Measuring the Frailty Phenotype and its Association with Cognition in Mid-Life and Older Age

Abstract

Background:

Frailty is directly linked to physical robustness and cognitive decline in older age. The Fried Frailty phenotype (FP) is a construct composed of five core symptoms that has been studied predominately in older age. There is little research contrasting the psychometric properties of the FP in mid-life versus older age.

Objective:

We compared the psychometric properties of the FP in mid-life and older age and investigated relationships between the FP and cognition.

Methods:

Frailty and neuropsychological assessments were completed on 361 adults, between 45 and 92 years of age, without primary neurological disorders. Confirmatory factor analysis was used to examine FP, indicated by Grip Strength, Gait Speed, Physical Activity, Fatigue, and Weight Loss. Measurement invariance was tested in mid-life (45–64 years) versus older age (≥65 years). Associations were examined between FP and language, executive functions, memory, processing speed, and visuospatial domains as well as a Generalized Cognition factor. Age was tested as a moderator of these associations.

Results:

Weight Loss was a poor indicator of FP. Factor loadings were comparable across age groups; however, Fatigue was disproportionately higher among those in mid-life. FP was negatively associated with all cognitive domains and remained invariant across age groups.

Conclusion:

Results support the construct validity of the FP and document its stable associations with poorer cognition in middle and older life. Future research investigating central features of frailty earlier in life may offer avenues for developing targeted prevention measures and better characterization of individuals with elevated dementia risk.

Keywords

Aging cognition confirmatory factor analysis frailty measurement invariance neuropsychology quantitative methods

INTRODUCTION

Frailty is broadly defined as an age-associated decline in reserve and resistance to stressors, marked by decreased strength, lean body mass, activity level, endurance, and ambulation [1, 2]. Frailty is associated with diverse negative health outcomes including increased risk for institutionalization, multimorbidity [3], and mortality [4, 5], as well as a two-fold increase in long-term healthcare costs [6]. Frailty is also associated with global cognitive decline [7] and age-related brain disorders such as Alzheimer’s disease and vascular dementia [8 –11]. Given evidence that neurodegenerative disease often begins a decade or more before clinical signs emerge, there are growing efforts to identify and employ preventative measures earlier in life [12]. Despite well-supported links between frailty and incidence of dementia, much of the extant frailty research has been completed with older adults [7]. Frailty and its association with cognition in mid-life (i.e., 45 to 64 years of age) has received little attention. Considering the implications for public health and healthcare costs, investigating relevant features of frailty earlier in life may offer an avenue for developing targeted prevention measures and better characterization of individuals with elevated dementia risk.

Measurement of frailty

Multiple approaches exist to assess frailty [13]; however, the Frailty phenotype (FP) [1] is the most widely cited measure to date. The FP includes five clinical criteria: weakness, as measured by grip strength, slowed gait, low physical activity, unintentional weight loss, and self-reported exhaustion/fatigue. Based on this framework, individuals who meet three or more criteria are considered “Frail”, those with one to two criteria are “Prefrail” and those with zero criteria are considered “Robust.” As frailty is directly associated with the aging process, the phenotype was developed in older adults and most investigators have studied the core symptoms in those aged 65 years and above [1 , 15]. In fact, research on the construct validity of the FP has been almost exclusively conducted among populations aged 65 and above [14 –16].

Early evidence of the co-occurrence of the five FP symptoms provides important preliminary support for the FP [14]. More recently, confirmatory factor analysis (CFA) was applied to establish construct validity of the FP among adults aged 65 years and above, demonstrating that all five frailty symptoms load significantly onto a continuous latent factor of Frailty [15]. Of the five indicators, gait speed exhibited the strongest loading, followed by grip strength, exhaustion/fatigue, physical activity, and weight loss [15].

Given increasing recognition that select interventions may slow, reduce, or reverse symptoms, especially among individuals meeting Prefrail criteria, there is growing interest in studying the FP among younger adults [3 , 18]. To date, there is limited research examining the validity of the FP among those younger than 65 years of age; however, preliminary evidence in adults 37 to 73 years of age suggests frailty, regardless of age, is associated with multimorbidity and mortality [3]. Further, Prefrail and Frail classifications were associated with greater mortality rates across all ages, supporting the predictive validity of the FP throughout adulthood. However, it has yet to be determined whether these symptoms are expressions of the same underlying syndrome in mid-life versus older years. One way to address this important question is to directly test the construct validity of the FP in mid-life and to compare its measurement properties to older adults.

Measurement invariance refers to equality across a set of psychometric properties of a construct across groups [19], including race, ethnicity, culture, language, sex, gender, and age. Measurement non-invariance can occur at several levels, each of which has unique consequences on the validity of the underlying measure. Critically, non-invariance can result in measurement-based bias [20, 21], potentially reducing the validity of the measure, particularly when considered across groups.

While latent class analyses have been leveraged to assess the FP as a discrete syndrome, categorizing otherwise continuous phenomena results in a loss of statistical information [22], and is increasingly recognized as a limitation in extant research spanning psychiatry, psychology, and neuroscience [23, 24]. In contrast, CFA allows for validation of a theory-driven Frailty continuum, providing more nuance in this syndrome as well as increased statistical power to evaluate its association with relevant covariates and outcomes. Additionally, latent class analyses begin with the assumption the categorical construct is invariant across categories, whereas CFA allows more flexible and widely implemented techniques for identifying and rectifying non-invariance.

The Frailty Phenotype and cognition

Studies examining the association between frailty and cognitive functioning are predominantly confined to older adults aged 65 years and above, with most relying on screening measures for cognitive impairment, rather than full neuropsychological assessments. While cognitive screening measures are valuable clinical tools, they preclude more nuanced examinations of specific cognitive domains [7, 25]. These brief assessments are valuable but may lack the sensitivity to detect subtle cognitive impairment, particularly among individuals with high premorbid functioning [26]. At present, extant research is conflicted, with some researchers reporting the FP is most strongly associated with executive dysfunction [7] whereas others find the FP is predictive of accelerated decline in memory systems, perceptual speed, and visuospatial skills [27] and less so with deficits in executive functions [25]. Critically, there is little research directly evaluating if links between frailty and cognition are stable across the lifespan. This is highly pertinent given recent evidence of the deleterious effects of frailty on other health outcomes in mid-life [28], indicating possible earlier onset of frailty and an urgent need to examine the effects of frailty earlier in the life course. Consequently, research utilizing comprehensive neuropsychological batteries across a broader age range is necessary to further characterize specific cognitive deficits associated with frailty.

In the current study, we begin to address these gaps in the literature by examining the construct validity of the FP and assessing its relationship to cognition across a broad age range of adults using a comprehensive neuropsychological battery. The first aim tests the psychometric properties of a continuous, latent measure of the FP among mid-life adults (45–64 years) and evaluates whether the FP construct is commensurate in older age (≥65 years). The second aim assesses the relationship between the FP and cognition, and whether frailty-cognition associations are consistent across mid-life and older age, which has yet to be critically examined in extant research.

METHODS

Participants

Participants were recruited from the University of Miami McKnight Brain Research Registry and volunteers from the community. The registry consisted primarily of patients evaluated in the Department of Neurology for subjective cognitive complaints. Eligible participants were 1) 45 years or older, 2) without major neurological disease (e.g., Parkinson’s disease, multiple sclerosis, epilepsy, severe traumatic brain injury), major psychiatric disturbances (e.g., schizophrenia, major depressive disorder), or significant sensory deficit based on a retrospective chart review and/or clinical interview, 3) a native English or Spanish speaker, and 4) willing and capable to complete cognitive testing and frailty assessment. Participants were recruited between March of 2016 and May of 2018.

A total of 361 participants (63.7% Female) were eligible and included in the study (Table 1). On average, participants were approximately 68 years old (range: 45–92 years), and nearly two-thirds (65.4%) were aged 65 years or older. The sample was predominantly Caucasian (88.6%), followed by African American (5.8%), another race (2.8%), and Asian American (2.2%). Hispanic/Latino (49%) and non-Hispanic/Latino individuals (51%) were equally represented, with one-third (33.2%) completing the evaluation in Spanish. With regard to education, 27% completed a graduate degree, 25% graduated college, 25% completed some college, 13% graduated high school, and 10% did not complete high school. The majority of the sample (76.5%) was recruited from the university’s clinical services.

Table 1

Sample characteristics and descriptive statistics (N = 361), with statistical comparisons between Older (n = 236) and Mid-Life (n = 125) adults

	Overall		Older Age		Mid-Life		p
Continuous variables	M	(SD)	M	(SD)	M	(SD)
Age	67.8	(10.6)	74.0	(6.5)	56.1	(5.6)	< 0.001
Education	14.9	(3.3)	14.6	(3.6)	15.4	(2.9)	0.034
Body Mass Index	27.8	(5.6)	27.4	(5.7)	28.6	(5.4)	0.051
BDI-II	11.5	(10.4)	9.7	(9.3)	14.9	(11.5)	< 0.001
Mini Mental Status Exam	26.7	(3.1)	26.2	(3.3)	27.6	(2.2)	< 0.001
Frailty symptoms
Weight Loss (K)	–0.001	(0.1)	–	(0.1)	–0.004	(0.1)	0.502
Fatigue	2.9	(3.8)	3.1	(4.0)	2.6	(3.5)	0.210
Gait Speed (seconds)	5.2	(1.5)	5.3	(1.5)	5.1	(1.6)	0.400
Grip Strength (lbs/in²)	19.8	(7.8)	19.6	(7.9)	20.2	(7.6)	0.470
Physical Activity (median Kcal)	3512	(4028)	3329	(3918)	3846	(4219)	0.265
Categorical variables	N	(%)	n	(%)	n	(%)
Gender (Female)	230	(63.7)	148	(62.7)	82	(65.6)	0.587
Race							0.002
Caucasian	320	(88.6)	218	(92.4)	102	(81.6)
African American	21	(5.8)
Asian American	8	(2.2)
Other/Unknown	10	(2.8)
Ethnicity (Hispanic/Latino)	177	(49.0)	112	(47.5)	65	(52.0)	0.411
Language (English)	241	(66.8)	87	(36.9)	33	(26.4)	0.045
Recruited from clinic	276	(76.5)	163	(69.1)	113	(90.4)	< 0.001
Medical Conditions
Diabetes	78	(21.6)	53	(22.5)	25	(20.0)	0.589
Hypertension	184	(51.0)	134	(56.8)	50	(40.0)	0.002
Hyperlipidemia	161	(44.6)	119	(50.4)	42	(33.6)	0.002
History of stroke	22	(6.1)	12	(5.1)	10	(8.0)	0.271
Kidney disease	20	(5.5)	13	(5.5)	7	(5.6)	0.971
Cancer	51	(14.1)	41	(17.4)	10	(8.0)	0.015
Heart disease	54	(15.0)	44	(18.6)	10	(8.0)	0.007
Arthritis	37	(10.2)	28	(11.9)	9	(7.2)	0.164
Tobacco use							< 0.001
Never	219	(60.7)	133	(56.4)	86	(68.8)
Previous	120	(33.2)	95	(40.3)	25	(20.0)
Current	14	(3.9)	6	(2.5)	8	(6.4)

Independent samples t-tests were used for group comparisons on continuous variables, while χ² tests were used for all other group comparisons; K = (weight in previous year –current weight)/(weight in previous year).

Procedures

Participants completed a comprehensive frailty assessment [1] as well as a comprehensive neuropsychological evaluation, which included a structured clinical interview, cognitive testing, and self-report mood questionnaires. Evaluations were carried out by trained examiners under the supervision of licensed neuropsychologists. At the community recruitment sites, volunteer participants provided written informed consent before completing the same neuropsychological evaluation and frailty assessment. All study procedures were approved by the university’s Institutional Review Board.

Prior to neuropsychological testing, language competency was evaluated by our team of bilingual clinicians to determine English versus Spanish language for the evaluation. Considerations included participant preference, language of education, use and exposure (e.g., first spoken language, language spoken at home, language of media consumption), as well as performance on objective measures of proficiency (e.g., reading tasks) and clinician judgement. Validated Spanish versions of self-report inventories and neuropsychological tests were used when available. For instance, verbal fluency was assessed using letters PTM in Spanish speakers rather than FAS, the standard for English speakers [29].

Measures

Grip Strength

Grip Strength was measured using the Jamar Hydraulic Hand Dynamometer and calculated as the average score across three trials in the dominant hand. Linear models were run with Grip Strength scores regressed on participant body mass index (BMI), separately for men and women. Model residuals were computed, which results in a measure of Grip Strength adjusted for gender and BMI.

Gait Speed

Gait Speed was operationalized as the average time (seconds) to walk a 15-foot straight line at routine pace across three trials. As with Grip Strength, the residuals approach was utilized to control for height and gender.

Physical Activity

Weekly caloric expenditure was assessed using the Community Healthy Activities Model Program for Seniors (CHAMPS) Questionnaire for Older Adults, 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 [30]. The CHAMPS was translated into Spanish and back translated to English by the bilingual research team in order to maximize linguistic equivalence and culturally appropriate verbiage.

Fatigue

Fatigue was assessed with participant self-reports to two questions: frequency in the last week that 1) “I could not get going”, and 2) “I felt that everything I did was an effort”. Following the methods of Wu and colleagues (2018), symptom duration responses were coded as follows: “less than 1 day” = 0; “1 to 2 days” = 1.5; “3 to 4 days” = 3.5; “5 to 7 days” = 6. Fatigue scores represented the summed value of both coded questions. Consistent with the methods of Fried et al. (2001), these questions were selected from the Center for Epidemiological Studies Depression Scale, which has been validated in both English and Spanish [31]. The questions were also reviewed by the bilingual research team for cultural appropriateness.

Weight Loss

Weight Loss was assessed by the K-coefficient, representing the percentage of weight loss from the past year, calculated as (weight one year prior –current weight)/(weight one year prior).

Neuropsychological evaluation

Participants completed a standard battery of clinical neuropsychological tests with established reliability and validity, as described in detail elsewhere [32, 33]. Individual measures were grouped by five cognitive domains. Processing Speed was measured by Trail Making Test Part A (TMT-A) completion time and the Wechsler Adult Intelligence Scale –Fourth Edition (WAIS-IV) Coding subtest score. TMT-A scores were inverted, such that higher scores would reflect better performance, and the direction of TMT-A scores would be consistent with other neuropsychological measures. Executive Functioning was based on Wisconsin Card Sorting Test (WCST) total perseverative errors, Trail Making Test Part B (TMT-B) completion time, and WAIS-IV Digit Span Total scores. Like TMT-A, both WCST perseverative errors and TMT-B scores were reversed (i.e., multiplied by –1) such that higher scores represented better performance. The Memory domain was based on delayed recall scores from the California Verbal Learning Test- Second Edition (CVLT-II) and Brief Visual Memory Test –Revised (BVMT-R). Visuospatial skills were assessed by the WAIS-IV Block Design subtest and Judgment of Line Orientation task. The Language domain comprised the Boston Naming Test scores and Controlled Oral Word Association category fluency (animals) test. Composite scores were constructed based on the domain groupings detailed above. The standardized residuals were calculated via linear models regressing raw scores for each cognitive measure on all covariates noted below. Composite scores were then calculated as the average of all standardized residuals among domain-specific measures.

Covariates

Performance on each neuropsychological measure was adjusted for several covariates through linear regression, and residualized neuropsychological scores were utilized in subsequent analyses. Covariates included age, sex, years of formal education, race, ethnicity, recruitment source, language of cognitive testing, depressive symptoms, BMI, tobacco use, and medical history of diabetes (Type I or II), hypertension, hyperlipidemia, stroke, kidney disease, cancer (other than melanoma), arthritis, and heart disease. Due to a relatively low proportion of racial minorities, race was dichotomized as White or non-White; ethnicity was coded as Hispanic or non-Hispanic. Language of assessment was coded as English (0) or Spanish (1). Current depressive symptoms were assessed via Beck Depression Inventory –Second Edition (BDI-II), a continuous measure of depressive symptom severity. Self-reported tobacco use history was coded as 1 (never), 2 (previous), or 3 (current). Each medical history variable was dichotomized as denied (0) or endorsed (1).

Statistical analyses

Analyses were complete using SPSS version 24.0 (IBM Corp., Armonk, NY, USA) and R version 3.6.1. Each variable was examined for outliers and normality. Confirmatory factor analysis (CFA) was utilized to test the factor structure of the FP using lavaan version 0.6–4 [34]. Data were assumed missing at random and full information robust maximum likelihood estimation was used. In line with best practices [35], decisions of model fit were determined from several criteria, including non-significant chi-square (χ²) significance tests (p > 0.05), comparative fit index (CFI) values greater than 0.95, as well as root mean square error of approximation (RMSEA) and standardized root mean square residual (SRMR) values less than 0.08 [36]. Determining appropriate sample size in CFA is not straightforward as the number of indicators per factor, the strength of factor loadings, and the magnitude of covariances and regression paths affect sample size requirements [37, 38]; however, the present sample exceeded minimum recommendations and fell within acceptable sample size ranges [38, 39].

Measurement invariance

Measurement invariance (MI) indicates that scores from the operationalization of a construct have the same meaning in different groups and is a prerequisite to validly comparing measures and constructs across populations [19, 40]. We assessed MI of the FP between mid-life (45–64 years) and older age (≥65 years) cohorts by constraining model parameters to be equal across groups and evaluating decreases in model fit; meaningful decreases in fit indicate a violation of MI and are determined by a significant χ² difference test, decreases of the CFI by more than 0.01, and decreases in the RMSEA by more than 0.015 [41].

Measurement invariance was tested in sequential order, with each step building on the previous by applying additional parameter constraints at each step [42]. Configural invariance was evaluated first, followed by metric and scalar invariance. Overall, if constraining parameters produced significant misfit, partial MI was determined by freeing some constrained parameters to reduce misfit to a non-significant degree. Factor means were also compared by constraining the intercepts of the FP to zero and comparing model fit to the final scalar model.

Structural equation models

First, a structural equation model (SEM) was specified with the latent FP factor predicting each of the five neurocognitive composite scores independently. Next, group differences were evaluated using tests of structural invariance by constraining each unstandardized regression path. Significant decreases in fit resulting from constrained paths indicate that the relationship between the FP and the respective cognitive domain are moderated by group [35].

A set of exploratory models were tested to examine specificity in associations between components of the FP and overall cognition. A latent construct of Generalized Cognition was indicated by each cognitive composite, and measurement invariance across mid-life and older age groups was explored using the same procedures previously described. Next, each FP indicator was entered as a predictor of the Generalized Cognition latent variable, and structural invariance was tested between age cohorts.

RESULTS

Descriptive statistics

Missing data counts ranged from 2.2% –8.3% for FP measures and 0.5% –1.4% for cognitive composite scores. Hispanic participants reported significantly higher levels of Fatigue (t(329) = 3.26, p = 0.001), suggesting higher frailty, compared to non-Hispanic participants. On the other hand, Hispanic participants also reported greater engagement in Physical Activity (t(329) = 4.99, p < 0.001), and demonstrated stronger Grip Strength (t(329) = 2.14, p = 0.033) and faster Gait Speed t(329) = 2.73, p = 0.006), suggesting lower levels of frailty. There were no significant differences in any FP criteria between White and non-White participants (p≥0.085).

Table 1 depicts descriptive statistics and statistical comparisons between mid-life and older age on frailty, cognitive, demographic, and medical variables. On average, the older age cohort was less educated, more English speaking and White, and included a larger proportion of participants recruited from the community. The older age cohort also scored slightly worse on the MMSE, indicating poorer global cognition, reported less depression symptoms, and had significantly higher rates of hypertension, hyperlipidemia, cancer, heart disease, and tobacco use.

The present samples were additionally characterized as Robust (0 criteria met), Prefrail (1–2 criteria met), and Frail (3 or more criteria met) using cutoff scores defined by Fried and colleagues [1]. The mean number of FP symptoms was not significantly different (p = 0.954) between older (M = 1.30, SD = 1.10) and mid-life adults (M = 1.29, SD = 1.10). Significantly more mid-life participants met criteria for Frail levels of Fatigue, while more older adults met criteria for Frail levels of Grip Strength (Table 2).

Table 2

Number of participants (N = 361) meeting each Frailty phenotype criteria (adapted from Fried et al. [1]) and within each classification, with statistical comparisons between Older (n = 236) and Mid-Life (n = 125) adults

	Overall		Older Age		Mid-Life		p
	N	(%)	n	(%)	n	(%)
Frailty criteria
Weight Loss	47	(13.2)	32	(13.7)	15	(12.3)	0.704
Fatigue	115	(33.5)	58	(26.0)	57	(47.5)	< 0.001
Gait Speed	61	(17.9)	42	(18.8)	19	(16.4)	0.589
Grip Strength	206	(59.5)	145	(63.9)	61	(51.3)	0.023
Physical Activity (Kcal)	26	(7.2)	18	(8.1)	8	(7.1)	0.731
Frailty phenotype							0.821
Robust	84	(23.3)	53	(26.4)	31	(29.5)
Prefrail	170	(47.1)	114	(56.7)	56	(53.3)
Frail	52	(14.4)	34	(16.9)	18	(17.1)

Weight Loss: K = (weight in previous year –current weight)/(weight in previous year). If K≥0.05 (at least 5% of previous year’s body weight), then the Frail weight loss criterion was met. Fatigue: Using the Center for Epidemiological Studies –Depression (CES–D) scale, the following two statements are read. (a) I felt that everything I did was an effort; (b) I could not get going. The question is asked “How often in the last week did you feel this way?” 0 = rarely or none of the time (< 1 day), 1 = some or a little of the time (1–2 days), 2 = a moderate amount of the time (3–4 days), or 3 = most of the time. Subjects answering “2” or “3” to either question were categorized as Frail by the exhaustion criterion. Gait Speed: Walking time is assessed over 15 feet. The average was computed across three trials. Mean walking speed scores (seconds) were compared against standardized cutoff times stratified by sex and height (see Fried et al. [1]). Walking speeds falling below these thresholds indicated the criterion was met. Grip Strength: Assessed with a hand dynamometer bilaterally. Average scores (lbs/in²) across three trials were compared against standardized cutoff times stratified by sex and body mass index (see Fried et al., 2001). Scores falling below these thresholds indicated the criterion was met. Physical Activity: Kcals per week expended are calculated using the CHAMPS standardized algorithm, stratified by gender. Men under 383 Kcals expended per week are classified as Frail; women under 270 Kcals expended per week are classified as Frail.

Frailty: Confirmatory factor analysis and measurement invariance

First, a CFA with full-information robust maximum likelihood estimation was built using the five previously identified indicators of the FP: Gait Speed, Grip Strength, Fatigue, Physical Activity, and Weight Loss. The model fit the data well (χ²(5) = 7.66, p = 0.176, CFI = 0.97, RMSEA = 0.04, SRMR = 0.03) although Weight Loss did not load significantly onto the FP factor (p = 0.663). With Weight Loss removed, the model maintained acceptable fit (χ²(2) = 1.66, p = 0.435, CFI = 1.00, RMSEA = 0.00, SRMR = 0.02), and this model was retained. Gait Speed (λ= 0.56), Grip Strength (λ= 0.62), Fatigue (λ= 0.34), and Physical Activity (λ= 0.33), all loaded significantly onto the FP factor.

Next, the MI of the FP was assessed across mid-life and older age groups (Table 3). The configural model fit the data well, supporting the tenability of the FP in both mid-life and older adults. The metric model also fit the data well, and the χ² difference test was not significant, confirming the FP maintains the same interpretation across groups. However, the scalar model fit significantly worse than the metric model. To explore partial scalar invariance, each intercept was freed one-at-a-time to determine which constrained parameter harmed fit most. Freeing the intercept of Fatigue resulted in a model with acceptable fit, which was not significantly worse fitting than the metric model. These findings indicated that the mid-life cohort reported higher mean fatigue scores at the same level of frailty, compared to the older age participants. Lastly, the mean model fit significantly worse than the partial scalar model, indicating the older age cohort had higher overall levels of frailty than the mid-life cohort.

Table 3

Model fit and measurement invariance (MI) comparisons across mid-life and older age

Model	χ²	df	p	CFI	RMSEA	SRMR	Δχ ²	df	p
Frailty Phenotype
Configural	5.24	4	0.264	0.99	0.04	0.03
Metric	7.66	7	0.147	0.96	0.06	0.05	5.00	3	0.172
Scalar	48.17	10	< 0.001	0.57	0.15	0.09	42.99	3	< 0.001
Partial Scalar^a	12.37	9	0.193	0.96	0.05	0.05	1.29	2	0.526
Mean	25.42	10	0.005	0.83	0.09	0.08	11.65	1	< 0.001
Generalized Cognition
Configural	17.75	10	0.059	0.99	0.07	0.03
Metric	19.16	14	0.159	0.99	0.05	0.03	1.08	4	0.897
Scalar	21.62	18	0.249	0.99	0.01	0.04	2.48	4	0.648
Mean	21.92	19	0.288	0.99	0.03	0.04	0.27	1	0.604

^aIntercept of Fatigue was freed. Fit is compared to the invariant metric model.

Frailty and neuropsychological composite scores: Structural equation model

After establishing metric and partial scalar invariance, the FP was used to predict five neurocognitive composite scores: Language, Visuospatial, Processing Speed, Executive Functioning, and Memory. First, each composite was predicted from the FP latent variable via SEM. Next, each model was examined across age groups for invariance in structural paths. Fit statistics for each model with path coefficients, are reported in Table 4.

Table 4

Model fit and structural regression paths of models predicting indicators of Generalized Cognition from the Frailty phenotype as well as indicators of the Frailty phenotype predicting Generalized Cognition

		Model Fit Statistics						Structural Paths
	Model	χ²	df	p^a	CFI	RMSEA	SRMR	b	SE	β	p^b
Predicting Cognitive Composites from the Frailty Phenotype^c
Language	9.77	5	0.082	0.95	0.05	0.03	–0.13	0.04	–0.21	0.002
Visuospatial Skills	1.94	5	0.857	1.00	0.00	0.01	–0.17	0.06	–0.24	0.004
Delayed Memory	2.50	5	0.777	1.00	0.00	0.02	–0.11	0.05	–0.15	0.047
Processing Speed	2.44	5	0.786	1.00	0.00	0.02	–0.17	0.06	–0.25	0.003
Executive Functioning	5.47	5	0.361	0.99	0.02	0.03	–0.16	0.04	–0.20	0.004
Predicting Generalized Cognition from Indicators of the Frailty Phenotype
Fatigue	14.94	9	0.093	0.99	0.05	0.03	–0.02	0.01	–0.16	0.009
Gait Speed	10.71	9	0.296	0.99	0.02	0.02	–0.06	0.02	–0.17	0.005
Physical Activity	17.75	9	0.038	0.98	0.05	0.03	–0.09	0.10	–0.05	0.388
Grip Strength	13.90	9	0.126	0.99	0.04	0.02	–0.02	0.01	–0.19	0.002
Weight Loss	11.43	9	0.247	0.99	0.03	0.02	–0.12	0.44	–0.02	0.781

CFI will equal 1 and RMSEA will equal 0 when χ² values are smaller than model degrees of freedom. ^arefers to the p-value of the χ² test of model fit. ^brefers to the p-value of the estimate. ^cthe Frailty phenotype latent variable does not include Weight Loss as an indicator.

All models fit the data well and revealed significant paths from the FP to each cognitive composite. The FP was associated with poorer abilities in areas of Language, Visuospatial, Memory, Processing Speed, and Executive Functioning (Table 4). There were no significant differences in fit when structural paths were constrained to be equal across age groups (ps≥0.319).

Given the consistent relationship between the FP and neuropsychological composite scores, we sought to examine whether the FP would be similarly associated with a latent factor comprised of all five cognitive domains. First, we tested a latent factor Generalized Cognition, as indicated by each cognitive composite score, via CFA. The single-factor model fit the data well; Language (λ= 0.60), Visuospatial (λ= 0.69), Memory (λ= 0.59), Processing Speed (λ= 0.74), and Executive Functioning (λ= 0.76) all loaded significantly onto the factor.

Second, invariance of the Generalized Cognition latent variable was tested across age groups. The configural, metric, scalar models all fit the data well, supporting invariance of the Generalized Cognition factor between age groups (Table 3).

After establishing MI of Generalized Cognition and partial MI of the FP, structural invariance of the association between the FP and Generalized Cognition was tested across age groups. An overall SEM with the FP predicting Generalized Cognition fit the data well (χ²(26) = 32.44, p = 0.179, CFI = 0.99, RMSEA = 0.03, SRMR = 0.03) and revealed a negative relationship between the FP and Generalized Cognition (b = –0.10, SE = 0.03, β= –0.27, p = 0.002; Fig. 1). A multigroup SEM analysis revealed good model fit when the path between the FP and Generalized Cognition was constrained to be equal in mid-life and older age groups and was not significantly worse in fit than the unconstrained model (Δ χ²(1) < 0.01, p = 0.977). Taken together, the negative association between the FP and Generalized Cognition was found to be consistent across mid-life and older age samples.

Fig. 1

Structural equation model (SEM), with standardized factor loadings and regression paths, demonstrating the negative relationship between the Frailty phenotype and Generalized Cognition latent variables. Exh., Exhaustion/Fatigue; Gait, Gait Speed; CH, Community Healthy Activities Model Program for Seniors (CHAMPS) Questionnaire; Grip, Grip Strength; Lang, Language Composite Score; VS, Visuospatial Composite Score; Mem, Memory Composite Score; PS, Processing Speed Composite Score; EF, Executive Functioning Composite Score.

To examine specificity in FP indicators, we conducted exploratory post hoc analyses evaluating the relationship between individual FP indicators and the Generalized Cognition factor (Table 4). Results indicated that Gait Speed and Grip Strength had the strongest effect, followed by Fatigue. The independent effects of self-reported Physical Activity and Weight Loss were not significant.

DISCUSSION

The primary aim of this study was to ascertain whether the Frailty phenotype (FP) [1], as developed and studied in older age, has comparable psychometric properties to those in mid-life, a time when it is assumed people are more vigorous and robust. To accomplish this, our study assessed the validity of the FP as a continuous construct and explored whether its measurement properties differ between those in mid-life (45–64 years) and older age (65+). Factor analyses demonstrated all Frailty indicators except Weight Loss loaded significantly onto the latent factor, and the strength of loadings did not differ across age groups. These findings support the construct validity of at least four of the five FP criteria and provide preliminary evidence that the FP reflects the same underlying syndrome across age groups.

Fatigue was more frequently endorsed among those in mid-life relative to later life, an unexpected finding given significantly greater levels of frailty among those in the older cohort. Critically, fatigue is uniquely associated with poorer cognitive functioning [43] and is the FP criterion most predictive of increased healthcare costs [6]. The current results are consistent with mounting research showing that fatigue levels are higher among relatively younger adults [3]. While the reason for this remains unclear, one possibility is that it has been shown that depressive symptoms decline from mid-life to older age [44]. Depression, which often includes symptoms of lethargy, sleepiness, or insomnia, may be a strong covariate of fatigue as conceptualized in the FP. In support of this hypothesis, we observed that depression scores were significantly higher in mid-life compared to older participants. Alternatively, higher reports of fatigue in mid-life may be attributed to greater environmental demands found among those below the traditional retirement age. These might include child-rearing, employment, and other social responsibilities. Similarly, it may be that older adults may adaptively limit their engagement with strenuous activities in order to conserve energy and reduce fatigue. Consistent with this notion, our data suggested that some adults in mid-life engaged in more physical exertion compared to older adults, albeit these differences did not reach statistical significance. Survival biases and cohort effects should also be considered as explanatory factors. Nonetheless, these findings indicate that fatigue is an especially salient feature of frailty as expressed in mid-life. From a measurement perspective, these findings suggest that age differences in fatigue may lead to bias in FP total scores, if not accounted for. Thus, future research should consider adjusting for age and potentially depressive symptoms when measuring fatigue as an indicator of the FP.

We observed some variability in the strengths of individual FP loadings. The two objective physical measures (Grip Strength and Gait Speed) represented the strongest loadings, followed by self-reported Fatigue and Physical Activity, whereas the loading for Weight Loss was weak and non-significant. Similarly, previous population-based research [15] found that weight loss was the weakest, albeit statistically significant, contributor to frailty. These inconsistent findings may be reflective of difficulty measuring weight change in a unidimensional frailty measurement model. Moreover, while weight loss is traditionally identified as a key symptom of frailty, there is evidence to suggest that weight gain is also associated with other FP criteria. Obese older adults have been found to exhibit commensurate reductions in walking speed, strength, and muscle mass as non-obese Frail individuals [45]. Some individuals experience simultaneous muscle loss with weight gain, referred to as sarcopenic obesity [46], a condition that has garnered attention as an important health concern [47]. It is also noteworthy that the measure of weight loss in the present study only assessed weight lost over the past year; however, weight fluctuations may not progress uniformly or linearly, and K may fail to capture long term weight changes. In sum, the contribution of weight change to frailty may be moderated by other health factors, and not simply representable by declines in total weight.

The second aim of the present study was to elucidate the association between frailty and specific domains of cognitive functioning, and to test whether these relationships differ between those in mid-life and older age. Our findings showed that the FP was negatively associated with all cognitive domains while controlling for medical comorbidities and psychosocial factors, suggesting a diffuse, rather than domain-specific, association with cognition. A generalized effect of the FP on cognitive functioning was further supported by follow-up analyses wherein a latent factor of Generalized Cognition was modeled. Lastly, analyses revealed that of the five FP criteria, Gait Speed and Grip Strength had the strongest association with Generalized Cognition, a finding that is consistent with previous research[7, 27].

The present study adds to the growing body of literature showing that frailty is negatively associated with broad cognitive abilities [27, 48], supporting the notion that frailty is a negative prognostic indicator for cognitive impairment and incidence of dementia [8–11 , 27]. Proposed mechanisms underlying the relationship between frailty and cognition have included Alzheimer’s disease neuropathology, increased cardiovascular risk, hormonal dysregulation, nutritional deficiencies, chronic inflammation, and mood disorders [7]. Given the multifaceted processes linking frailty and cognitive impairment, it is perhaps not surprising to find frailty exerts an expansive impact on cognition.

We found that the negative associations between frailty and cognitive functioning were similarly represented among participants both in mid-life and older age. Prior to this study, there has been limited research directly investigating possible age differences in this relationship.

However, one study found that the effects of frailty on memory and attention were dissociable between relatively younger and older adults, with negative effects of frailty on executive functioning restricted to those aged 75 years and above, while associations with memory were isolated to those aged 50 to 64 years [48]. Given that a certain degree of frailty may be considered a normative part of the aging process, particularly in later years, an important question for future longitudinal research is whether the prognosis is worse for those exhibiting earlier symptoms of frailty in mid-life.

One strength of this study was the measurement of frailty as a continuous latent factor. The benefits of this approach include the ability to capture greater differentiation along the frailty continuum as well as increased statistical power to elucidate possible associations with cognition. The utility of a comprehensive neuropsychological battery also allowed for refined assessment of the relationship between frailty and specific cognitive domains. While the study sample was very diverse regarding its ethnic representation (i.e., nearly 50% Hispanic), it was predominantly comprised of a clinic sample presenting with cognitive complaints but without a history of major neurological disease. Similarly, individuals referred for neuropsychological evaluation in mid-life are not likely representative of all mid-life adults. Therefore, the ability to generalize these findings to a broader community may be limited. Another limitation is that there may be additional cultural/ethnic factors impacting the expression and measurement of frailty. While we controlled for race, ethnicity, and language of testing to account for cultural and linguistic factors that might potentially influence our findings, we recognize these proxies are not exhaustive. For example, we were unable to control for acculturation status among participants born outside the U.S. Additionally, we acknowledge that subjective symptoms, such as fatigue, may differ in terms of how they are perceived, manifested, and reported between Hispanics and non-Hispanics [49]. Finally, we could not account for potential medication effects on frailty due to inadequate medication documentation for many of the participants. While we attempted to mitigate the likelihood of this limitation confounding the current results by excluding significant conditions at greatest risk for cognitive complications due to polypharmacy, this remains an important consideration of age-related cognitive and physical frailty research.

Conclusion

This study demonstrates that frailty in mid-life and older age is an expression of a similar underlying construct. This was evidenced by both invariance of the FP across age cohorts as well as commensurate negative effects of frailty on cognitive abilities across these age groups. Of note, Fatigue is a particularly important symptom to consider in mid-life, whereas weight-loss does not appear to be a defining feature of the FP in either mid-life or older age. While additional research is necessary to further elucidate possible prognostic differences in frailty across the age spectrum, these findings provide novel support for the importance of frailty assessment beginning in mid-life, and possibly earlier. Early detection and treatment of frailty may have a profound impact on reducing the personal and societal burdens associated with medical comorbidities and cognitive impairment.

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. Materials to reproduce analyses will be made available upon request. Data are not publicly available due to the inclusion of confidential and potential sensitive information. This study was not preregistered.

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

Authors’ disclosures available online ().

References

Fried

, Tangen

, Walston

, Newman

, Hirsch

, Gottdiener

, Seeman

, Tracy

, Kop

, Burke

, McBurnie

, Cardiovascular Health Study Collaborative Research Group (2001) Frailty in older adults: Evidence for a phenotype. J Gerontol A Biol Sci Med Sci 56, M146–M156.

Dent

, Kowal

, Hoogendijk

(2016) Frailty measurement in research and clinical practice: A review. Eur J Intern Med 31, 3–10.

Hanlon

, Nicholl

, Jani

, Lee

, McQueenie

, Mair

(2018) Frailty and pre-frailty in middle-aged and older adults and its association with multimorbidity and mortality: A prospective analysis of 493 737 UK Biobank participants. Lancet Public Health 3, e323–e332.

Gutman

, Stark

, Donald

, Beattie

(2001) Contribution of self-reported health ratings to predicting frailty, institutionalization, and death over a 5-year period. Int Psychogeriatr 13(Supp 1), 223–231.

Hewitt

, Carter

, Vilches-Moraga

, Quinn

, Braude

, Verduri

, Pearce

, Stechman

, Short

, Price

, Collins

, Bruce

, Einarsson

, Rickard

, Mitchell

, Holloway

, Hesford

, Barlow-Pay

, Clini

, Myint

, Moug

, McCarthy

, Davey

, Jones

, Lunstone

, Cavenagh

, Silver

, Telford

, Simmons

, Mutasem

TEJ

, Singh

, Paxton

, Harris

, Galbraith

, Bhatti

, Edwards

, Duffy

, Bisset

, Alexander

, Garcia

, Sangani

, Kneen

, Lee

, McGovern

, Guaraldi

(2020) The effect of frailty on survival in patients with COVID-19 (COPE): A multicentre, European, observational cohort study. Lancet Public Health 5, e444–e451.

Hajek

, Bock

J-O

, Saum

K-U

, Matschinger

, Brenner

, Holleczek

, Haefeli

, Heider

, König

H-H

(2018) Frailty and healthcare costs-longitudinal results of a prospective cohort study. Age Ageing 47, 233–241.

Robertson

, Savva

, Kenny

(2013) Frailty and cognitive impairment–a review of the evidence and causal mechanisms. Ageing Res Rev 12, 840–851.

Buchman

, Boyle

, Wilson

, Tang

, Bennett

(2007) Frailty is associated with incident Alzheimer’s disease and cognitive decline in the elderly. Psychosom Med 69, 483–489.

Gray

, Anderson

, Hubbard

, LaCroix

, Crane

, McCormick

, Bowen

, McCurry

, Larson

(2013) Frailty and incident dementia. J Gerontol A Biol Sci Med Sci 68, 1083–1090.

10.

Kojima

, Taniguchi

, Iliffe

, Walters

(2016) Frailty as a predictor of Alzheimer disease, vascular dementia, and all dementia among community-dwelling older people: A systematic review and meta-analysis. J Am Med Dir Assoc 17, 881–888.

11.

Solfrizzi

, Scafato

, Frisardi

, Seripa

, Logroscino

, Maggi

, Imbimbo

, Galluzzo

, Baldereschi

, Gandin

, Di Carlo

, Inzitari

, Crepaldi

, Pilotto

, Panza

, Italian Longitudinal Study on Aging Working Group (2013) Frailty syndrome and the risk of vascular dementia: The Italian Longitudinal Study on Aging. Alzheimers Dement 9, 113–122.

12.

Ngandu

, Lehtisalo

, Solomon

, Levälahti

, Ahtiluoto

, Antikainen

, Bäckman

, Hänninen

, Jula

, Laatikainen

, Lindström

, Mangialasche

, Paajanen

, Pajala

, Peltonen

, Rauramaa

, Stigsdotter-Neely

, Strandberg

, Tuomilehto

, Soininen

, Kivipelto

(2015) A 2 year multidomain intervention of diet, exercise, cognitive training, and vascular risk monitoring versus control to prevent cognitive decline in at-risk elderly people (FINGER): A randomised controlled trial. Lancet 385, 2255–2263.

13.

Faller

, Pereira

, do

, de Souza

, Nampo

, Orlandi

, de

, Matumoto

(2019) Instruments for the detection of frailty syndrome in older adults: A systematic review. PLoS One 14, e0216166.

14.

Bandeen-Roche

, Xue

Q-L

, Ferrucci

, Walston

, Guralnik

, Chaves

, Zeger

, Fried

(2006) Phenotype of frailty: Characterization in the women’s health and aging studies. J Gerontol A Biol Sci Med Sci 61, 262–266.

15.

, Geldhof

, Xue

Q-L

, Kim

, Newman

, Odden

(2018) Development, construct validity, and predictive validity of a continuous frailty scale: Results from 2 large US cohorts. Am J Epidemiol 187, 1752–1762.

16.

Mezuk

, Lohman

, Dumenci

, Lapane

(2013) Are depression and frailty overlapping syndromes in mid- and late-life? A latent variable analysis. Am J Geriatr Psychiatry 21, 560–569.

17.

Rockwood

, Song

, Mitnitski

(2011) Changes in relative fitness and frailty across the adult lifespan: Evidence from the Canadian National Population Health Survey. CMAJ 183, E487–E494.

18.

Mitnitski

, Rockwood

(2016) The rate of aging: The rate of deficit accumulation does not change over the adult life span. Biogerontology 17, 199–204.

19.

Meredith

(1993) Measurement invariance, factor analysis and factorial invariance. Psychometrika 58, 525–543.

20.

Meredith

, Teresi

(2006) An essay on measurement and factorial invariance. Med Care 44, S69.

21.

Borsboom

(2006) When does measurement invariance matter? Med Care 44, S176–S181.

22.

MacCallum

, Zhang

, Preacher

, Rucker

(2002) On the practice of dichotomization of quantitative variables. Psychol Methods 7, 19–40.

23.

Cuthbert

, Insel

(2013) Toward the future of psychiatric diagnosis: The seven pillars of RDoC.. BMC Med 11, 126.

24.

Henderson

, van Lierop

, McLean

, Uszler

, Thornton

, Siow

Y-H

, Pavel

, Cardaci

, Cohen

(2020) Functional neuroimaging in psychiatry—aiding in diagnosis and guiding treatment. What the American Psychiatric Association does not know.. Front Psychiatry 11, 276.

25.

Brigola

, Rossetti

, dos Santos

, Neri

, Zazzetta

, Inouye

, Pavarini

SCI

(2015) Relationship between cognition and frailty in elderly: A systematic review. Dement Neuropsychol 9, 110–119.

26.

Chan

, Khan

, Oliver

, Gill

, Werring

, Cipolotti

(2014) Underestimation of cognitive impairments by the Montreal Cognitive Assessment (MoCA) in an acute stroke unit population. J Neurol Sci 343, 176–179.

27.

Boyle

, Buchman

, Wilson

, Leurgans

, Bennett

(2010) Physical frailty is associated with incident mild cognitive impairment in community-based older persons. J Am Geriatr Soc 58, 248–255.

28.

, Ploner

, Karlsson

, Liu

, Magnusson

PKE

, Pedersen

, Hägg

, Jylhävä

(2019) The frailty index is a predictor of cause-specific mortality independent of familial effects from midlife onwards: A large cohort study. BMC Medicine 17, 94.

29.

Rey

, Feldman

, Rivas-Vazquez

, Levin

, Benton

(1999) Neuropsychological test development and normative data on Hispanics. Arch Clin Neuropsychol 14, 593–601.

30.

Stewart

, Mills

, King

, Haskell

, Gillis

, Ritter

(2001) CHAMPS physical activity questionnaire for older adults: Outcomes for interventions. Med Sci Sports Exerc 33, 1126–1141.

31.

Ruiz-Grosso

, Loret de Mola

, Vega-Dienstmaier

, Arevalo

, Chavez

, Vilela

, Lazo

, Huapaya

(2012) Validation of the Spanish Center for Epidemiological Studies Depression and Zung Self-Rating Depression Scales: A comparative validation study. PLoS One 7, e45413.

32.

Lezak

, Howieson

, Loring

, Hannay

, Fischer

(2004) Neuropsychological Assessment. 4th ed. Oxford University Press.

33.

Zucchella

, Federico

, Martini

, Tinazzi

, Bartolo

, Tamburin

(2018) Neuropsychological testing. Pract Neurol 18, 227–237.

34.

Rosseel

(2012) lavaan: An r package for structural equation modeling. J Stat Softw 48, 1–36.

35.

Kline

(2015) Principles and Practice of Structural Equation Modeling, Fourth Edition, Guilford Publications.

36.

, Bentler

(1999) Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Struct Equ Modeling 6, 1–55.

37.

Maccallum

, Widaman

, Zhang

, Hong

, Maccallum

, Zhang

, Hong

, Sample size in factor analysis. Psychol Methods 4, 84–99.

38.

Wolf

, Harrington

, Clark

, Miller

(2013) Sample size requirements for structural equation models: An evaluation of power, bias, and solution propriety. Educ Psychol Meas 73, 913–934.

39.

Bentler

, Chou

C-P

(1987) Practical issues in structural modeling. Sociol Methods Res 16, 78–117.

40.

Meade

, Lautenschlager

(2004) A comparison of item response theory and confirmatory factor analytic methodologies for establishing measurement equivalence/invariance. Organ Res Methods 7, 361–388.

41.

Chen

(2007) Sensitivity of goodness of fit indexes to lack of measurement invariance. Struct Equ Modeling 14, 464–504.

42.

Thompson

, Green

(2013) Evaluating between-group differences in latent variable means. In, Information Age Publishing. Structural equation modeling: A second course, pp. 163–218.

43.

Banerjee

, Slugh

, Kaur

, Sun-Suslow

, McInerney

, Sun

, Levin

(2020) Neuropsychological correlates of subjective fatigue in non-demented older adults and the moderating effect of physical activity. Neuropsychol Dev Cogn B Aging Neuropsychol Cogn 27, 254–269.

44.

Hasin

, Goodwin

, Stinson

, Grant

(2005) Epidemiology of major depressive disorder: Results from the National Epidemiologic Survey on Alcoholism and Related Conditions. Arch Gen Psychiatry 62, 1097–1106.

45.

Villareal

, Banks

, Siener

, Sinacore

, Klein

(2004) Physical frailty and body composition in obese elderly men and women. Obes Res 12, 913–920.

46.

Stenholm

, Harris

, Rantanen

, Visser

, Kritchevsky

, Ferrucci

(2008) Sarcopenic obesity: Definition, cause and consequences. Curr Opin Clin Nutr Metab Care 11, 693–700.

47.

Zamboni

, Mazzali

, Fantin

, Rossi

, Di Francesco

(2008) Sarcopenic obesity: A new category of obesity in the elderly. Nutr Metab Cardiovasc Dis 18, 388–395.

48.

Robertson

, Savva

, Coen

, Kenny

R-A

(2014) Cognitive function in the prefrailty and frailty syndrome. J Am Geriatr Soc 62, 2118–2124.

49.

Cordero

, Loredo

, Murray

, Dimsdale

(2012) Characterizing fatigue: The effects of ethnicity and acculturation. J Appl Biobehav Res 17, 59–78.