Beyond the Health Deficit Count: Examining Deficit Patterns in a Deficit-Accumulation Frailty Index

Abstract

Objective:

Exploring deficit patterns among frail people may reveal subgroups of different prognostic importance.

Design:

Analysis of National Health and Aging Trends Study

Setting:

Community

Participants:

Community dwelling older adults with mild to moderate frailty (deficit-accumulation frailty index [FI] of 0.25–0.40) (n=1821).

Measurements:

Latent class analysis identified distinct clinical subgroups based on comorbidity (range:0–10), NHATS dementia classification, and short physical performance battery (SPPB) (range:0–12). Survival analyses compared 5-year mortality by subgroups.

Results:

Three latent classes existed: Class 1 (n=831, mean FI=0.30) had 2.7% probable dementia, high comorbidities (mean=3.6), and low physical impairment (SPPB mean=9.9); Class 2 (n=734, mean FI=0.32) had 6.9% probable dementia, low comorbidities (mean=2.8), and moderate physical impairment (SPPB mean=6.2); Class 3 (n=256, mean FI=0.34) had 20.7% probable dementia, low comorbidities (mean=2.4), and high physical impairment (SPPB mean=2.0). Compared to Class 1, Classes 2 and 3 experienced higher 5-year mortality (C2: 1.28 [95%CI, 1.00–1.62]; C3: 1.87 [95% CI, 1.29–2.73]).

Conclusion:

Deficit patterns among the mild-to-moderately frail provide additional prognostic information and highlight opportunities for preventive interventions.

INTRODUCTION

A deficit-accumulation frailty index (FI) quantifies the total burden of health deficits across multiple clinical domains (e.g., diagnoses, disability, cognitive function, and physical function) as a proportion.¹ Cut-points are often used to classify people into robust (FI <0.15), pre-frail (0.15 ≤ FI < 0.25), mild-to-moderately frail (0.25 ≤ FI ≤ 0.40), and severely frail categories (FI >0.40).^{2, 3} Because deficit-accumulation frailty indices do not require measurement of specific health deficit items, but rather consider cumulative burden from equally weighted items, they offer flexibility to be used in various surveys and routine health care databases (e.g., administrative claims data).^4–7

While people in the extremes of frailty are clinically distinguishable in deficit patterns (i.e., those with severe frailty have deficits in almost every clinical domain), those with mild-to-moderate frailty may have a deficit burden across heterogeneous clinical domains, where the deficits in no single clinical domain dominate.¹ As an FI is increasingly used in clinical settings to inform prognosis and influence clinical decision-making^{8, 9}, further investigation into deficit patterns may provide additional insight into individualized clinical management. For example, despite a comparable FI level, a person with many comorbidities but little cognitive or physical function impairment compared to one with fewer comorbidities but high physical function impairment may require different clinical management. Yet little is known about the deficit patterns of people with mild-to-moderate FI and the extent to which these patterns relate to mortality.

This study aimed to identify clinically different groups in a nationally-representative cohort of older adults with mild-to-moderate frailty. We used latent class analysis to identify underlying deficit patterns and examined the association of latent classes within mild-to-moderate frailty with 5-year mortality.

METHODS

Data source

We used National Health and Aging Trends Study (NHATS) data, 2011–2016.¹⁰ NHATS is sponsored by the National Institute on Aging in cooperation with the Johns Hopkins Bloomberg School of Public Health. NHATS is a nationally-representative sample of 8245 Medicare beneficiaries aged 65 years or older, which was assembled in 2011 using a stratified three-stage sampling design with oversampling of the oldest ages and non-Hispanic Black people.¹⁰ Subsequent annual assessments were conducted to assess health status and mortality. We used baseline assessment data from 2011 and proxy-reported mortality from follow-up assessments. The study protocol was determined to be exempt by our Institutional Review Board.

Study population

Our target population was community dwelling Medicare beneficiaries aged 65+ with mild-to-moderate frailty. Of the 7197 community dwelling NHATS participants (87.3% of the NHATS sample), 6871 had sufficient data to calculate a 35-item deficit-accumulation FI and 2110 participants had a mild-to-moderate FI between 0.25 and 0.40. We excluded anyone missing information on the NHATS dementia classification (n=18) and the Short Physical Performance Battery (SPPB) (n=273). The final sample included 1821 participants.

Measurements

1. Frailty index to define target population: We calculated a 35-item deficit-accumulation FI following standard procedures¹¹ used in previous analyses of NHATS.¹² Items were selected in clinical domains of comorbidities, cognition function, and physical function.

2. Clinical domains for latent class analysis: Latent classes were defined (see methods below) using three clinical domains of comorbidity burden (comorbidity count), cognitive function determined by the NHATS dementia classification, and physical function (SPPB). These clinical domains contribute to the FI calculation and represent the potential for distinct clinical management. The comorbidity index was a count of ten self-reported diagnosed health conditions (hypertension, arthritis, diabetes, dementia, cancer, myocardial infarction, osteoporosis, heart disease, lung disease, and stroke). The NHATS dementia classification has three categories (no dementia, possible dementia, and probable dementia) derived from scores on three cognitive tests: orientation (e.g., today’s date, naming the president and vice president; range 0–8), memory (e.g., immediate and delayed word recall; range 0–20), and executive function (e.g., clock drawing; range 0–5). Scores below 1.5 standard deviations of the mean on at least two domains were indicative of probable dementia or the case where it is likely a person has dementia.¹³ Physical function was characterized by the SPPB score, a measure of lower extremity function based on gait speed, standing balance, and chair stands (range 0–12, a lower score indicates greater mobility impairment).¹⁴

3. Baseline characteristics of interest: We included age, sex, race/ethnicity, educational attainment, marital status, annual household income, social isolation, and region of residence. Social isolation was determined as an aggregate measure of social network size, if the participant lived with or talked to someone, attended religious services, or participated in other social activities.¹⁵

4. All-cause mortality based on up to 5 years of annual follow up: In the event a participant has died, a proxy is interviewed about the participant’s last month of life, including date of death. Survival time was measured from date of baseline assessment to death (n=339), last known follow-up date, or December 31, 2016. Participants alive at follow-up were considered censored observations.

Statistical Analysis

Latent class analysis

Latent class analysis was performed using gsem (Stata version 15) to identify distinct unobserved groups by the comorbidity index (continuous), NHATS dementia classification (ordinal), and SPPB score (continuous).^{16, 17} We tested the fit of one to five latent classes by comparing BIC across unweighted models because survey weighting procedures invalidated the assumptions for maximum likelihood precluding use of Akaike Information Criterion (AIC) or Bayesian Information Criterion (BIC). We then applied survey weights to fit our final model (See Supplemental Table 1). Differences in characteristics of latent classes were tested by chi-square test and analysis of variance for categorical and continuous variables, respectively. Multinomial logistic regression predicted membership in Classes 2 and 3 relative to Class 1 based on age, sex, race/ethnicity, marital status, annual household income, and social isolation. Kaplan-Meier method estimated five-year mortality for each latent class. Multivariable Cox proportional hazards regression assessed the risk of 5-year mortality by latent class, after adjusting for age, sex, race/ethnicity, marital status, annual household income, and FI. The assumption of proportional hazards across classes was satisfied based on Schoenfeld residuals. All analyses accounted for the NHATS complex sampling design and were weighted to reflect national estimates.¹⁸

RESULTS

We studied 1821 participants, representing an estimated 8040261 older adults with mild-to-moderate frailty in the United States. Overall, the mean FI was 0.31 (standard error [SE]: 0.001). The mean number of comorbidities was 3.2 (SE: 0.03), 5.3% of the population had probable dementia, 13.2% had possible dementia, and 81.5% had no dementia, the mean SPPB was 7.7 (SE: 0.07).

Latent class analysis

Comparing the BIC, we identified three latent classes (Supplemental Table 1 and Supplemental Figure 1). Mean FI was similar across the groups: Class 1 (n=831) mean FI=0.30 (SE: 0.002), Class 2 (n=734) mean FI=0.32 (SE: 0.002) and Class 3 (n=256) mean FI=0.34 (SE: 0.003) (Table 1). Class 1 was multimorbid—having low prevalence of probable dementia (2.7%), high comorbidity burden (mean 3.56, SE: 0.04), and high physical function (SPPB mean 9.93, SE: 0.05). Class 2 was cognitively impaired— having moderately high prevalence of probable dementia (6.9%), but physically functional (SPPB mean 6.16, SE: 0.05) with low comorbidity (mean 2.85, SE: 0.04). Class 3 was cognitively and physically impaired— having high prevalence of probable dementia (20.7%), and poor physical function (SPPB mean 1.99, SE: 0.10), but low comorbidities (mean 2.39, SE: 0.08).

Table 1.

Baseline Characteristics of Latent Classes Among Older Adults with Mild-Moderate Frailty (n=1821)

Characteristics	Class 1 831 (45.6%)a	Class 2 734 (40.3%)a	Class 3 256 (14.1%)a	P-value
Frailty Index, Mean (SEb)	0.30 (0.002)	0.32 (0.002)	0.34 (0.003)	<.001
NHATS dementia classification domain scores, Mean (SE)
Orientationc	6.28(0.07)	5.81(0.08)	4.96(0.13)	<.001
Memoryd	8.64(0.13)	7.31(0.14)	5.50(0.21)	<.001
Executive functione	3.67(0.03)	3.33(0.05)	2.91(0.09)	<.001
Comorbidity index
Mean (SE)	3.56(0.04)	2.85(0.04)	2.39(0.08)	<.001
SPPBf
Mean (SE)	9.93(0.05)	6.16(0.05)	1.99(0.10)	<.001
Age group
65–74	370(56.0)	199(37.1)	46(24.4)	<.001
75–79	190(21.7)	157(23.4)	45(21.6)
80–84	166(14.6)	212(23.6)	73(26.0)
85–89	78(6.2)	111(11.7)	45(16.7)
90 +	27(1.5)	55(4.1)	47(11.3)
Sex
Male	379(45.4)	262(36.2)	87(34.3)	<.001
Female	452(54.6)	472(63.8)	169(65.7)
Race/Ethnicity
Non-Hispanic White	622(83.1)	498(80.2)	157(73.8)	0.001
Non-Hispanic Black	131(5.6)	176(9.3)	69(11.9)
Other	78(11.3)	60(10.5)	30(14.2)
Educational attainment
High School or less	486(57.7)	513(67.7)	178(66.8)	<.001
Some college or more	344(42.3)	220(32.3)	77(33.2)
Marital status
Married/Partner	538(68.9)	346(51.4)	84(38.2)	<.001
Separated/Divorced	66(8.1)	98(13.1)	24(10.8)
Other (Widowed, never married)	227(23.0)	290(35.6)	148(51.0)
Annual household income
<= $13,200	152(16.1)	203(24.5)	100(36.3)	<.001
>$13,200-$44,999	410(48.9)	374(51.4)	113(43.2)
>=$45,000	269(35.0)	157(24.1)	43(20.5)
Social Isolation
Isolated	255(30.8)	288(38.9)	124(43.9)	<.001
Integrated	576(69.2)	446(61.1)	132(56.1)
Region of residence
Northeast	149(18.5)	145(21.8)	55(22.3)	0.037
Midwest	184(22.5)	184(23.3)	68(27.3)
South	319(37.3)	288(38.5)	92(34.0)
West	179(21.7)	117(16.3)	41(16.5)
Self-reported diagnosis
Hypertension	684(81.8)	552(73.6)	177(64.9)	<.001
Arthritis	587(72.1)	498(68.1)	145(55.5)	<.001
Diabetes	309(37.0)	204(29.2)	71(29.4)	0.006
Dementia	26(2.6)	18(2.2)	--	0.782
Cancer	325(38.6)	198(28.1)	45(19.0)	<.001
Myocardial infarction	222(25.9)	112(14.4)	42(15.9)	<.001
Osteoporosis	229(29.4)	155(23.4)	46(18.4)	0.001
Heart disease	261(31.0)	137(18.9)	39(15.0)	<.001
Lung disease	198(25.4)	108(15.5)	29(11.2)	<.001
Stroke	112(12.9)	86(11.7)	19(7.3)	0.079

Note. All proportions are weighed to reflect national estimates. – indicates cell size smaller than allowable by NHATS

^an(%) out of n=1821; Class 1: multimorbid, yet functional; Class 2: cognitively impaired; Class 3: cognitively and physically impaired;

^bSE=sampling scheme adjusted standard error;

^cOrientation: range 0–8, lower scores indicate higher impairment;

^dMemory: range 0–20, lower scores indicate higher impairment;

^eExecutive function: range 0–5, lower scores indicate higher impairment;

^fSBBP=Short Physical Performance Battery: range 0–12, lower scores indicate higher impairment

Characteristics of each class

Compared with Classes 1 and 2, Class 3 was comprised of higher proportions of people age 90 years or older (Class 1: 1.5%; Class 2: 4.1%; Class 3: 11.3%), Hispanic and other racial/ethnic groups (11.3%; 10.5%; 14.2%), isolated people (30.8%; 38.9%; 43.9%), and those with annual household income below $45,000 (65.0%; 75.9%; 79.5%) (Table 1).

Adjusted odds ratios (aOR) and 95% CIs in Figure 1 show that, compared to Class 1, members of Classes 2 and 3 were more likely to be older, of non-Hispanic Black race, and divorced or separated (Class 2) or in another marital situation (Class 3). In addition, Class 3 was less likely to have household incomes greater than $13,200. (See also Supplemental Table 2)

Figure 1.

Demographic Characteristics Associated with Membership to Latent Classes 2: Cognitively Impaired and 3: Cognitively and Physically Impaired, Compared to Class 1: Multimorbid, yet Functional

Note: All analyses are multinomial logistic regression models weighted to reflect national estimates. Reference groups are as follows: Age <80, Male, Non-Hispanic White, Married/Partnered, Annual income <13,200, socially integrated.

Five-year mortality

Mean follow-up time was 43.6 months (standard deviation [SD] 19.0 months). Mean follow-up was 46.4 months (SD 18.4 months) for censored events and 31.2 months (SD 16.7 months) for those who died. The 5-year incidence of mortality was 24.1%.

Kaplan-Meier estimates of five-year mortality by latent class were (Figure 2): Class 1: 18.6%; Class 2: 25.1%; Class 3: 40.7%. After multivariable adjustment, Classes 2 and 3 experienced a higher rate of 5-year mortality compared to Class 1 (adjusted hazard ratio (aHR), Class 2: 1.28 [95%CI, 1.00–1.62]; Class 3: 1.87 [95% CI, 1.29–2.73]).

Figure 2.

Kaplan-Meier Curve Estimating Probability of Five-year Mortality by Latent Classes (Class 1: Multimorbid, yet Functional; Class 2: Cognitively Impaired; Class 3: Cognitively and Physically Impaired)

DISCUSSION

We identified three latent classes among mild-to-moderately frail older adults which were characterized by being multimorbid (Class 1), cognitively impaired (Class 2), or cognitively and physical impaired (Class 3). Despite similar deficit-accumulation FI levels, having both probable dementia and physical impairment, is associated with greater 5-year mortality risk than having multimorbidity or cognitive impairment alone.

The heterogeneity of the three latent classes highlights important differences in prognostic information that would otherwise remain unknown relying solely on the FI. Our findings were also consistent with other research demonstrating high prevalence of single domain impairment¹⁹, which is associated with increased mortality risk if the domain is worse cognitive impairment, as in Class 2, or the domain is a growing number of comorbidities, as in Class 1^{20, 21}. However Class 3, dominated by concomitant cognitive and physical function impairment, had nearly twice the risk of 5-year mortality as Class 1, a risk consistent with the increases mortality risk for both cognitive impairment and SPPB scores below 10²⁴, independently and also with previous research demonstrating excess mortality risk with frailty phenotype concurrent with cognitive impairment compared to frailty phenotype without cognitive impairment^{22, 23} Therefore by demonstrating that people with mild to moderate frailty have similar FI, but based on composition of deficits they have distinct 5-year mortality risk, we were able to directly address the lack of specificity in the type of frailty that FI measures, an issue raised by Xue and Varadhan.²⁵

Consequently, observing deficit patterns in key clinical domains, may enhance the utility of a deficit-accumulation FI -- and potentially other frailty measures, such as Fried phenotype, where heterogeneity is likely present at each frailty stage²⁶ -- for clinical management²⁷ by offering opportunities for preventive interventions to improve frailty or lower mortality risk. For example, research on physical training interventions have shown improved SPPB and reduction in frailty, which could benefit those with physical impairments in Class 3.²⁸ The same is true for comorbidity burden as in Class 1; optimizing disease management and care quality has been shown to prevent excess disease burden for people with arthritis²⁹ and reduce mortality risk for people with diabetes³⁰ or hypertension.³¹ Moreover, dementia, such as in Class 2 and 3, is a progressive, gradual decline in cognitive and functional abilities³² and likely explains the excess estimated mortality.³³ Though known dementia interventions do not reduce mortality, certain interventions may improve quality of life for people living with dementia, for example, discontinuing psychotropic medications, and ensuring a balanced nutritious diet and exercise.³⁴ Our findings emphasize the importance of identifying deficit patterns that could be useful to health care providers in recognizing potentially vulnerable subgroups (e.g., those with cognitive and physical function impairment) and could provide a basis for individualized clinical management.

Strengths and limitations

Latent class analysis is an accepted method for identifying unobserved subgroups, however results can vary based on the classification criteria and algorithm used. We relied on standardized assessments of cognitive function and physical function, and comorbidity for classification criteria. The comorbidity index³⁵, NHATS dementia classification¹³, and SPPB¹⁴ have demonstrated validity in other studies. However, number of comorbidities may have been under-reported due to self-report, which may reduce the impact of the comorbidity index. In addition, we chose not to adjust for demographic characteristics (as typically done in conventional latent class analysis) because our focus was to create groups as if a clinician would evaluate based on the observed health characteristics (comorbidity burden, cognitive and physical function) without statistical adjustment. Even when we adjusted for sociodemographic variables, the associations between those variables and groups were similar to our analysis. Because our sample included community-dwelling Medicare beneficiaries with mild-to-moderate frailty, results may not generalize to institutionalized Medicare beneficiaries, whom, we have noted, are less likely to demonstrate deficit patterns such as those described here.

Conclusions

Examining deficit patterns provides additional prognostic information among older adults with mild-to-moderate frailty. Knowing the specific deficits uncovers important targets for intervention that may be useful for individualized clinical management for this vulnerable population.