Physical Frailty Phenotype Criteria and Their Synergistic Association on Cognitive Functioning

Abstract

Background

Frailty (physical frailty phenotype [PFP]) and its criteria (slow gait, weakness, weight loss, low activity, and exhaustion) are each associated with cognitive dysfunction. The extent to which the PFP is associated with cognition beyond that expected from its component parts remains uncertain.

Method

We used the National Health and Aging Trends Study to quantify associations between PFP criteria and cognitive performance (level/change) using adjusted mixed effects models. We tested whether frailty was associated with excess cognitive vulnerability (synergistic/excess effects, Cohen’s d) beyond criteria contributions by assessing interactions between each criterion and frailty.

Results

Among 7439 community-dwelling older adults (mean age = 75.2 years) followed for a mean of 3.2 years (SE = 0.03), 14.1% were frail. The PFP and PFP criteria were all associated with lower baseline cognitive performance, among which slow gait (−0.31 SD, SE = 0.02) and frailty (−0.23 SD, SE = 0.02) were strongest. Only slow gait (−0.03 SD/year, SE = 0.01), frailty (−0.02 SD/year, SE = 0.01), weight loss (−0.02 SD/year, SE = 0.01), and weakness (−0.02 SD/year, SE = 0.01) were associated with cognitive decline. Frailty was associated with cognitive performance above and beyond each criterion (excess effects ranging from −0.07 SD [SE = −0.05] for slow gait to −0.23 SD [SE = 0.03] for weakness); the same was not true for cognitive decline. Slow gait was the only criterion associated with cognitive change among both frail and nonfrail participants (frail: Cohen’s d/year = −0.03, SE = 0.01; nonfrail: Cohen’s d/year = −0.02, SE = 0.01).

Conclusions

PFP is an important frailty measure that is cross-sectionally associated with lower cognitive performance, but not with subsequent cognitive decline, above and beyond its criteria contributions. Further research into the construct of frailty as a “syndrome” correlated with cognition and other adverse outcomes is needed.

Physical frailty is a syndrome (1–3), distinct but related to comorbidity and disability (4), occurring in approximately 10%–15% of community-living older adults (3,5). It is described as a deterioration in physiologic reserve as a result of an underlying multisystem dysfunction that is manifested by the body’s inability to recover efficiently to acute and everyday stressors (1–3). While a consensus on a clear definition of frailty is lacking, with at least 67 identified frailty instruments (6), including the commonly investigated Deficit Accumulation Index (7,8), the physical frailty phenotype (PFP) (1) has been shown to be the most widely used measure of frailty (6), especially for its utility in etiologic research given its biologic underpinnings (3,6). The PFP is defined by the presence of any 3 of 5 criteria: slow gait speed, weakness, unintentional weight loss, low activity, and exhaustion. These indicators were identified to reflect the Cycle of Frailty based on geriatric observation and theory of multisystem dysregulation in relation to many adverse outcomes including falls, disability, hospitalizations, and mortality (1), and have since been validated as syndromic (2). However, the indicators are assumed to be equally important in measuring the phenotype (2), despite evidence demonstrating their role as independent predictors of adverse outcomes with varying strengths (9–14). It remains unclear how much value the construct of the phenotype adds for predicting adverse outcomes above and beyond the independent contributions of the individual criteria that define its measure (2).

This study aims to critically evaluate this property of the PFP within the context of cognitive decline. Using cognition as a case study to test this property is ideal for several reasons. Firstly, there is a growing interest in the relationship between physical frailty and cognition (15–21); a greater understanding of the frailty–cognition link presents optimistic opportunities to develop effective intervention and prevention strategies for both frailty and cognitive impairment (15,22,23). Secondly, studies have found that physical frailty is associated with poorer cognitive performance and steeper cognitive decline in older adults (17,22,24–26), and that the association may be stronger for executive function compared to other domains such as memory or orientation (27). Thirdly, and most importantly, each of the 5 PFP criteria was also identified as independently associated with cognitive performance and/or decline (1,22,28–37). As such, it remains unclear whether the PFP components interact synergistically in relation to cognition, or whether each of the 5 PFP criteria is independently driving the association with cognition. This begs the question: Is frailty—a unitary construct of multisystem dysfunction—associated with cognitive performance and decline? Or are there distinct biological processes—as represented by the 5 separate criteria—that are driving the relationship with cognition?

To critically evaluate this question, we used data from the National Health and Aging Trends Study (NHATS), a large, nationally representative sample of U.S. Medicare beneficiaries aged 65 years and older followed prospectively. Our goal was to assess if the PFP as a whole is associated with levels and rates of change in cognitive performance to a degree exceeding the contributions of its independent components. First, we assessed and compared predictive value of each individual PFP criterion on cognitive level and change. Additionally, we tested whether the presence of the PFP is associated with excess cognitive vulnerability above and beyond the individual effects of its 5 criteria. A deeper examination of the synergistic properties of the PFP in correlation with cognition will inform our theoretical understanding of the construct of frailty for purposes of studying etiologies underlying cognitive dysfunction.

Method

Study Design

We leveraged NHATS, a nationally representative, prospective cohort study of U.S. Medicare beneficiaries aged 65 years and older. All participants were followed annually for a maximum of 5 years (2011–2016), and data were collected via 2-hour in-person interviews.

Demographic factors such as age, sex, race/ethnicity, education, and annual income were considered, in addition to count of comorbidities (score ranging from 0 to 7) from self-reported medical conditions; specifically, participants were asked whether a doctor had ever told them they had: arthritis, diabetes, heart disease, high blood pressure, lung disease, osteoporosis, and stroke.

Frailty

In this study, baseline frailty is our exposure of interest, and was measured using the PFP, which was developed in the Cardiovascular Health Study (CHS) (1), and cross-validated in the Women’s Health and Aging Studies (WHAS) (2). The PFP criteria were previously operationalized in NHATS (5,15) using validated interview and performance measures of functioning and guidance from previously published studies (1,2). The 5 Fried criteria were defined in NHATS as follows: (i) exhaustion was based on self-reported as recently having low energy or being easily exhausted enough to limit activity; (ii) low physical activity was based on self-reported never having walked for exercise nor engaged in vigorous activity in the past month; (iii) unintentional weight loss was based on self-reported having lost 10 or more points within the prior year or having a body mass index (BMI) ≤18.5 kg/m²; (iv) slow gait was defined as at or below the 20th percentile of weighted population distribution within 4 sex-by-height categories using the first of 2 usual-pace walking trials on and unintentional weight loss or shrinking; and (v) weakness was defined as at or below the 20th percentile within 8 sex-by-BMI categories, using maximum dominant hand grip strength over 2 trials (5).

The PFP criteria were scored as 0 or 1 representing the absence or presence of the component. Consistent with prior studies and recommended practice, participants were scored “0” for each respective criterion if they were not tested because of safety concerns, if they were ineligible due to recent surgery or pain, or if they were unable to complete a task (5). Criteria were then summed to create a total PFP score ranging from 0 to 5, and participants were defined as frail if they had a score of 3 or more.

Cognitive Function

Four domain-specific cognitive performance measures for 3 domains (memory, executive function, and orientation) were assessed in NHATS during the in-person interviews. A word-list memory test (38) required participants to recall 10 words immediately after the list of words was presented (Immediate Word Recall), and again after about a 5-minute delay (Delayed Word Recall). The number of words recalled at a given time point was used as either an immediate or delayed word-recall score, ranging from 0 to 10, where higher scores represent more words recalled. A Clock Drawing Test (CDT) for executive function was also administered to participants. Each participant was given a 2-minute time limit to complete a picture of a clock face telling the time “10 after 11.” Clocks were rated according to standard criteria, where higher scores represent more complete and accurate drawings (39). Orientation to time and history was measured by asking participants for the current day, month, year, and day of the week, as well as to name the current President and Vice President.

For our analyses, we standardized scores of all 4 component cognitive tests using their baseline means and standard deviations, and then averaged them into a global cognitive composite score, as described in prior studies (40). Both level and change of global cognitive function were considered in this study; results for domain-specific cognitive outcomes are available in the Supplementary Materials (Supplementary Table 1 and Supplementary Figures 1–4).

Descriptive Statistics

Descriptive statistics were computed for baseline demographic and health characteristics by frailty status. We present raw numbers and weighted percentages for categorical variables, and weighted means and standard errors for continuous variables. We tested differences in participant characteristics by frailty status using t tests from logistic regressions incorporating survey weights from the sample design to compare means for all continuous variables and frequency distributions for all categorical variables by frailty status. Weighted means and standard errors were further generated for baseline cognitive levels by presence and absence of each PFP criterion.

Comparing Associations and Predictive Performance: PFP versus PFP Criteria

A series of adjusted multiple linear regression models with fixed and random effects for people and time were used to model repeated measures of cognitive performance as a function of covariates including continuous age, sex (female/male), race (White/Black/Hispanic/other), education (less than or equal to eighth grade/ninth to twelfth grade, no diploma/HS diploma, or equivalent/some college but no degree/associates or bachelor’s/graduate degree), income (quartiles), and number of comorbidities (categorical: 0, 1, 2, 3, or 4+). Number of comorbidities was used in the models (instead of the presence or absence of specific comorbidities) in order to avoid controlling for potential mediators of the relationships between the PFP/PFP criteria and cognitive performance and decline. To better control for age over the 65–106 year age-span of participants in this study, we used splines with knots at 75, 85, and 95 years (41). In addition to comparing strength of associations between the PFP/PFP criteria and baseline cognitive performance/subsequent cognitive change, we assessed the predictive value of the PFP and the PFP criteria in 6 separate models by comparing Akaike information criteria (AICs) and Bayesian information criteria (BICs), where lower AICs and BICs indicate greater predictive value.

Excess Effect of Frailty on Cognition: Examining PFP Criteria in Separate Models

Next, we determined the associations between each specific PFP criterion and cognitive performance among frail versus nonfrail participants in separate models, controlling for covariates, by adding interaction terms with the PFP, with no PFP main effect. $\begin{array}{ll}{\displaystyle E(\mathrm{C}\mathrm{o}\mathrm{g}{\mathrm{n}}_{ij}|X_{ij}(b_{0i},b_{1i})=({\mathrm{\beta }}_0+b_{0i})+{\mathrm{\beta }}_1\left(\mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{o}{\mathrm{n}}_k\right)}& \\ +{\mathrm{\beta }}_2(\mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{o}{\mathrm{n}}_k\ast \mathrm{P}\mathrm{F}\mathrm{P})+({\mathrm{\beta }}_3+b_{1i})\left(\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}\right)\\ +{\mathrm{\beta }}_4(\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}\ast \mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{o}{\mathrm{n}}_k)+{\mathrm{\beta }}_5(\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}\ast \mathrm{P}\mathrm{F}\mathrm{P}\ast \mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{o}{\mathrm{n}}_k)\\ +\sum {\mathrm{\beta }}^{11}m=6\left(\mathrm{C}\mathrm{o}\mathrm{v}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{e}{\mathrm{s}}_m\right)+{\epsilon }_{ij}\end{array}$

No PFP main effect was included to avoid erroneous extrapolation due to an implicit assumption that the association of frailty with cognitive level or change is the same within every pattern of criterion responses. This is particularly important in subsequent analyses where we include multiple PFP criteria and the PFP in the same model. For example, if a frailty main effect is added, given that being positive for three or more criteria defines someone as frail, and two or fewer defines someone as nonfrail, the model would inaccurately allow for no association that can be empirically observed. For nonfrail participants, β ₁ and β ₄ quantify the cognitive level and cognitive change, respectively, for each specific PFP criterion. For frail participants, (β ₁ + β ₂) and (β ₄ + β ₅) quantify the cognitive level and cognitive change, respectively, for each specific PFP criterion. Thus, β ₂ represents the excess effect on cognitive performance by those who are positive for the specific frailty criterion and who also have frailty, accounting for covariates. Analogously, β ₅ represents the excess effect on cognitive change by those who are positive for the specific frailty criterion and who also have frailty, accounting for covariates.

Excess Effect of Frailty on Cognition: Examining PFP Criteria in a Single Model

We subsequently determined the independent associations between each criterion and cognitive performance among frail and nonfrail participants, controlling for all other PFP criteria in addition to covariates in a single model:

$${\displaystyle \begin{array}{l}{\displaystyle \begin{array}{l}{\displaystyle E(\mathrm{C}\mathrm{o}\mathrm{g}{\mathrm{n}}_{ij}|X_{ij}(b_{0i},b_{1i})=({\mathrm{\beta }}_0+b_{0i})+{\sum }^{5}k=1[{\mathrm{\beta }}_k\left(\mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{o}{\mathrm{n}}_k\right)}\\ +{\mathrm{\beta }}_{k+5}(\mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{o}{\mathrm{n}}_k\ast \mathrm{P}\mathrm{F}\mathrm{P})+{\displaystyle ({\mathrm{\beta }}_{11}+b_{1i})\left(\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}\right)}\\ +{\mathrm{\beta }}_{k+11}(\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}\ast \mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}{\mathrm{a}}_k)+{\mathrm{\beta }}_{k+16}(\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}\ast \mathrm{P}\mathrm{F}\mathrm{P}\ast \mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}{\mathrm{a}}_k)]\\ +\sum {\mathrm{\beta }}^{27}m=22\left(\mathrm{C}\mathrm{o}\mathrm{v}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{e}{\mathrm{s}}_m\right)+{\epsilon }_{ij}\end{array}}\end{array}}$$

For nonfrail participants, β _k and β _k+11 quantify the cognitive level and cognitive change, respectively, for each of the 5 PFP criteria. For frail participants (β _k + β _k+5) and (β _k+11 + β _k+16) quantify the cognitive level and cognitive change, respectively, for each of the 5 PFP criteria. β _k+5 represents the excess effect on cognitive performance by those who are positive for the specific PFP criterion and who also have frailty, accounting for all other PFP criteria and covariates. β _k+16 represents the excess effect on cognitive change by those who are positive for the specific PFP criterion and who also have frailty, accounting for all other PFP criteria and covariates.

Sensitivity Analyses

Given that reliability of self-reported data may be questionable if cognitive impairment is detected, we conducted a sensitivity analysis to assess whether results remained robust among a subgroup of participants who were not cognitively impaired, which has been previously operationalized in NHATS (15). Additionally, given that smoking remains an important risk factor across all ages (42), coupled with the fact that the potential relationship between frailty and cognitive performance may be vascular in nature (15), we conducted a sensitivity analysis additionally adjusting for smoking status (ever vs never smoker).

Results

Study Population

Of the 7439 community-dwelling older adult NHATS participants followed for a weighted mean of 3.2 years (SE = 0.03), the weighted mean age was 75 years (SE = 0.10), 56.4% were women, and 81.5% self-reported as White. At baseline, 14.1% (n = 1313) had frailty (Table 1).

Table 1.

Baseline Demographic and Health Characteristics by Frailty Status Among Community-Dwelling Older Adults in the National Health and Aging Trends Study (NHATS) (n = 7439)

		Nonfrail (n = 6126)	Frail (n = 1313)
Characteristic	Overall (n = 7439)	Population Size: 29 665 848	Population Size: 4 872 200
Age (years)	75.23 (0.10)	74.62 (0.10)	78.94 (0.24)
Female (%)	4320 (56.44)	3479 (55.25)	841 (63.72)
Race (%)
White	5129 (81.53)	4346 (82.72)	783 (74.27)
Black	1641 (8.20)	1263 (7.59)	378 (11.93)
Hispanic	440 (6.68)	322 (6.00)	118 (10.81)
Other	229 (3.59)	195 (3.69)	34 (2.99)
Income (%)
1st quartile	2302 (25.37)	1683 (22.68)	619 (41.74)
2nd quartile	2020 (25.67)	1618 (24.60)	402 (32.22)
3rd quartile	1778 (26.04)	1580 (27.45)	198 (17.44)
4th quartile	1339 (22.92)	1245 (25.27)	94 (8.60)
Education (%)
8th grade or less	953 (10.14)	643 (8.50)	310 (20.13)
9th–12th Grade (no diploma)	1042 (11.36)	790 (10.33)	252 (17.64)
HS diploma or equivalent	2041 (27.58)	1694 (27.41)	347 (28.59)
Some college but no degree	1483 (21.54)	1285 (22.18)	198 (17.60)
Associates or bachelor’s degree	1182 (17.94)	1034 (18.89)	148 (12.18)
Graduate degree	714 (11.44)	665 (12.68)	49 (3.86)
Number of comorbidities (%)
0 comorbidity	1256 (18.37)	1188 (20.56)	68 (5.04)
1 comorbidity	2184 (30.19)	1952 (32.47)	232 (16.27)
2 comorbidities	2048 (26.65)	1681 (26.48)	367(27.68)
3 comorbidities	1163 (14.86)	843(13.30)	320 (24.35)
4+ comorbidities	788 (9.93)	462(7.18)	326 (26.66)
Global Cognition Composite Score	0.20 (0.01)	0.26 (0.01)	−0.26 (0.03)
Clock Drawing Test Score	0.16 (0.02)	0.25 (0.02)	−0.39 (0.04)
Immediate Word Recall	0.18 (0.21)	0.25 (0.02)	−0.33 (0.04)
Delayed Word Recall	0.20 (0.02)	0.25 (0.02)	−0.22 (0.04)
Orientation to Date and Time	0.15 (0.02)	0.24 (0.02)	−0.41 (0.04)
Frailty criteria
Weight loss	1313 (15.00)	658 (9.28)	655 (49.55)
Exhaustion	2371 (29.80)	1243 (20.69)	1068 (85.15)
Low physical activity	1621 (30.47)	1496 (22.33)	1058 (79.94)
Slow gait	1793 (19.91)	806 (10.58)	987 (75.00)
Weakness	1593 (19.80)	790 (12.32)	803 (68.73)

Note: Sampling weights of sample design were imposed to generate representative baseline prevalence estimates. Raw numbers and weighted percentages (%) for categorical characteristics, as well as weighted means and standard deviations for continuous characteristics, are presented. Comorbidities were self-reported and included history of cancer, hip fracture, heart disease, high blood pressure, arthritis, osteoporosis, diabetes, lung disease, or stroke. Standardized scores are presented for each of the cognitive tasks: clock drawing, immediate word recall, delayed word recall, and orientation to date and time. The global cognition composite score was created by standardizing each of the 4 cognitive tests to a mean of 0 and a standard deviation of 1 based on the baseline visit, and taking the average of those standardized scores.

Baseline PFP Criteria Prevalence Estimates

Among the 7439 community-dwelling older adult NHATS participants, low physical activity (30.5%) and exhaustion (29.8%) were the most prevalent PFP criteria, followed by slow gait (19.9%), weakness (19.8%), and unintentional weight loss (15.0%) (Table 1). Among nonfrail participants, low physical activity (22.3%) and exhaustion (20.7%) were the most prevalent, affecting nearly a quarter of participants, followed by weakness (12.3%), slow gait (10.6%), and unintentional weight loss (9.3%). As expected, PFP criteria were more prevalent among frail participants, each affecting a majority of frail community-dwelling older adults (exhaustion: 85.2%; low physical activity: 79.94%; slow gait speed: 75.0%; weakness: 68.7%); unintentional weight loss was the one exception, which had the lowest prevalence, yet was still affecting approximately half of frail participants (49.6%) (Table 1).

Baseline Global Cognitive Performance by Frailty and PFP Criteria

Among community-living older adults, average unadjusted global cognitive scores at baseline were lower and below the population mean for frail participants (Cohen’s d = −0.262 SD, SE = 0.01) compared to nonfrail participants (Cohen’s d = 0.260 SD, SE = 0.03). Across all 5 PFP criteria, global and domain-specific cognitive performance was lower for those who were positive for the criteria compared to those who were negative for the criteria. Participants who tested positive for unintentional weight loss, slow gait, and weakness had cognitive scores consistently below the mean across all global and domain-specific cognitive outcomes, while those who tested negative for the criteria had cognitive scores consistently above the mean (results not shown).

Associations and Predictive Performance of PFP versus PFP Criteria on Global Cognition

After adjusting for demographic and health characteristics, participants who tested positive for the PFP criteria or for frailty had substantially lower cognitive levels at baseline and exhibited greater declines in global cognitive function than those who tested negative (Table 2). Slow gait speed (Cohen’s d = −0.306 SD, SE = 0.02), frailty (Cohen’s d = −0.226 SD, SE = 0.02), and weakness (Cohen’s d = −0.142 SD, SE = 0.02) were most strongly associated with lower cognitive performance, followed by low physical activity (Cohen’s d = −0.092 SD, SE = 0.02) and exhaustion (Cohen’s d = −0.065 SD, SE = 0.02), respectively. Additionally, slow gait speed (Cohen’s d = −0.026 SD/year, SE = 0.01), frailty (Cohen’s d = −0.019 SD/year, SE = 0.01), unintentional weight loss (Cohen’s d = −0.017 SD/year, SE = 0.01), and weakness (−0.016 SE/year, SE = 0.01) were most strongly associated with decline in cognitive function, respectively. Exhaustion (Cohen’s d = −0.003 SD/year, SE = 0.004) and low physical activity (Cohen’s d = −0.007 SD/year, SE = 0.004) were not associated with cognitive change.

Table 2.

Adjusted Associations Between Frailty-Related Variables (PFP and PFP criteria) and Global Cognitive Performance (Cohen’s d, SD) and Change (Cohen’s d, SD/year) Among Community-Dwelling Older Adults in the National Health and Aging Trends Study (NHATS), 2011–2016 (n = 7439)

	Cognitive Level	Cognitive Change
	Cohen’s d (SE)	Cohen’s d, SD/year (SE)	AIC	BIC
Frailty (PFP)	−0.226 (0.020)**	−0.019 (0.006)**	35493.0	35706.2
Weight loss	−0.128 (0.019)**	−0.017 (0.005)**	35168.3	35381.2
Exhaustion	−0.065 (0.016)**	−0.003 (0.004)	35595.7	35808.8
Low physical activity	−0.092 (0.015)**	−0.007 (0.004)	35600.3	35813.5
Slow gait	−0.306 (0.018)**	−0.026 (0.005)**	32483.8	32694.9
Weakness	−0.142 (0.018)**	−0.016 (0.005)**	33746.7	33958.7

Notes: AIC = Akaike information criteria; BIC = Bayesian information criteria; PFP = physical frailty phenotype. Frailty was operationalized using the PFP. Results from a series of multiple linear regression models with random slopes (time) and intercepts (person) testing the associations between frailty-related variables and cognitive level and change in separate models. All models were controlled for age, sex, race, education, income, and number of comorbidities.

**Significance at a cutoff level of p = .05.

Notably, slow gait speed explained more of the variance in global cognitive performance (AIC = 32483.8; BIC = 32694.9), followed by weakness (AIC = 33746.7; BIC = 33958.7), unintentional weight loss (AIC = 35168.3; BIC = 35381.2), frailty (AIC = 35493.0; BIC = 35706.2), exhaustion (AIC = 35595.7; BIC = 35808.8), and low physical activity (AIC = 35600.3; BIC = 35813.5), respectively (Table 2).

Excess Effect of Frailty on Cognition: Results for PFP Criteria in Separate Models

The PFP was associated with lower global cognitive performance at baseline above and beyond 4 of the 5 PFP criteria, including unintentional weight loss (excess effect Cohen’s d = −0.079 SD, SE = 0.04), exhaustion (excess effect Cohen’s d = −0.220, SE = 0.03), low physical activity (excess effect Cohen’s d = −0.188, SE = 0.03), and weakness (excess effect Cohen’s d = −0.234, SE = 0.03) (Table 3). Specifically, frail community-dwelling older adults who were also positive for the specific PFP criterion had substantially lower cognitive scores at baseline compared to nonfrail older adults who were positive for the respective PFP criterion (unintentional weight loss: −0.231 vs −0.052; exhaustion: −0.202 vs 0.018; low physical activity: −0.217 vs −0.029; weakness: −0.276 vs −0.042). The sole exception was with slow gait speed; the PFP was only borderline associated with lower cognitive performance above and beyond slow gait speed (excess effect Cohen’s d = −0.065, SE = 0.03), such that both frail (Cohen’s d = −0.340, SE = 0.03) and nonfrail (Cohen’s d = −0.275, SE = 0.03) community-dwelling older adults who were also positive for slow gait speed exhibited substantially lower cognitive performance.

Table 3.

Global Cognitive Level at Baseline (Cohen’s d, SD) and Cognitive Change (Cohen’s d, SD/year) by Frailty Status (PFP) and Its Criteria Among Community-Dwelling Older Adults in the National Health and Aging Trends Study (2011–2016, n = 7439) in Separate Models

	Global Cognitive Level at Baseline
	Nonfrail β 1	Frail β 1 + β 2	Excess Effect β 2
Criterion	Cohen’s d (SE)	Cohen’s d (SE)	Cohen’s d (SE)
Weight loss	−0.052 (0.024)**	−0.231 (0.029)**	−0.179 (0.036)**
Exhaustion	0.018 (0.018)	−0.202 (0.024)**	−0.220 (0.026)**
Low physical activity	−0.029 (0.017)*	−0.217 (0.024)**	−0.188 (0.026)**
Slow gait	−0.275 (0.026)**	−0.340 (0.026)**	−0.065 (0.034)*
Weakness	−0.042 (0.021)**	−0.276 (0.026)**	−0.234 (0.031)**
	Global Cognitive Change
	Nonfrail β 4	Frail β 4 + β 5	Excess Effect β 5
	Cohen’s d/year (SE)	Cohen’s d/year (SE)	Cohen’s d/year (SE)
Weight loss	−0.019 (0.006)**	−0.015 (0.009)*	0.004 (0.011)
Exhaustion	0.0003 (0.005)	−0.012 (0.007)*	−0.012 (0.008)
Low physical activity	−0.002 (0.005)	−0.021 (0.007)**	−0.019 (0.008)**
Slow gait	−0.024 (0.007)**	−0.030 (0.008)**	−0.006 (0.010)
Weakness	−0.018 (0.006)**	−0.013 (0.008)	0.005 (0.010)

Notes: PFP = physical frailty phenotype. Frailty was operationalized using the PFP. Results from a series of multiple linear regression models with random slopes (time) and intercepts (person) testing the associations between the individual frailty criteria and cognitive level and change among frail versus nonfrail participants. All models were controlled for age, sex, race, education, income, and number of comorbidities. Excess effect refers to the association between frailty and cognitive performance/decline above and beyond each of the respective criteria (ie, for those who are positive for frailty and positive for the criterion of interest).

$\begin{array}{l}E({\text{Cogn}}_{ij}|X_{ij}\left(b_{0i},b_{1,}\right)=\left({\beta }_0+b_{0i}\right)+{\beta }_1\left({\text{Criterion}}_k\right)+{\beta }_2\left({\text{Criterion}}_k*\text{PFP}\right)+\left({\beta }_3+b_{1i}\right)\left(\text{Time}\right)+{\beta }_4\left(\text{Time}*{\text{Criterion}}_k\right)+{\beta }_5\left(\text{Time}*\text{PFP}*{\text{Criterion}}_k\right)\\ {\displaystyle \sum {\beta }^{11}m=6\left({\text{Covariates}}_m\right)+{\epsilon }_{ij}}\end{array}$

*Significance at a cutoff level of p = .10. **Significance at a cutoff level of p = .05.

In few cases was the PFP associated with cognitive change above and beyond each of its criteria (Table 3). The PFP was only associated with global cognitive decline above and beyond low physical activity (Cohen’s d = −0.019 SD/year, SE = 0.01), such that frail community-dwelling older adults who had low physical activity experienced greater declines in cognitive function (Cohen’s d = −0.021 SD/year, SE = 0.01) compared to those who were nonfrail and had low physical activity (Cohen’s d = −0.002 SD/year, SE = 0.01). Notably, slow gait was the sole criterion that was strongly associated with global cognitive change among both frail (Cohen’s d = −0.030 SD/year, SE = 0.01) and nonfrail participants (Cohen’s d = −0.024 SD/year, SE = 0.01).

Excess Effect of Frailty on Cognition: Adjusting for All PFP Criteria in a Single Model

After additional adjustment for all other criteria in the same model, frailty was associated with lower baseline cognitive performance above and beyond weakness (Table 4). Specifically, those who were frail and weak had substantially lower cognitive scores at baseline (Cohen’s d = −0.176 SD, SE = 0.04) compared to those who were nonfrail but weak (Cohen’s d = −0.069 SD, SE = 0.02), and this difference was considerable (excess effect Cohen’s d = −0.106 SD, SE = 0.04). Frailty was not associated with lower baseline global cognitive performance above and beyond unintentional weight loss (Cohen’s d = −0.004 SD, SE = 0.04), exhaustion (Cohen’s d = 0.021 SD, SE = 0.04), low physical activity (Cohen’s d = 0.028 SD, SE = 0.04), and slow gait speed (Cohen’s d = −0.106 SD, SE = 0.01).

Table 4.

Cognitive Level at Baseline (Cohen’s d, SD) and Cognitive Change (Cohen’s d, SD/year) by Frailty Status (PFP) and Its Criteria Among Community-Dwelling Older Adults in the National Health and Aging Trends Study (2011–2016, n = 7439) in the Same Model

	Cognitive Level at Baseline
Criterion	Nonfrail β k Cohen’s d (SE)	Frail β k + β k+5 Cohen’s d (SE)	Excess Effect β k+5 Cohen’s d (SE)
Weight loss	−0.090 (0.026)**	−0.094 (0.036)**	−0.004 (0.044)
Exhaustion	−0.009 (0.018)	0.012 (0.041)	0.021 (0.044)
Low physical activity	−0.044 (0.018)**	−0.016 (0.040)	0.028 (0.043)
Slow gait	−0.276 (0.028)**	−0.232 (0.037)**	0.045 (0.045)
Weakness	−0.069 (0.023)**	−0.176 (0.037)**	−0.106 (0.042)**
	Cognitive Change
Criterion	Nonfrail β k+11 Cohen’s d/year (SE)	Frail β k+11 + β k+16 Cohen’s d/year (SE)	Excess Effect β k+16 Cohen’s d/year (SE)
Weight loss	−0.022 (0.007)**	−0.007 (0.012)	0.015 (0.013)
Exhaustion	−0.001 (0.005)	0.007 (0.013)	0.008 (0.014)
Low physical activity	−0.001 (0.005)	−0.021 (0.013)	−0.019 (0.014)
Slow gait	−0.029 (0.007)**	−0.030 (0.012)**	−0.001 (0.014)
Weakness	−0.019 (0.007)**	0.012 (0.012)	0.031 (0.013)**

Notes: PFP = physical frailty phenotype. Frailty was operationalized using the PFP. Results from a series of multiple linear regression models with random slopes (time) and intercepts (person) testing the associations between the individual frailty criteria and cognitive level and change among frail versus nonfrail participants. All models were controlled for age, sex, race, education, income, number of comorbidities, and each of the other PFP criteria. Excess effect refers to the association between frailty and cognitive performance/decline above and beyond each of the respective criteria (ie, for those who are positive for frailty and positive for the criterion of interest).

$E(\mathrm{C}\mathrm{o}\mathrm{g}{{\mathrm{n}}_{ij}}_{}|{X_{ij}}_{}(b_{0i},b_{1i})=({\mathrm{\beta }}_0+b_{0i})+{\sum }^5{{}_k}_{=1}[{\mathrm{\beta }}_k\left(\mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{o}{\mathrm{n}}_k\right)+$ $\begin{array}{l}{{\mathrm{\beta }}_k}_{+5}(\mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{o}{\mathrm{n}}_k\ast \mathrm{P}\mathrm{F}\mathrm{P})+({\mathrm{\beta }}_{11}+{\mathrm{b}}_{1i})(\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e})+{{\mathrm{\beta }}_k}_{+11}(\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}\ast \mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}{\mathrm{a}}_k)+{{\mathrm{\beta }}_k}_{+16}(\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}\ast \mathrm{P}\mathrm{F}\mathrm{P}\ast \end{array}$ $\begin{array}{l}\mathrm{C}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}{\mathrm{a}}_k)]+\sum {\mathrm{\beta }}^{27}{{}_m}_{=22}(\mathrm{C}\mathrm{o}\mathrm{v}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{e}{\mathrm{s}}_m)+{\epsilon }_{ij}\end{array}$ **Significance at a cutoff level of p = .05.

Additionally, after adjustment for all PFP criteria, frailty was also associated with global cognitive change above and beyond the contributions of its weakness (Table 4). Specifically, those who were frail and weak did not exhibit declines in cognitive performance (Cohen’s d = 0.012 SD/year, SE = 0.01) compared to those who were nonfrail but weak (Cohen’s d = −0.019 SD/year, SE = 0.01), and this difference was significant (excess effect Cohen’s d = −0.031 SD/year, SE = 0.01). Frailty was not associated with cognitive decline above and beyond unintentional weight loss (Cohen’s d = 0.015 SD, SE = 0.01), exhaustion (Cohen’s d = 0.008 SD, SE = 0.01), low physical activity (Cohen’s d = −0.019 SD, SE = 0.01), and slow gait speed (Cohen’s d = −0.001 SD, SE = 0.01).

Sensitivity Analyses

Results remained robust across both sensitivity analyses. Specifically, after conducting analyses only among participants without cognitive impairment (n = 4344, 58.4%), gait speed remained more strongly associated with lower cognitive performance (Cohen’s d = −0.111 SD) and cognitive decline (Cohen’s d = −0.044 SD), and had the lowest AIC and BIC compared to the PFP and the other 4 PFP criteria, indicating greater predictive value (AIC = 17805.4, BIC = 18006.0) (Supplementary Table 2). We also found that excess effects were strengthened, such that the PFP remained strongly associated with lower cognitive performance above and beyond all 5 component parts (Supplementary Table 3). After additionally adjusting for smoking status as a potential confounder, we found similar results. Gait speed was still the component most strongly associated with lower cognitive performance (Cohen’s d = −0.287, p < .001) and cognitive decline (Cohen’s d = −0.016, p = .03), and had the lowest AIC and BIC compared to the PFP and the other 4 PFP criteria, indicating greater predictive value (AIC = 16111.7, BIC = 16312.5) (Supplementary Table 4). Importantly, the PFP remained strongly associated with lower cognitive performance above and beyond all 5 components (Supplementary Table 5).

Discussion

In this study, we found that the association between frailty, as measured by the PFP, and cognitive performance (Cohen’s d = −0.226 SD, SE = 0.02) and change (Cohen’s d = −0.019, SE = 0.01) was comparable to that between gait speed and cognitive performance (Cohen’s d = −0.306 SD, SE = 0.02) and change (Cohen’s d = −0.016, SE = 0.01), but that gait speed explained more of the variance in cognitive function (AIC = 32483.8, BIC = 32694.9) than frailty (AIC = 35493.0, BIC = 35706.2) and all other PFP criteria. After further examination into the synergistic properties of the PFP, we found that frail community-dwelling older adults who were positive for a PFP criterion had substantially lower cognitive performance scores at baseline than nonfrail older adults who were positive for the same PFP criterion. Specifically, we found that the PFP was associated with cognitive level above and beyond the contributions of 4 of its 5 criteria, including unintentional weight loss (excess effect Cohen’s d = −0.179 SD, SE = 0.04), exhaustion (excess effect Cohen’s d = −0.220 SD, SE = 0.03), low physical activity (excess effect Cohen’s d = −0.188 SD, SE = 0.03), and weakness (excess effect Cohen’s d = −0.234 SD, SE = 0.03). Though frailty was only borderline associated with lower cognitive levels above and beyond slow gait speed (excess effect Cohen’s d = −0.065 SD, SE = 0.03), after conducting the sensitivity analyses by additionally controlling for smoking (excess effect Cohen’s d = −0.096 SD, SE = 0.05) and by conducting the analyses only among those who were not cognitively impaired (excess effect Cohen’s d = −0.082 SD, SE = 0.03), this association was strengthened and became significant. In few cases was frailty associated with cognitive change above and beyond its individual criteria. Specifically, frailty was only associated with cognitive decline above and beyond physical activity (Cohen’s d = −0.019 SD/year, SE = 0.008). Notably, gait speed was the only criterion that proved an important predictor of longitudinal change among both frail (Cohen’s d = −0.340 SD/year, SE = 0.03) and nonfrail participants (Cohen’s d = −0.275 SD/year, SE = 0.03), with little difference between both groups (excess effect = −0.065 SD/year, SE = 0.03).

Our findings corroborate existing studies that have found that the phenotype and each of its 5 PFP criteria are individually related to lower cognitive performance among older adults (1,21,22,28–37). Interestingly, in this study, only frailty and the 3 objectively/partially objectively measured criteria (unintentional weight loss, weakness, and slow gait speed) were associated with cognitive change, despite the substantial evidence that has demonstrated that low physical activity and exhaustion may also be associated with cognitive change (30,35,36,43). It is possible that the self-reported nature of physical activity and exhaustion may not be as reliable, particularly among those with cognitive dysfunction, as evidenced in the sensitivity analyses conducted among participants who were not cognitively impaired, whereby associations were strengthened.

With a lack of a gold standard or direct measure of frailty, investigating frailty as a predictor of adverse outcomes with multifactorial etiologies, like cognitive dysfunction, is especially challenging. Prior to this study, there lacked a critical evaluation of the PFP as a measure of frailty associated with adverse outcomes above and beyond its component parts (2). While this study only evaluated this assumption of the PFP within the context of cognitive dysfunction, it revealed the capabilities of the measure for capturing a vulnerability distinct from the separate indicators it relies on. As evidenced from this study, despite the independent associations of its component parts with cognitive function, the PFP did add value above and beyond those individual contributions as it relates to lower cognitive performance. This finding underscores that the PFP remains an important measure of frailty that should be considered for enhancing our understanding of cognitive dysfunction and its etiologies.

Unlike the cross-sectional associations with cognitive performance, findings for the association between frailty and cognitive decline above and beyond its component parts were less consistent. It is critical to note that this does not invalidate the phenotype as a syndrome; change in cognitive performance is not a gold standard for frailty. There are 2 possible reasons that may account for this discrepancy. The first is a methodological consideration. It may be a function of lack of power due to the presence of 3-way interactions; a larger sample may be needed to confidently address this question. The second may suggest that the PFP can serve as a risk factor of lower cognitive performance only, but may need further investigation as a risk factor for subsequent cognitive decline; it is possible that separate biological processes, as represented by the separate indicators, may be mediating the relationship between frailty and cognitive decline. For example, concordant with findings from prior studies, gait speed, especially, is featured as an important indicator associated with cognitive change among both frail and nonfrail participants. This finding underscores its significance as a risk factor for cognitive decline in its own right, involving a complex integration of motor, perceptual, and cognitive processes such as attention, memory, and executive function (28,44,45). Subsequent investigations into potential subpatterns of PFP indicators and how they may relate to cognitive decline are warranted; such studies may pinpoint potential underlying biological mechanisms linking these indicators and cognitive decline for purposes of developing targeted intervention strategies.

Setting aside potential lack of power, when all criteria were examined together in the same model to assess the independent excess effects of frailty over its component parts, the magnitude of the estimates may inform a deeper understanding of the mechanisms of the PFP as a measure of frailty. Interestingly, weakness consistently appeared to be the sole indicator greatly affected by the presence of frailty, controlling for all other criteria. This finding could potentially support prior hypotheses that the different indicators are manifested hierarchically, reflecting frailty severity. For example, a prior study conducted among in the Women’s Health and Aging Study II found that weakness is a common first manifestation among all indicators of the PFP (3,46). Findings from this study support this observation, by demonstrating that after controlling for all other components, the PFP is associated with lower cognitive performance and decline above and beyond weakness, but not over other components that may be manifested downstream in the frailty process. Another study among participants from 2 population-based cohorts, InCHIANTI Study in Italy and the Longitudinal Aging Study Amsterdam (LASA) in the Netherlands, found that exhaustion appeared to be the earliest indicator that discriminated frail versus nonfrail participants (9 years prior to onset), though weakness was also an early indicator of frailty onset (6 years prior to onset of frailty) (47). There are 2 potential explanations for the discrepancy in findings related to exhaustion. First, exhaustion was measured differently. In both WHAS II and NHATS, exhaustion was based on self-report of “low energy” or feelings of tiredness/exhaustion, whereas in the InCHIANTI and LASA studies, exhaustion was assessed using 2 questions from the Center for Epidemiology Studies-Depression (CES-D) scale: “I felt that everything I did was an effort in the last week” and “I could not get going in the last week” (47). It is possible results from WHAS II and NHATS are more similar because the constructs that are captured are more closely aligned based on the measures used compared to those found in the InCHIANTI and LASA studies. The second is, given that lnCHIANTI and LASA were population-based studies like NHATS, discrepancies in the initial manifestations of frailty components may be attributable to different case-mixes presented in the United States versus in Italy or the Netherlands; it may be that with varying health profiles as a reflection of different lifestyles, organ-specific etiologic pathways with different rates of frailty progression are more prominently being captured by these different populations as opposed to the systemic dysregulations of physiologic aging (46). Further investigation into the hierarchical ordering of PFP criteria may be warranted to better understand the early warning signs of increasing vulnerability in early frailty development among other large, diverse samples.

This study has limitations. As mentioned previously, one limitation is that many of the approaches herein used multiple interactions which may have limited the power to observe significant findings. However, even where significance is not obtained, in many cases, the magnitude of the observed excess effect coefficients tended to be larger than the coefficients of the criteria, which warrants follow-up to this question in larger studies. Another limitation is that physical activity was measured via self-report, which is susceptible to social desirability bias, recall bias due to the limited abilities to remember details over the past month, and lacks the ability to capture a comprehensive range of activities in which older adults engage in on a daily basis; it is therefore possible that use of objective measures may lead to different inferences. Additionally, this study invites the common challenge concerned with using an assortment of cognitive performance measures with varying sensitivities to study global cognitive performance. Though validated, screening tools for global cognitive impairment exist, such as the Montreal Cognitive Assessment (48) and the Modified Mini-Mental State Examination (49), such assessments are lacking in NHATS. Domain-specific tests in NHATS were chosen for their speed and ease of use. These qualities are especially desirable for large, epidemiologic studies that must consider minimizing the burden for participants who are asked to undergo hours of in-person testing in diverse evaluation settings. Another consideration is that, despite statistical significance in many instances in this study, standardized coefficients represent small effects based on the standard that estimates of 0.10 are small, and 0.30 are moderate in size (50).

Nevertheless, this large study of U.S. Medicare beneficiaries aged 65 years and older validates the PFP as a measure that captures a distinct syndrome with synergistic properties. For purposes of risk stratification, slow gait speed may be a simpler, quicker, and more efficient screening tool to identify vulnerable older adults aged 65 years and older with lower cognitive performance and risk of subsequent cognitive decline. Frailty as measured by the PFP can sometimes be challenging to measure given time constraints, imprecision in self-reported items, and inability to complete a task (51). However, both the PFP and slow gait speed are important risk factors of low cognitive performance, and should be considered for enhancing our understanding of etiologies underlying cognitive dysfunction. Despite the existing relationships between the individual PFP criteria and cognition, based on our findings, the phenotype does add value in the study of cognitive performance above and beyond the confluence of its component parts. However, further research into the unitary construct of frailty as a “syndrome” correlated with cognitive decline and other adverse outcomes is needed. Lingering questions remain as to the synergistic effects of different subpatterns of PFP criteria occurrence among older adults as they relate to cognitive performance, which may lend insight into operational choices of cutoffs defining the presence of frailty. Innovative approaches, such as machine learning methods, are warranted to address this question efficiently.

Funding

This work was supported by the National Institute on Aging at the National Institutes of Health (T32AG000247 and K01AG064040 to N.M.C., R03AG053743 to Q.-L.X., P30AG021334 and P50AG005146 to K.B.-R., and K01AG050699 to A.L.G.).

Author Contributions

N.M.C.: participated in concept design, data analysis, interpretation, drafting, critical revision, and approval of the article. K.B.-R.: participated in concept design, interpretation, critical revision, and approval of the article. Q.-L.X.: participated in interpretation, critical revision, and approval of the article. M.C.C.: participated critical revision and approval of the article. A.R.S.: participated in concept design, critical revision, and approval of the article. A.L.G.: participated in concept design, interpretation, critical revision, and approval of the article.

Conflict of Interest

None declared.