Evaluating Distinctions in the Assessment of Late-Life Disability

Abstract

Background

By including categories for accommodations and reduced frequency, to supplement earlier classifications of difficulty and dependence, a new 5-category functional scale has the potential to distinguish finer gradations in disability but the hierarchical nature and advantages of this scale over alternative scales are uncertain.

Methods

Using data from the National Health and Aging Trends Study (N = 7,609), we conducted a series of: cross-sectional analyses that focused on the hierarchical consistency of responses in the 5-category scale; and longitudinal analyses that focused on predictive accuracy for mortality and functional dependence, comparing the 5-category scale with three simpler scales, having only three or four categories.

Results

Although there was considerable variability across the seven functional activities, the prevalence of inconsistencies in the hierarchy of the 5-category scale (eg, reports difficulty but no accommodations) was relatively high. In addition, the predictive accuracy of the 5-category scale for mortality and functional dependence was only modestly better than that of the two 3-category scales and was comparable to that of the 4-category scale. Finally, when evaluated as summative disability scores, there was little difference in predictive accuracy between the 5-category scale and three alternative scales.

Conclusions

Despite inconsistencies in its hierarchy, the 5-category scale is more effective in stratifying risk for mortality and functional dependence than the two 3-category scales but not the 4-category scale. When assessing changes over time, however, the addition of questions on accommodations and reduced frequency to those on difficulty and dependence, to form a summative disability score, offers little benefit and increases the burden of the assessment.

Functional assessment is the cornerstone of geriatric practice (1). Assessing functional status, including self-reported ability to perform daily activities such as bathing, dressing, transferring, and walking, serves as the basis for ascertaining and quantifying the severity of disability among older persons. Late-life disability is associated with increased mortality (2,3) and leads to additional adverse outcomes, such as nursing home placement and greater use of home care services (4–7), all of which place a substantial burden on older persons, informal caregivers, and health care resources (8–10). Over the next 30 years, spending on long-term care for the elderly is projected to more than double from $160.7 to $346.1 billion (11).

Despite a burgeoning literature, the science of functional assessment has generally lagged behind advances in the epidemiology, prevention and treatment of disability (12–16). In prior work (3), we found that participant responses to questions about difficulty and dependence (ie, personal assistance) provide complementary information that together depicts the continuum of disability more fully than does the response to either question alone. These results informed the development of functional scales that have served as disability outcomes in longitudinal studies and clinical trials, with each activity in the scale scored as 0 (no difficulty or need for help), 1 (difficulty but no need for help), and 2 (need for help) (17–20). In the Women Health and Aging Study II, an additional category of modification, which was intermediate to difficulty and no help, was found to predict the subsequent need for personal assistance (21).

More recently, Freeman and colleagues proposed a 5-category hierarchy of late-life disability, with activities scored as 0 (fully able), 1 (accommodated declines by using assistive technology or environmental features), 2 (reduced frequency but has no difficulty), 3 (has difficulty [despite accommodations] but receives no personal assistance), and 4 (receives personal assistance) (22). By including categories for accommodations and reduction in frequency, to supplement earlier classifications of difficulty and dependence, this new hierarchy has the potential to distinguish finer gradations in disability.

Based on data from the National Health and Aging Trends Study (NHATS), a nationally representative sample of older Americans, about a third of persons aged 65 years or older were fully able to carry out their self-care and mobility activities, a fifth received personal assistance, and nearly half did not require personal assistance but were not fully able by virtue of accommodations, reduced frequency, or difficulty (22). Although the hierarchy across these latter three categories was generally supported through a series of cross-sectional analyses, clinical experience suggests that these categories might confer comparable vulnerability on the pathway to functional dependence and death. In the setting of equivalent reductions in physical, cognitive, or other capabilities, some older persons might first try to accommodate, while others might first reduce frequency, while the remainder might first report difficulty. If these three categories were truly hierarchical, one would expect persons who report difficulty with a task to also report accommodations and reduced frequency and those who report reduced frequency to also report accommodations but the consistency of responses across these categories has not been previously evaluated. An additional and perhaps more rigorous approach for determining the value of distinguishing among these categories would be a longitudinal analysis that evaluated associations with relevant outcomes.

The goals of the current study were to assess the hierarchical nature of the NHATS late-life disability scale and to evaluate the advantages of this scale over alternative scales that have only three or four categories. To accomplish these goals, we conducted a series of cross-sectional analyses that focused on the hierarchical consistency of responses and longitudinal analyses that focused on predictive accuracy for subsequent mortality and functional dependence. We postulated that the predictive accuracy of the 5-category hierarchical scale would not be much better than that of three simpler scales that (i) considered accommodations, reduced frequency, and difficulty to be comparable indicators of vulnerability; (ii) included difficulty as the sole indicator of vulnerability; or (iii) combined accommodations and reduced frequency into a single category of modification. Because of known sex differences in late-life disability (23–25), we repeated the analyses separately in men and women.

Methods

Study Population

Publically available, deidentified data, together with the sensitive data files, were used from NHATS (22). On September 30, 2010, NHATS drew a random sample of persons 65 years or older living in the contiguous United States from the Medicare enrollment file, with oversampling of non-Hispanic blacks and those 90 years or older. Baseline (or Round 1) interviews, completed in May through November 2011, yielded a sample of 8,245 persons with a weighted response rate of 71.3%. The sample for the current analysis includes the 7,609 participants who were living in settings other than nursing homes. Proxy respondents were interviewed when the participant could not respond (n = 583 or 5.8% [weighted]). The Johns Hopkins University Institutional Review Board approved the NHATS protocol and all participants provided written informed consent.

Data Collection

In-person assessments were completed annually for 4 years (Rounds 1–4) by trained research staff.

Assessment of covariates

Round 1 data were collected on demographic characteristics, including age (in years), sex, race/ethnicity, education, and living situation; 10 self-reported, physician diagnosed chronic conditions, including myocardial infarction, hypertension, arthritis, osteoporosis, diabetes, lung diseases, stroke, dementia, cancer, and hip fracture (since age 50); physical performance; and cognitive status. Physical performance was assessed with the short physical performance battery (26,27), with scores ranging from 0 (lowest) to 12 (highest), whereas cognitive status was assessed with a 10-word recall (28).

Assessment of disability

Disability was assessed in four self-care activities (eating, getting cleaned up, using the toilet, and dressing) and three mobility activities (going outside, getting around inside, and getting out of bed) (22). For each activity, participants were asked about their use of devices or environmental modifications (canes, walkers, wheelchairs, scooters, grab bars, bath or shower seat, eating, and dressing devices) and receipt of help from another person during the prior month. Participants who ever performed the activity without personal assistance were asked whether they had difficulty completing the activity alone (with the particular devices or environmental modifications, if used). For activities other than getting out of bed, using toilet, and eating, participants were also asked about changes in the frequency with which they performed the activity in the past year.

NHATS scales

Based on responses to these questions, four or five hierarchical categories (scored 0 to 4) were created for each activity: fully able (0: no device use, reduction in activities, difficulty, or assistance); accommodations (1: device use but no reduction in activities, difficulty, or assistance); reduced activities (2: reductions in activities but no difficulty or assistance); difficulty (3: difficulty performing activities by oneself, when using devices, if needed, but no assistance); and assistance (4: help from another person or, rarely, not doing the particular activity) (22). A 5-category hierarchical scale, indicating the lowest level of functioning across all seven activities, was created (22). In addition, scores for the seven activities were summed to form a summative functional scale, with scores from 0 (best) to 28 (worst).

Alternative scales

For comparison, three sets of alternative scales were created. In the first, accommodations, reduced activities and difficulty were combined into a single category (or indicator) of vulnerability (score = 1), with the fully able (score = 0) and assistance (score = 2) categories unchanged. This led to creation of a 3-category hierarchical scale (fully able for all activities, vulnerable in any activity, assistance in any activity) and a summative scale, with scores from 0 (best) to 14 (worst). In the second, the sole indicator of vulnerability was difficulty (score = 1); fully able, accommodations, and reduced activity were combined into a single category denoted as independent (score = 0) and assistance (score = 2) was unchanged. This led to creation of a 3-category hierarchical scale (independent in all activities, difficulty in any activity, assistance in any activity) and a summative scale, with scores from 0 (best) to 14 (worst). In the third, accommodations and reduced frequency were combined, leading to a 4-category hierarchical scale (fully able for all activities [score = 0], modification in any activity [score = 1], difficulty in any activity [score = 2], and assistance in any activity [score = 3]) and a summative scale, with scores from 0 (best) to 21 (worst). For efficiency, the two sets of 3-category scales are referred to simply as version c (combined indicator of vulnerability) and version d (difficulty as sole indicator of vulnerability). The latter scale and 4-category scale provided a frame of reference for comparisons with previously published research (3,17–21).

Longitudinal Outcomes

The two longitudinal outcomes included death and functional dependence. Deaths were ascertained by research staff during and between the annual assessments. Month and year, but not day, of death were available from the NHATS sensitive data files. Functional dependence, ascertained during the annual assessments, was defined as the need for personal assistance with one or more of the seven self-care or mobility activities. Participants who were living in a nursing home at the time of a follow-up assessment were also categorized as functionally dependent, in part because interviews were not completed among participants in a nursing home.

Statistical Analysis

Baseline (ie, Round 1) characteristics were summarized using percentages and means (SE) and these values were compared between men and women.

For each of the seven self-care and mobility activities, the prevalence of potential inconsistencies in the NHATS hierarchical disability scale was estimated among participants who were not fully able in the specific activity at Round 1 and these values were compared between men and women using the Rao–Scott chi-square (29). Potential inconsistencies included reporting difficulty but no reduction, reporting difficulty but no accommodations, and reporting reduction but no accommodations. These analyses were repeated using data from Round 2.

Kaplan–Meier curves for survival were generated for the 5-category scale and three alternative scales. The multivariable associations between each of these scales and time to death were evaluated using proportional hazards models that were adjusted for age (in years), race/ethnicity, and education and for the overall sample, sex. The Akaike Information Criterion (AIC) value was used to compare the fit of the three models. The model with the lowest AIC value has the best fit (or predictive accuracy). These analyses were repeated for the corresponding summative scales. To facilitate comparisons, the scores for the two 3-category summative scales (range 0–14) were multiplied by two, and the score for the 4-category summative scale (range 0–21) was multiplied by 1.33 so that the hazard ratios could express an equivalent increment in risk as for the 5-category summative scale (range 0–28).

Among participants who did not require assistance at Round 1, the rates of functional dependence per 100 person-years were calculated for each level of the 5-category scale and three alternative scales. The denominators included the follow-up time (up to 3 years, excluding years when functional dependence could not be determined), while the numerators included time in the dependent state, that is, 1 year for each round meeting the operational definition. This included the development of new functional dependence and persistence of prior functional dependence. The multivariable associations between each of these scales and functional dependence were evaluated using repeated measures logistic regression models that were adjusted for age (in years), race/ethnicity, and education and for the overall sample, sex. The area under the (receiver operating characteristic) curve was used to estimate each model’s predictive accuracy. The model with the highest value has the best predictive accuracy. These analyses were repeated for the corresponding summative scales. To facilitate comparisons, the scores for the summative scales were converted to percentages of the maximum possible scores so that the odds ratios could express an equivalent increment in risk for the 3-category (maximum score: 7), 4-category (maximum score: 14), and 5-category (maximum score: 21) summative scales.

All analyses were performed using the SAS Survey Procedures and account for the complex sample design (SAS, version 9.4; SAS Institute, Cary, NC). Hence, all values other than group sizes are weighted.

Results

The Round 1 characteristics of the study participants are provided in Table 1. Compared with men, women were older and had lower educational attainment, were more likely to live alone, and had more chronic conditions but lower short physical performance battery scores.

Table 1.

Characteristics of Community-Living Participants (N = 7,609), Overall and by Sex, Round 1^a

Characteristics	All Participants		Men		Women		p Valueb
	N	% or Mean (SE)	N	% or Mean (SE)	N	% or Mean (SE)
Agec	7,609	75.3 (0.10)	3,171	74.5 (0.11)	4,438	75.9 (0.14)	<.001
Female sex	4,438	56.6	NA		NA		NA
Race/ethnicity
White, non-Hispanic	5,186	81.4	2,197	81.9	2,989	81.1	.461
Black, non-Hispanic	1,662	8.2	647	7.5	1,015	8.7
Hispanic	454	6.8	196	6.9	258	6.7
Other	225	3.6	101	3.7	124	3.5
Education
< High school	2,047	21.8	850	21.9	1,197	21.7	<.001
High school or equivalent	2,069	27.6	732	23.3	1,337	30.9
> High school	3,397	50.6	1,556	54.8	1,841	47.4
Lives alone	2,519	30.2	683	19.4	1,836	38.5	<.001
Number of chronic conditions (0–10)	7,608	2.4 (0.02)	3,170	2.2 (0.03)	4,438	2.5 (0.02)	<.001
SPPB score (0–12)	6,578	8.3 (0.06)	2,753	8.9 (0.07)	3,825	7.8 (0.08)	<.001
Word Recall (0–10)	7,064	2.7 (0.06)	2,979	2.7 (0.08)	4,085	2.7 (0.06)	.620

Note: NA = not applicable; SPPB = short physical performance battery.

^a N’s are unweighted; means, percentages and standard errors are weighted. ^bBased on Rao–Scott chi-square for categorical variables and t test for continuous measures; values account for complex sampling design. ^cMonth and year, but not day, of birth were available from the NHATS sensitive data files.

Table 2 shows the Round 1 prevalence of potential inconsistencies in the NHATS hierarchical disability scale. Among all participants, values ranged from 20.4% (bathing) to 65.9% (dressing) for reports difficulty but no reduction, from 5.7% (using toilet) to 67.4% (dressing) for reports difficulty but no accommodations, and from 5.7% (bathing) to 24.0% (going outside) for reports reduction but no accommodations. Inconsistencies tended to be more common in men than women. These differences were statistically significant for reports difficulty but no accommodations for getting around inside, getting out of bed, and using the toilet and for reports reduction but no accommodations for bathing. Comparable results were observed for Round 2 (Supplementary Table 1).

Table 2.

Prevalence of Potential Inconsistencies in NHATS Hierarchical Disability Scale, Overall and by Sex, Among Community-Living Participants Who Were Not Fully Able in Each Activity, Round 1^a

Type of Inconsistency/Activityb	All Participants			Men			Women			p Valued
	Sample Sizec	Potential Inconsistency		Sample Sizec	Potential Inconsistency		Sample Sizec	Potential Inconsistency
		N	Prevalence, % (SE)		N	Prevalence, % (SE)		N	Prevalence, % (SE)
Reports difficulty but no reduction
Going outside	2,402	551	24.4 (1.0)	881	198	25.8 (1.9)	1,521	353	23.5 (1.2)	.310
Getting around inside	2,458	918	39.8 (1.1)	842	329	42.6 (2.2)	1,616	589	38.2 (1.2)	.060
Bathing	3,295	722	20.4 (0.8)	1,222	280	22.2 (1.2)	2,073	442	19.2 (1.1)	.063
Dressing	1,528	995	65.9 (1.4)	542	359	66.2 (3.0)	986	636	65.7 (1.7)	.881
Reports difficulty but no accommodations
Going outside	2,403	376	18.1 (1.1)	880	135	18.3 (1.6)	1,523	241	18.0 (1.3)	.903
Getting around inside	2,437	631	30.2 (1.0)	835	240	33.5 (1.7)	1,602	391	28.4 (1.2)	.014
Getting out of bed	1,942	995	57.5 (1.2)	676	382	62.6 (1.7)	1,266	613	54.5 (1.5)	<.001
Eating	639	425	66.3 (2.4)	226	155	68.3 (3.6)	413	270	65.1 (2.8)	.413
Bathing	3,287	407	11.0 (0.7)	1,221	152	11.7 (1.0)	2,066	255	10.6 (1.0)	.385
Using toilet	3,331	168	5.7 (0.5)	1,022	73	8.3 (1.1)	2,309	95	4.5 (0.5)	<.001
Dressing	1,529	1,039	67.4 (1.6)	542	366	67.7 (2.8)	987	673	67.2 (1.8)	.881
Reports reduction but no accommodations
Going outside	2,975	651	24.0 (1.0)	991	234	25.2 (1.7)	1,984	417	23.4 (1.1)	.291
Getting around inside	2,621	335	13.9 (0.9)	879	115	14.0 (1.7)	1,742	220	13.8 (0.9)	.876
Bathing	3,739	238	5.7 (0.5)	1,349	102	7.0 (0.7)	2,390	136	5.0 (0.6)	.024
Dressing	1,781	279	15.2 (1.0)	638	83	13.6 (1.8)	1,143	196	16.2 (1.3)	.275

Note: NHATS = National Health and Aging Trends Study.

^aSample sizes and N’s are unweighted; percentages and standard errors are weighted. ^bParticipants were not asked about changes in frequency (ie, reduction) for getting out of bed, using toilet, and eating. ^cNumber of participants for whom both items in the potentially inconsistent pair were not missing and who were not fully able for the specific activity. ^dRao–Scott chi-square for difference in the weighted prevalence of potential inconsistency between men and women; accounts for complex sample design.

Of the 7,609 participants, 1,081 (10.9%) died over a median follow-up of 36.5 months. Figure 1 provides Kaplan–Meier survival curves for the 5-category scale (Panel A), two 3-category scales (Panels B and C), and 4-category scale (Panel D). For each of the four scales, the assistance subgroup had the lowest survival, with probabilities that were clearly distinct from those of the other subgroups. In contrast, differences in survival among the other subgroups were relatively small. This was particularly apparent for the 5-category scale where the survival curves for the reduction and difficulty subgroups were nearly superimposed. Comparable results were observed for men and women (Supplementary Figure 1).

Figure 1.

Kaplan–Meier curves for survival. (A) 5-category scale; (B) 3-category scale, version c; (C) 3-category scale, version d; (D) 4-category scale. As described in the Methods section, accommodations, reduced frequency and difficulty were combined into a single indicator of vulnerability for version c, whereas difficulty served as the sole indicator of vulnerability for version d.

The multivariable associations between each of the functional scales and time to death are provided in Table 3. The results were consistent with those of the Kaplan–Meier analyses. For the 5-category scale, the adjusted hazard ratios for the difficulty and reduction subgroups were nearly indistinguishable. The point estimate for the accommodations subgroup was within the 95% confidence interval for the reduction subgroup among all participants and among both men and women, indicating that the corresponding hazard ratios did not differ significantly. For the 3-category scales, the adjusted hazard ratios for version d were smaller than those of version c, reflecting the higher mortality rate in the reference group (ie, independent), which included participants in the accommodations and reduction subgroups, in addition to those who were fully able. For the 4-category scale, the hazard ratios were modestly greater for difficulty than modification. Although the differences were relatively small, model fit based on the AIC values was best for the 5-category scale, intermediate for the 4-category scale and version c of the 3-category scale, and worst for version d of the 3-category scale. Similar results were observed for men and women. When evaluated as summative scores, ranging from 0 to 28 (Supplementary Table 2), the adjusted hazard ratios per each 1-point increment were comparable for the four scales, although model fit for version c of the 3-category scale was better than those of the other three scales based on the AIC values. There were no sex-specific differences in these associations or model fit.

Table 3.

Multivariable Associations Between Categorical Functional Scales at Round 1 and Time to Death, Overall and by Sex^a

Functional Scale	All Participants			Men			Women
	N	% Deceased	HR (95% CI)	N	% Deceased	HR (95% CI)	N	% Deceased	HR (95% CI)
5-category
Assistance	1,716	30.4	6.27 (4.46, 8.82)	506	34.7	5.63 (3.86, 8.23))	1,210	28.4	7.42 (4.01, 13.7)
Difficulty	1,507	10.1	2.28 (1.71, 3.04)	618	11.7	2.02 (1.43, 2.84)	889	8.9	2.75 (1.55, 4.86)
Reduction	470	10.3	2.32 (1.59, 3.37)	170	12.6	2.21 (1.30, 3.76)	300	8.9	2.59 (1.48, 4.52)
Accommodations	1,854	7.0	1.59 (1.11, 2.28)	741	9.0	1.56 (1.02, 2.38)	1,113	5.6	1.72 (0.98, 3.02)
Fully able	2,061	3.5	ref	1,136	4.5	ref	925	2.2	ref
Model AICb			123,516,278			52,820,405			65,483,542
3-category, version c
Assistance	1,716	30.4	6.23 (4.43, 8.77)	506	34.7	5.60 (3.83, 8.18)	1,210	28.4	7.36 (3.98, 13.6)
Vulnerablec	2,831	8.5	1.94 (1.45, 2.59)	1,529	10.5	1.81 (1.30, 2.51)	2,302	7.2	2.20 (1.29, 3.74)
Fully Able	2,061	3.5	ref	1,136	4.5	ref	925	2.2	ref
Model AICb			123,564,265			52,834,926			65,521,529
3-category, version d
Assistance	1,716	30.4	4.37 (3.51, 5.45)	506	34.7	4.21 (3.18, 5.58)	1,210	28.4	4.59 (3.37, 6.25)
Difficulty	1,507	10.1	1.62 (1.30, 2.02)	618	11.7	1.53 (1.11, 2.09)	889	8.9	1.72 (1.25, 2.37)
Independentd	4,385	5.5	ref	2,047	6.6		2,338	4.5
Model AICb			123,616,729			52,872,233			65,534,818
4-category
Assistance	1,716	30.4	6.24 (4.44, 8.78)	506	34.7	5.61 (3.84, 8.19)	1,210	28.4	7.39 (3.99, 13.7)
Difficulty	1,507	10.1	2.28 (1.71, 3.03)	618	11.7	2.01 (1.43, 2.84)	889	8.9	2.74 (1.55, 4.85)
Modificatione	2,324	7.6	1.73 (1.22, 2.43)	911	9.7	1.67 (1.11, 2.52)	1,413	6.3	1.89 (1.10, 3.25)
Fully able	2,061	3.5	ref	1,136	4.5	ref	925	2.2	ref
Model AICb			123,536,178			52,828,644			65,495,865

Note: AIC = Akaike Information Criterion; CI = confidence interval; HR = hazard ratio; ref = reference group.

^aProportional hazards models were adjusted for age in years, race/ethnicity, and education and for the overall sample, sex. The percentages and hazard ratios represent weighted values, and the N’s represent the number of participants at risk. ^bLower values indicate better model fit. ^cIncludes accommodations, reduction, and difficulty. ^dIncludes fully able, accommodations, and reduction. ^eincludes accommodations and reduction.

Among participants who did not require assistance at Round 1, the rate of functional dependence (development and persistence) over the 3-year follow-up period was 10.6 per 100 person-years. As shown in Table 4, the rates of functional dependence among all participants increased across the 5-category and 4-category scales, from fully able (4.5 per 100 person-years) to difficulty (22.0 per 100 person-years), and was considerably higher for the vulnerable and difficulty groups than the fully able and independent groups for versions c and d of the 3-category scales, respectively. The multivariable odds ratios followed a similar pattern. Predictive accuracy, as assessed by area under the (receiver operating characteristic) curve, was highest for the 5-category and 4-category scales, intermediate for version d of the 3-category scale, and lowest for version c of the 3-category scale. Comparable results were observed for men and women. When evaluated as summative scores (Supplementary Table 3), the adjusted odds ratios per each 10 percent increment were lower for the 5-category scale than for any of the other scales. Predictive accuracy, however, was lowest for version d of the 3-category scale and comparable for the 5-category scale, 4-category scale, and version c of the 3-category scale.

Table 4.

Multivariable Associations Between Categorical Functional Scales at Year 1 and Functional Dependence Over 3 y, Overall and by Sex^a

Functional Scale	All Participants			Men			Women
	Outcome Rate Per 100 Person-Years	OR (95% CI)	AUCb	Outcome Rate Per 100 Person-Years	OR (95% CI)	AUCb	Outcome Rate Per 100 Person-Years	OR (95% CI)	AUCb
5-category
Difficulty	22.0	4.69 (3.80, 5.79)		17.8	3.99 (2.89, 5.51)		24.9	5.39 (4.00, 7.27)
Reduction	13.3	2.36 (1.70, 3.26)		12.9	2.61 (1.50, 4.55)		13.6	2.34 (1.57, 3.46)
Accommodations	9.1	1.66 (1.28, 2.15)	0.739	8.3	1.67 (1.17, 2.38)	0.727	9.6	1.72 (1.26, 2.36)	0.741
Fully able	4.5	ref		4.4	ref		4.6	ref
3-category, version c
Vulnerablec	14.5	2.78 (2.25, 3.45)	0.712	12.6	2.62 (1.93, 3.56)	0.708	15.7	2.99 (2.26, 3.96)	0.707
Fully able	4.5	ref		4.4	ref		4.6	ref
3-category, version d
Difficulty	22.0	3.26 (2.82, 3.76)	0.731	17.8	2.91 (2.29, 3.71)	0.715	24.9	3.53 (2.90, 4.29)	0.734
Independentd	7.2	ref		6.3	ref		7.9	ref
4-category
Difficulty	22.0	4.68 (3.80, 5.78)		17.8	3.99 (2.89, 5.50)		24.9	5.38 (3.99, 7.26)
Modificatione	9.9	1.79 (1.39, 2.29)	0.738	9.2	1.84 (1.30, 2.58)	0.726	10.3	1.83 (1.35, 2.49)	0.740
Fully able	4.5	ref		4.4	ref		4.6	ref

Note: AUC = area under the receiver operating characteristic curve; CI = confidence interval; OR = odds ratio; ref = reference group.

^aAll values are weighted to account for the complex sampling design. Odds ratios were adjusted for age (in years), race/ethnicity and education and for the overall sample, sex. ^bHigher values are better; potential range 0.5–1. ^cIncludes accommodations, reduction, and difficulty. ^dIncludes fully able, accommodations, and reduction. ^eIncludes accommodations and reduction.

Discussion

The current study evaluated the hierarchical nature of the NHATS 5-category late-life disability scale and compared the predictive accuracy of this scale with that of three alternative scales that have only three or four categories. Although there was considerable variability across the different functional activities, the prevalence of inconsistencies in the hierarchy of the NHATS scale was relatively high. In addition, the predictive accuracy of the NHATS scale for mortality and functional dependence was only modestly better than that of the two 3-category scales and was comparable to that of the 4-category scale. Finally, when evaluated as summative scores, there was little difference in predictive accuracy between the NHATS scale and three alternative scales. These results, which were generally consistent between men and women, raise questions about the hierarchical nature of the NHATS 5-category scale and suggest that it may not offer significant benefits over simpler scales having only three or four categories.

The supposition underlying the NHATS late-life disability scale is that the categories of accommodations, reduced frequency, and difficulty represent a hierarchy, from least impaired to most impaired, on the pathway to functional dependence and death. If these three categories were truly hierarchical, one would expect persons who report difficulty with a task to also report accommodations and reduced frequency and those who report reduced frequency to also report accommodations. The high prevalence of inconsistencies, however, suggests that this hierarchy is often not maintained, reflecting the inherent complexity of the disabling process. For example, as older persons develop functional problems, they may first report difficulty with no accommodations. They may then begin to restrict their activities. As they come to terms with their functional problems, they may subsequently accept equipment accommodations, thereby improving in the hierarchy even as their disability progresses.

Our results provide only partial support for the supposition that these three categories (accommodations, reduced frequency, and difficulty) confer comparable vulnerability on the pathway to functional dependence and death. While the risk of death was comparable for difficulty and reduction, the risk of functional dependence was greater for difficulty than reduction and the lowest risk for each of these two outcomes was observed for accommodations. Of the five categories in the hierarchy, reduction may be the most problematic. The category is not applicable for three of the seven activities (getting out of bed, using toilet, and eating), its prevalence is considerably lower than that of the other four categories, its test–retest reliability is uncertain and, in an earlier report (22), the graded relationships between it and the adjacent categories were rather weak for a series of physical and cognitive capacity measures.

By including categories for accommodations (primarily through use of adaptive equipment) and reductions in frequency of completing a subset of activities, to supplement traditional classifications of difficulty and dependence, the NHATS scale attempts to distinguish finer gradations in disability. A similar approach was taken in the Women Health and Aging Study (WHAS) II, although accommodations and reductions were combined into a single category denoting modification (21). Whether the benefits of adding questions about accommodations and reductions outweigh the increased time and burden of instrument administration will likely depend on the goal of the assessment. One role of functional scales is to serve as disability outcomes in longitudinal studies and clinical trials. Our longitudinal results of the summative scores, with mortality and functional dependence as outcomes, suggest that the NHATS 5-category scale and WHAS II 4-category scale offer little additional benefit over either of the 3-category scales, which were fairly comparable in terms of predictive accuracy. When choosing between the latter two scales, version d may be preferable to version c for two reasons. First, version c, for which vulnerability is defined on the basis of accommodations, reduced frequency, and difficulty (combined), requires many more questions than version d, for which vulnerability is defined solely on the basis of difficulty. Second, version d has been shown to be responsive to clinically meaningful changes in prior longitudinal studies and clinical trials (17,18).

If the objective of the assessment is to stratify risk for subsequent functional dependence, the NHATS 5-category scale might be preferable to either of the 3-category scales since it generates a monotonic gradient in risk and identifies a very low risk group of older persons who are fully able. These advantages, however, were not as apparent when the outcome was mortality. In contrast, the 4-category scale generated monotonic gradients in risk for both mortality and functional dependence, while identifying a very low risk group of older persons who are fully able.

Our study has many strengths. Because participants were drawn from a nationally representative sample, our results should be broadly generalizable to the population of community-living older persons. Second, the amount of missing data on the functional measures was relatively small and attrition for reasons other than death was modest. Third, our results on potential inconsistencies were comparable for two separate waves of data (Rounds 1 and 2). Fourth, our analyses were run separately for men and women and our results demonstrated relatively few sex-specific differences. However, because data on accommodations, reduced frequency, and difficulty were limited to seven self-care and mobility activities, we cannot comment on household (or instrumental) activities, such as shopping, cooking, and paying bills.

In summary, despite inconsistencies in the hierarchy of the NHATS scale, the results of the current study support the use of this 5-category scale over two alternative 3-category scales for stratifying risk of functional dependence and (less so) mortality, although risk was stratified most effectively by a 4-category scale that combined accommodations and reduced frequency into a single category of modification. When assessing changes over time, however, the addition of questions on accommodations and reduced frequency to those on difficulty and dependence, to form a summative disability score, offers little benefit and increases the burden of the assessment. Given the complexity of the disabling process, additional research is needed to further advance the science of functional assessment.

Supplementary Material

Supplementary data are available at The Journals of Gerontology, Series A: Biological Sciences and Medical Sciences online.

Funding

The study was conducted at the Yale Claude D. Pepper Older Americans Independence Center (P30AG21342). T.M.G. is the recipient of an Academic Leadership Award (K07AG043587) from the National Institute on Aging.