = 73)		Psych(N = 47)		At-Risk(N = 74)		Pre-MCI(N = 87)		aMCI(N = 82)		MD MCI(N = 44)		MCI+(N = 64)		Major(N = 59)
	Est.	S.E.	Est.	S.E.	Est.	S.E.	Est.	S.E.	Est.	S.E.	Est.	S.E.	Est.	S.E.	Est.	S.E.
Language
NART/WAT	0.95	(0.08)	0.78	(0.15)	0.38	(0.12)	0.49	(0.08)	0.69	(0.11)	0.01	(0.19)	0.26	(0.11)	−0.11	(0.16)
BNT	0.35	(0.20)	−0.09	(0.17)	−0.84	(0.18)	−0.23	(0.14)	−0.70	(0.28)	−1.34	(0.20)	−1.12	(0.16)	−1.70	(0.15)
LF	0.22	(0.13)	0.09	(0.20)	−0.80	(0.14)	0.08	(0.16)	−0.52	(0.29)	−1.36	(0.18)	−0.85	(0.17)	−1.53	(0.19)
CF	0.24	(0.15)	−0.01	(0.18)	−0.70	(0.17)	−0.05	(0.13)	−1.18	(0.24)	−1.32	(0.21)	−0.96	(0.15)	−1.92	(0.21)
Visuospatial Reasoning
BD	0.50	(0.13)	0.21	(0.15)	−0.17	(0.15)	−0.03	(0.15)	−0.03	(0.13)	−0.80	(0.18)	−0.71	(0.14)	−1.20	(0.49)
JLO	0.24	(0.11)	0.04	(0.20)	−0.10	(0.18)	−0.22	(0.16)	−0.41	(0.14)	−0.86	(0.25)	−0.79	(0.21)	−1.09	(0.40)
ROCF – C	0.11	(0.30)	0.09	(0.21)	−1.08	(0.22)	−0.83	(0.17)	−2.27	(0.11)	−1.14	(0.24)	−1.63	(0.16)	−2.63	(0.10)
Learning & Memory
LM I	0.82	(0.12)	0.85	(0.18)	−0.18	(0.16)	0.00	(0.12)	−0.72	(0.23)	−0.43	(0.29)	−0.36	(0.18)	−1.78	(0.19)
LM II	0.70	(0.12)	1.10	(0.26)	−0.05	(0.17)	−0.11	(0.13)	−1.45	(0.28)	−0.40	(0.30)	−0.38	(0.20)	−2.14	(0.22)
CVLT – T1–5	0.22	(0.15)	0.30	(0.21)	−0.16	(0.27)	−0.49	(0.14)	−1.74	(0.17)	−1.00	(0.15)	−0.96	(0.14)	−2.43	(0.12)
CVLT – B	0.38	(0.28)	0.05	(0.22)	−0.27	(0.21)	−0.47	(0.15)	−1.04	(0.17)	−0.79	(0.19)	−0.65	(0.14)	−1.79	(0.16)
CVLT – SD	0.23	(0.14)	0.27	(0.22)	−0.30	(0.23)	−0.71	(0.14)	−2.13	(0.11)	−0.86	(0.18)	−0.99	(0.14)	−2.24	(0.12)
CVLT – LD	0.23	(0.16)	0.27	(0.18)	−0.31	(0.27)	−0.90	(0.16)	−2.41	(0.16)	−1.04	(0.15)	−1.15	(0.14)	−2.61	(0.14)
ROCF – D	0.07	(0.35)	0.10	(0.22)	−1.33	(0.24)	−0.93	(0.18)	−2.31	(0.14)	−1.39	(0.20)	−1.73	(0.17)	−2.76	(0.09)
ROCF – R	−0.20	(0.31)	−0.25	(0.23)	−0.90	(0.24)	−0.91	(0.17)	−1.89	(0.23)	−1.27	(0.27)	−1.48	(0.18)	−2.29	(0.17)
BVMT – IR	0.28	(0.14)	−0.91	(0.15)	−0.85	(0.18)	−2.04	(0.11)	−2.50	(0.11)	−2.30	(0.13)	−2.01	(0.16)	−2.85	(0.04)
BVMT – DR	0.80	(0.12)	−0.35	(0.17)	−0.36	(0.19)	−1.91	(0.16)	−2.71	(0.10)	−2.06	(0.13)	−1.78	(0.18)	−2.81	(0.06)
Processing Speed
Coding	0.54	(0.12)	0.13	(0.17)	−0.17	(0.11)	0.32	(0.14)	−0.03	(0.12)	−1.03	(0.15)	−0.87	(0.14)	−1.53	(0.37)
GPB (DH)	−0.23	(0.14)	−0.96	(0.17)	−0.62	(0.16)	−0.67	(0.15)	−1.07	(0.13)	−1.64	(0.20)	−1.49	(0.14)	−1.83	(0.26)
GPB (NDH)	−0.24	(0.12)	−0.89	(0.14)	−0.71	(0.14)	−0.69	(0.14)	−0.88	(0.12)	−1.71	(0.17)	−1.64	(0.09)	−1.56	(0.20)
TMT – Part A	0.06	(0.13)	−0.07	(0.22)	−0.62	(0.15)	−0.27	(0.15)	−0.57	(0.16)	−1.80	(0.24)	−1.06	(0.18)	−1.82	(0.37)
Executive Functioning
DS – F	0.30	(0.14)	0.03	(0.27)	−0.61	(0.19)	−0.04	(0.16)	0.10	(0.15)	−0.98	(0.18)	−0.82	(0.13)	−1.07	(0.16)
DS – B	0.39	(0.12)	0.55	(0.27)	−0.55	(0.18)	0.30	(0.16)	0.37	(0.17)	−0.96	(0.12)	−0.56	(0.13)	−1.10	(0.13)
DS – S	0.27	(0.11)	0.13	(0.23)	−0.55	(0.14)	0.08	(0.15)	−0.19	(0.17)	−1.12	(0.19)	−0.71	(0.20)	−1.68	(0.17)
WCST – C	−0.67	(0.19)	−0.30	(0.23)	−0.45	(0.22)	−0.82	(0.21)	−1.68	(0.23)	−1.71	(0.43)	−1.27	(0.25)	−2.31	(0.42)
WCST – E	0.78	(0.23)	−0.16	(0.21)	−0.27	(0.17)	0.55	(0.21)	−0.15	(0.27)	−1.24	(0.30)	−0.74	(0.15)	−1.47	(0.30)
TMT – Part B	−0.04	(0.12)	0.00	(0.15)	−0.77	(0.13)	−0.27	(0.16)	−1.13	(0.22)	−2.51	(0.20)	−1.43	(0.17)	−2.73	(0.31)
Psychiatric Symptomatology
BDI-II	5.34	(0.74)	21.35	(1.82)	9.22	(1.11)	7.35	(0.77)	7.12	(0.84)	7.74	(1.38)	28.16	(1.52)	10.62	(2.63)
BAI	4.11	(0.95)	14.38	(1.59)	5.86	(0.94)	4.93	(0.74)	4.57	(0.76)	4.35	(0.93)	23.09	(1.96)	10.00	(3.38)
AES	29.13	(3.61)	35.57	(2.65)	30.60	(2.22)	27.43	(1.82)	32.26	(1.97)	33.99	(2.89)	48.24	(2.79)	34.22	(9.53)
FSI	6.44	(1.36)	30.53	(3.21)	16.22	(2.80)	7.85	(1.40)	7.30	(1.69)	11.94	(3.43)	41.01	(3.17)	19.13	(5.99)
Frailty Criteria
Exhaustion	5%	(0.03)	63%	(0.11)	33%	(0.10)	0%	(0.01)	26%	(0.06)	19%	(0.10)	79%	(0.08)	19%	(0.10)
Weight Loss	8%	(0.05)	16%	(0.09)	10%	(0.05)	12%	(0.05)	18%	(0.06)	0%	(0.01)	19%	(0.07)	11%	(0.08)
Inactivity	0%	(0.00)	10%	(0.06)	8%	(0.04)	7%	(0.04)	12%	(0.05)	12%	(0.07)	7%	(0.04)	12%	(0.07)
Slowness	0%	(0.00)	15%	(0.07)	5%	(0.05)	11%	(0.05)	29%	(0.07)	5%	(0.06)	43%	(0.08)	37%	(0.11)
Weakness	34%	(0.07)	78%	(0.08)	50%	(0.10)	66%	(0.09)	61%	(0.11)	63%	(0.11)	79%	(0.07)	71%	(0.12)

Bolded values indicate clinical significance.

The inclusion of psychiatric and frailty indicators elucidated two novel profiles, a Psychiatric/Pre-Frail profile, and a Multiple Cognitive Domains/Psychiatric/Frail profile. The Psychiatric/Pre-Frail profile was comprised of those identified as Healthy (55%), At-Risk (23%), and Pre-MCI (22%) in the six-profile solution. The majority of those reclassified as Multiple Cognitive Domains/Psychiatric/Frail were previously labeled Mildly Impaired, Multiple Domains (66%), with a minority labeled previously At-Risk (14%), Mildly Impaired, Amnestic (11%), Major Cognitive Impairment (6%), and Pre-MCI (3%).

The Psychiatric/Pre-Frail profile did not demonstrate impairments on any neuropsychological task but expressed clinically significant depressive symptoms, anxiety, and fatigue. The Multiple Cognitive Domains/Psychiatric/Frail profile demonstrated a cognitive profile similar to the Mildly Impaired, Multiple Domains profile in addition to elevated depressive symptoms, anxiety, fatigue, and apathy.

Auxiliary covariates

Profiles derived from the eight-profile solution were compared across auxiliary covariates including sociodemographic variables, continuous frailty metrics, and health factors (Table 5). First, Mini-Mental Status Exam (MMSE) scores were compared across profiles. The Major Cognitive Impairment profile demonstrated the lowest performance on this cognitive screen, followed by the MCI, Amnestic Type class. Notably, estimated means of the remaining six profiles all fell above clinical thresholds (≥ 26). None of the participants that fell within the Psychiatric/Pre-Frail or MCI profiles were rated as functionally dependent by the referring physician. With respect to age, participants within the Amnestic MCI and Pre-MCI profiles were older than those in the remaining classes, whereas persons within the Multiple Cognitive Domains/Psychiatric/Frail profile were youngest. In all profiles, males and females were well-represented. The Major Cognitive Impairment profile was most heavily female-represented (69%), while the Mildly Impaired, Multiple Domains profile included a greater proportion of males (58%).

Table 5.

Class-wise comparisons on auxiliary variables.

	Healthy(N = 73)	Psych(N = 47)	At-Risk(N = 74)	Pre-MCI(N = 87)	aMCI(N = 82)	MD MCI(N = 44)	MCI+(N = 64)	Major(N = 59)	Significantly different from:
	Healthy(N = 73)	Psych(N = 47)	At-Risk(N = 74)	Pre-MCI(N = 87)	aMCI(N = 82)	MD MCI(N = 44)	MCI+(N = 64)	Major(N = 59)	Healthy	Major
Demographics
MMSE	28.64	28.43	27.86	27.50	24.98	26.11	27.01	20.38
Age	67.65	65.00	62.63	71.97	74.01	67.85	61.90	68.60	2, 3, 4, 6	2, 3, 4
Female	59%	66%	64%	48%	48%	42%	69%	72%		2, 5, 6
Racial Minority	11%	2%	10%	9%	6%	2%	13%	20%	5, 8	2, 5, 8
Hispanic	29%	70%	48%	43%	57%	57%	68%	57%	1, 2, 34, 5, 8	7
Risk Factors
BMI	26.63	30.32	27.69	26.89	26.16	26.39	28.65	27.46	8	8
Systolic BP	130.38	128.47	130.95	135.06	137.35	133.84	135.48	133.87
Diastolic BP	68.10	73.64	79.55	74.84	69.51	77.09	76.32	73.74	3
Comorbidities (#)	1.46	3.23	2.11	1.84	2.30	2.14	2.50	1.64	2, 3, 4, 8	4, 8
Medications (#)	2.66	8.21	4.52	4.80	6.06	6.33	8.55	6.33	1, 2, 34, 5, 6	7
PSQI	4.31	9.12	6.93	6.74	4.49	4.79	10.79	6.96	1, 3, 46, 8	2, 4, 7
Epworth	5.77	9.67	7.03	6.28	5.97	6.16	9.70	5.49	4, 8	4, 8
ACE	0.91	2.12	1.84	0.98	0.36	0.75	2.72	1.37	2, 4	2, 3, 4, 8
Dev. Delays	9%	3%	2%	2%	5%	9%	11%	0%		4
Medical History
Hypertension	27%	39%	36%	42%	38%	29%	39%	32%
Hyperlipidemia	31%	41%	34%	34%	38%	32%	34%	25%
Diabetes	9%	20%	18%	12%	13%	14%	17%	14%
CAD	4%	7%	6%	3%	14%	7%	8%	3%	2
A-Fib	4%	4%	0%	8%	5%	7%	3%	3%
MI	0%	2%	0%	0%	1%	5%	0%	5%
Ischemia	0%	2%	0%	2%	4%	0%	5%	2%
TBI	0%	4%	7%	2%	2%	2%	5%	5%	3
OSA	5%	22%	14%	10%	11%	14%	13%	3%	8	3, 8
Migraine	6%	13%	3%	1%	1%	17%	18%	2%	4	4, 5, 8
COPD	0%	5%	1%	0%	4%	2%	0%	2%
Fibrosis	0%	9%	1%	0%	1%	2%	5%	0%	8	8
Depression	2%	61%	19%	16%	16%	18%	51%	15%	1, 2, 3, 45, 6, 8	4, 7, 8
Anxiety	4%	40%	9%	15%	8%	6%	39%	14%	4, 6, 8	4, 8

Significant differences are based on χ² estimated via the BCH method.

Across all profiles, mean BMI was consistently within overweight ranges (>25). There was a significant difference in BMI across profiles (χ²(7) = 18.27, p = 0.011). Those within the Psychiatric/Frail profile demonstrated the highest BMI, which was significantly greater than all other profiles (p ≤ 0.034) except the Multiple Cognitive Domains/Psychiatric/Frail profile (p = 0.186). Profile-wise systolic blood pressures were within pre-hypertensive ranges while diastolic blood pressures fell within normative ranges. There was a significant difference in diastolic (χ²(7) = 17.33, p = 0.015) but not systolic (χ²(7) = 5.24, p = 0.631) blood pressure. Diastolic blood pressure was highest for those within the At-Risk profile, only significantly greater than the Healthy and Amnestic MCI profiles.

Polypharmacy differed notably across profiles (χ²(7) = 86.72, p < 0.001). Differences were driven by the Healthy profile, which was significantly less medicated than all other classes (p ≤ 0.014). The Multiple Cognitive Domains/Psychiatric/Frail and Psychiatric/Frail profiles demonstrated the highest levels of polypharmacy. A similar pattern emerged with respect to medical comorbidities. Persons within the Psychiatric/Frail and Multiple Cognitive Domains/Psychiatric/Frail demonstrated the greatest medical burden, while those within the Major Cognitive Impairment and Healthy profiles were the lowest.

Auxiliary psychosocial covariates included sleep dysregulation, daytime somnolence, and aversive childhood events (ACEs). Sleep dysregulation was highest in the Multiple Cognitive Domains/Psychiatric/Frail and Psychiatric/Frail profiles, and above clinical thresholds in the Major Cognitive Impairment, At-Risk, and Pre-MCI. Daytime somnolence followed a similar pattern, with persons within the Multiple Cognitive Domains/Psychiatric/Frail and Psychiatric/Frail profiles exhibiting the greatest somnolence. Lastly, ACEs were most frequent in Multiple Cognitive Domains/Psychiatric/Frail and Psychiatric/Frail profiles, and least frequent in the Amnestic MCI profile.

Discussion

The early identification of cognitive impairment is a critical first step towards characterizing the prospective nature of diminishing cognition and developing interventions that will slow or arrest future decline. Yet to date, there is little agreement on how to best characterize presentations and possible subtypes of MCI. The present study aimed to expand models of MCI by replicating data-driven models of cognitive impairment patterns while extending such models to integrate concomitant risk factors known to drive cognition. We believe our sample was an ecologically valid representation of patients routinely seen in neurological practice. This demonstrates an important extension in understanding the variability within neuropsychological impairment and associating such variability with relevant clinical findings.

Our study expanded the literature on cognition and aging by considering two additional dimensions known to predict cognitive decline but that are rarely included in studies examining the phenotypic heterogeneity of MCI: psychiatric symptomatology and physical frailty. In doing so, we identified two subpopulations not discussed in the extent research, subpopulations which would be subsumed by other profiles without the inclusion of frailty and psychiatric symptoms. The first was a subpopulation experiencing primarily physical and psychiatric distress in the absence of cognitive impairment, which would be subsumed by the Healthy profile – 55% of those identified as Healthy based on cognitive data alone were reclassified into the Psychiatric/Frail profile. The second was a subpopulation experiencing simultaneous cognitive, physical, and psychiatric impairments, subsumed by the Mildly Impaired, Multiple Domains profile. These results show that more comprehensive neuropsychological assessment allows for better classification additional subpopulations that share common characteristics, providing a view that can help in diagnosis, evaluating prognosis, effective treatment planning, and study design.

In recent years, data-driven methodologies have illuminated the need for greater attention to the heterogeneity in MCI by revealing numerous single- and multiple-domain subtypes. This body of research has highlighted the importance of considering not only the severity of cognitive impairment, but also the specific neuropsychological domains which have been impacted,^11,13–15 and has validated such subtypes against neurologic correlates and prognosis.^6,19 Commonly observed single-domain subtypes have included Amnestic, Linguistic, Dysnomic, and Dysexecutive presentations, while multi-domain subtypes have included Memory/Language, Memory/Executive, Executive/Visuospatial, and Executive/Processing Speed.^6,15,19 Our findings aligned with this broad categorization by corroborating a single-domain Amnestic MCI profile and a general Multiple Domain MCI profile; however, we did not find support for the more nuanced neuropsychological subtypes previously described. One explanation for this discrepancy is that by increasing the breadth and depth of cognitive testing, we see a degree of overlap between profiles not apparent with a more circumscribed battery. Further, the specific thresholds with which one classifies a domain as impaired has considerable influence on the representation of MCI subtypes.^14,20 We relied on a relatively strict threshold for impairment based upon normative data,¹³ which may influence the type and number of domains reaching significance. Nonetheless, our findings underscore the importance of such research by supporting the presence of MCI subpopulations which meaningfully vary in primary deficits, severity of impairments, and concomitant physical and behavioral health factors.

Existing research in this field has largely addressed cognition in isolation, validating emerging profiles against known markers of neuropathology. A limitation of the extant research in this area is a reliance on brief neuropsychological batteries,¹⁵ compounded with the empirical reality that these exploratory data techniques require large, diverse samples. Many of these previous endeavors have leveraged publicly available databases, which offer large samples but are restricted in the breadth of cognitive data available. Our work contributes to this body of literature by expanding the depth of neuropsychological assessment while increasing the breadth of physical and behavioral indicators considered.

The present study does include several limitations of note. First, participants in this study did not have known classifications based on currently accepted diagnostic schemes prior to their evaluations. Consequently, it was not possible to examine congruence between our classifications and the current diagnostic labels of Mild and Major Neurocognitive Disorder. Additionally, these data were collected over a period of time in which the assessment of cerebrospinal fluid or plasma biomarkers was not routine clinical practice, particularly in those without objective cognitive decline. Consequently, this study was limited in its ability to examine differences of biomarker prevalence across neurocognitive profiles. We hypothesize that heterogeneity in MCI reflects the high prevalence of multiple pathologies (e.g., Alzheimer's disease with cerebrovascular disease).^56,57 Further research is necessary to understand how modern biomarkers of dementia pathologies relate to multidimensional neurocognitive profiles presented here and otherwise.^6,11,19,20

The current findings, when considered alongside extant literature, point to a need for a wider breadth of cognitive assessment and the consideration of additional domains. Ultimately, the utility of these data-driven approaches depends on their ability to predict subpopulation stability, change over time, and long-term outcomes (e.g., conversion to dementia, functional dependence, & mortality). Further work remains necessary to examine whether these subpopulations remain independent over time and to establish which profiles are most indicative of critical clinical outcomes (i.e., conversion & reversion). It is of particular interest to determine 1) if the Psychiatric/Frail profile represents a prodrome to cognitive decline and 2) if the Multiple Cognitive Domains/Psychiatric/Frail profile represents a population further along the dementia continuum compared to those without psychiatric symptoms or frailty.

Supplemental Material

sj-docx-1-alz-10.1177_13872877241290127 - Supplemental material for Exploring heterogeneity in mild cognitive impairment

Supplemental material, sj-docx-1-alz-10.1177_13872877241290127 for Exploring heterogeneity in mild cognitive impairment by Zachary T Goodman, Maria M Llabre, Sonya Kaur, Nikhil Banerjee, Katalina McInerney, Xiaoyan Sun, Anita Seixas Dias Saporta and Bonnie E Levin in Journal of Alzheimer's Disease

Footnotes

Acknowledgments

We thank the McKnight Brain Institute, the McKnight Brain Research Registry, and each of the participants who generously gave their time in support of this research. This study was not preregistered.

ORCID iD

Zachary T Goodman

Author contributions

Zachary T Goodman (Conceptualization; Formal analysis; Methodology; Writing – original draft; Writing – review & editing); Maria M Llabre (Conceptualization; Investigation; Methodology; Supervision; Writing – original draft; Writing – review & editing); Sonya Kaur (Conceptualization; Data curation; Investigation; Writing – review & editing); Nikhil Banerjee (Conceptualization; Data curation; Investigation; Writing – review & editing); Katalina McInerney (Conceptualization; Data curation; Formal analysis; Funding acquisition; Writing – review & editing); Xiaoyan Sun (Data curation; Funding acquisition; Writing – review & editing); Anita Seixas Dias Saporta (Data curation; Project administration; Writing – review & editing); Bonnie E Levin (Conceptualization; Data curation; Funding acquisition; Investigation; Project administration; Resources; Writing – original draft; Writing – review & editing).

Funding

This work was supported by The National Institutes of Health (T32-HL007426 to Z.T.G.).

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

Materials to reproduce analyses are available upon request. Data are not publicly available due to the inclusion of confidential, potentially sensitive information.

Supplemental material

Supplemental material for this article is available online.

References

Roberts

Knopman

. Classification and epidemiology of MCI. Clin Geriatr Med 2013; 29: 753–772.

Lopez

Becker

Chang

Y-F

, et al. Incidence of mild cognitive impairment in the Pittsburgh cardiovascular health study–cognition study. Neurology 2012; 79: 1599–1606.

Thomas

Edmonds

Eppig

, et al. MCI-to-normal reversion using neuropsychological criteria in the Alzheimer’s disease neuroimaging initiative. Alzheimers Dement 2019; 15: 1322–1332.

Yaffe

. Modifiable risk factors and prevention of dementia: what is the latest evidence? JAMA Intern Med 2018; 178: 281–282.

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.

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.

Dunne

Aarsland

Brien

, , et al. Mild cognitive impairment: the Manchester consensus. Age Ageing 2021; 50: 72–80.

Blanken

Dutt

, et al. Disentangling heterogeneity in Alzheimer’s disease: two empirically-derived subtypes. J Alzheimers Dis 2019; 70: 227–239.

Blanken

Jang

, et al. Distilling heterogeneity of mild cognitive impairment in the national Alzheimer coordinating center database using latent profile analysis. JAMA Netw Open 2020; 3: e200413.

10.

Libon

Xie

Eppig

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

11.

Machulda

Lundt

Albertston

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

12.

Edmonds

Delano-Wood

Jak

, et al. “Missed” mild cognitive impairment: high false-negative error rate based on conventional diagnostic criteria. J Alzheimers Dis 2016; 52: 685–691.

13.

Bondi

Smith

. Mild cognitive impairment: a concept and diagnostic entity in need of input from neuropsychology. J Int Neuropsychol Soc 2014; 20: 129–134.

14.

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.

15.

Edmonds

Smirnov

Thomas

, et al. Data-driven vs consensus diagnosis of MCI: enhanced sensitivity for detection of clinical, biomarker, and neuropathologic outcomes. Neurology 2021; 97: e1288–e1299.

16.

Petermann-Rocha

Lyall

Gray

, et al. Associations between physical frailty and dementia incidence: a prospective study from UK Biobank. Lancet Healthy Longev 2020; 1: e58–e68.

17.

Kim

Han

K-D

Baek

, et al. Association between physical activity and conversion from mild cognitive impairment to dementia. Alzheimers Res Ther 2020; 12: 136.

18.

. Depression, anxiety, and apathy in mild cognitive impairment: current perspectives. Front Aging Neurosci 2020; 12: 9.

19.

Delano-Wood

Bondi

Sacco

, et al. Heterogeneity in mild cognitive impairment: differences in neuropsychological profile and associated white matter lesion pathology. J Int Neuropsychol Soc 2009; 15: 906–914.

20.

Jak

Bondi

Delano-Wood

, et al. Quantification of five neuropsychological approaches to defining mild cognitive impairment. Am J Geriatr Psychiatry 2009; 17: 368–375.

21.

Lezak

Howieson

Bigler

, et al. Neuropsychological assessment, 5th ed. New York, NY, US: Oxford University Press, 2012.

22.

Nelson

Willison

. National Adult Reading Test (NART). Windsor, UK: NFER-Nelson, 1991.

23.

Del Ser

González-Montalvo

J-I

Martı´nez-Espinosa

, et al. Estimation of premorbid intelligence in Spanish people with the word accentuation test and its application to the diagnosis of dementia. Brain Cogn 1997; 33: 343–356.

24.

Kaplan

Goodglass

Weintraub

. Boston naming test. 2nd edition. Philadelphia: Lea & Febiger, 1983.

25.

Delis

Kaplan

Kramer

. Delis-Kaplan Executive Function System: Technical Manual. San Antonio, TX: The Psychological Corporation, 2001.

26.

Rey

Benton

. Multilingual Aphasia Examination-Spanish. Manual of Instructions. Iowa City: AJA Associates, 1991.

27.

Wechsler

. Wechsler Adult Intelligence Scale - Fourth Edition. San Antonio, TX: Pearson Assessments, 2008.

28.

Benton

Sivan

Hamsher

, et al. Contributions to neuropsychological assessment: A clinical manual. USA: Oxford University Press, 1994.

29.

Osterrieth

. Le test de copie d’une figure complexe; contribution à l’étude de la perception et de la mémoire. [Test of copying a complex figure; contribution to the study of perception and memory.]. Arch Psychol 1944; 30: 206–356.

30.

Rey

. L’examen psychologique dans les cas d’encéphalopathie traumatique. (Les problems.). [The psychological examination in cases of traumatic encepholopathy. Problems.]. Arch Psychol 1941; 28: 215–285.

31.

Wechsler

. Wechsler Memory Scale - Fourth Edition. San Antonio, TX: Pearson Assessments, 2009.

32.

Delis

Kramer

Kaplan

, et al. California Verbal Learning Test, Second Edition. San Antonio, TX: Pearson Assessments, 2000.

33.

Benedict

RHB

Schretlen

Groninger

, et al. Revision of the brief visuospatial memory test: studies of normal performance, reliability, and validity. Psychol Assess 1996; 8: 145–153.

34.

Matthews

Klove

. Instruction Manual for the Adult Neuropsychology Test Battery. Madison, WI: University of Wisconsin Medical School, 1964.

35.

Milner

. Effects of different brain lesions on card sorting: the role of the frontal lobes. Arch Neurol 1963; 9: 90–100.

36.

Beck

Steer

Brown

. Manual for the Beck Depression Inventory-II. San Antonio, TX: Psychological Corporation, 1996.

37.

Beck

Steer

. Beck Anxiety Inventory Manual. San Antonio, TX: The Psychological Corporation Harcourt Brace & Company, 1993.

38.

Marin

Biedrzycki

Firinciogullari

. Reliability and validity of the apathy evaluation scale. Psychiatry Res 1991; 38: 143–162.

39.

Hann

Jacobsen

Azzarello

, et al. Measurement of fatigue in cancer patients: development and validation of the fatigue symptom inventory. Qual Life Res 1998; 7: 301–310.

40.

von Glischinski

von Brachel

Hirschfeld

. How depressed is “depressed”? A systematic review and diagnostic meta-analysis of optimal cut points for the Beck depression inventory revised (BDI-II). Qual Life Res 2019; 28: 1111–1118.

41.

Fried

Tangen

Walston

, et al. Frailty in older adults: evidence for a phenotype. J Gerontol A Biol Sci Med Sci 2001; 56: M146–M156.

42.

Stewart

Mills

King

, et al. CHAMPS Physical activity questionnaire for older adults: outcomes for interventions. Med Sci Sports Exerc 2001; 33: 1126–1141.

43.

Radloff

. The CES-D scale: a self-report depression scale for research in the general population. Appl Psychol Meas 1977; 1: 385–401.

44.

Peugh

Fan

. Modeling unobserved heterogeneity using latent profile analysis: a Monte Carlo simulation. Struct Equ Model Multidiscip J 2013; 20: 616–639.

45.

Pastor

Gagné

. Mean and covariance structure mixture models. In: Structural equation modeling: a second course, 2nd ed. Charlotte, NC, US: IAP Information Age Publishing, 2013, pp.343–393.

46.

Dempster

Laird

Rubin

. Maximum likelihood from incomplete data via the EM algorithm. J R Stat Soc Ser B Methodol 1977; 39: 1–22.

47.

Nylund

Asparouhov

Muthén

. Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study. Struct Equ Model Multidiscip J 2007; 14: 535–569.

48.

Mendell

Rubin

. Testing the number of components in a normal mixture. Biometrika 2001; 88: 767–778.

49.

Wang

M-C

Deng

, et al. Performance of the entropy as an index of classification accuracy in latent profile analysis: a Monte Carlo simulation study. Acta Psychol Sin 2017; 49: 1473–1482.

50.

Woo

Jebb

Tay

, et al. Putting the “person” in the center: review and synthesis of person-centered approaches and methods in organizational science. Organ Res Methods 2018; 21: 814–845.

51.

Lubke

Neale

. Distinguishing between latent classes and continuous factors: resolution by maximum likelihood? Multivar Behav Res 2006; 41: 499–532.

52.

Spurk

Hirschi

Wang

, et al. Latent profile analysis: a review and “how to” guide of its application within vocational behavior research. J Vocat Behav 2020; 120: 103445.

53.

Jak

Urban

McCauley

, et al. Profile of hippocampal volumes and stroke risk varies by neuropsychological definition of mild cognitive impairment. J Int Neuropsychol Soc 2009; 15: 890–897.

54.

Bakk

Vermunt

. Robustness of stepwise latent class modeling with continuous distal outcomes. Struct Equ Model Multidiscip J 2016; 23: 20–31.

55.

Asparouhov

Muthén

. Auxiliary variables in mixture modeling: three-step approaches using Mplus. Struct Equ Model 2014; 21: 329–341.

56.

Kapasi

DeCarli

Schneider

. Impact of multiple pathologies on the threshold for clinically overt dementia. Acta Neuropathol (Berl) 2017; 134: 171–186.

57.

Kawas

Kim

Sonnen

, et al. Multiple pathologies are common and related to dementia in the oldest-old. Neurology 2015; 85: 535–542.

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

0.03 MB