An Incomplete Model of Disability: Discrepancies Between Performance-Based and Self-Reported Measures of Functioning

Abstract

Background

Appropriate conceptualization and measurement of disability are critical for population-focused resource allocation and policy development. Self-reported and performance-based measures of functioning have been used to represent disability. Variation in environmental context or self-perception of ability may influence self-reports; however, performance-based measures that attempt to control environmental context may not accurately capture real-world aspects of functioning. This study examined the agreement between self-report and performance-based measures of functioning within 4 domains among older adults.

Methods

Cross-sectional data from the 2021 National Health and Aging Trends Study was used. Self-reported and performance-based measures of functioning were assessed for vision, hearing, mobility, and memory domains. We examined the diagnostic characteristics of performance-based versus self-reported measures using sensitivity, specificity, and receiver operating characteristics curves. Differences in the agreement of these measures across sociodemographic groups were investigated using logistic regression.

Results

Among 2 442 respondents 71 years and older (mean 78.5 ± 5.3, 56% female participants), performance measures of hearing and mobility had high sensitivity (89% and 91%, respectively) and low/moderate specificity (36% and 63%, respectively). The sensitivity and specificity of vision measures were 71%. Memory measures had high specificity (89%) and low sensitivity (28%). Performance-based discrimination ranged from 0.59 (memory) to 0.78 (mobility). Agreement varied across sociodemographic factors.

Conclusions

Performance measures diverge from self-reported functioning among older adults. Discordance may reveal opportunities for environmental intervention where participants’ performance does not capture the full extent of barriers in their daily lives. Additional research is needed to investigate individual and environmental factors which could explain the observed differences.

Measuring functioning and disability within the population is critical for resource allocation, social services, program planning, and policy development. International disability rights movements have long advocated for “nothing about us, without us,” including how disability is defined and identified within the population (1). Using this phrase as a guidepost, self-reported measures of disability identity have been pointed to as the gold standard for disability measurement. However, asking people to self-report disability identity on a questionnaire poses challenges. Structural forces and assumptions about people with disabilities have led to societal stigma and power imbalances that may inhibit prideful disability identity development (2). To overcome these challenges, public health surveillance systems and research studies have a long history of measuring self-reported functioning—typified by questions about ability or difficulty—to capture disability. Questions about self-reported functioning have either focused on underlying physical, cognitive, and sensory domains or on common activities in daily life.

Performance-based measures of functioning are sometimes used in lieu of self-reports. Performance-based measures involve participants attempting to complete standardized tasks and an interviewer scoring their performance or recording test results. Often, performance-based measures have clinically relevant cut-points to identify potential limitations or impairments in body structures that affect functioning. While performance measures provide insight into an older adult’s ability to carry out tasks in standardized conditions, they may not take into account behavioral and environmental components that contribute to functioning in everyday life.

At present, a working definition of disability that has been adopted by many is the World Health Organization (WHO) International Classification of Functioning, Disability and Health (ICF) framework (3). The ICF framework defines disability as an umbrella term capturing the interaction between individual function and the surrounding environment. Therefore, functioning and disability are multifaceted and shaped by biological, social, and environmental factors. The ICF model considers that an individual can perform a task within a controlled, clinical setting; however, this may not translate into the ability to execute that task within their home, neighborhood, or community. Conversely, the ICF also recognizes that an individual may not be able to perform a task within a clinical setting; but may be able to execute the task within their home, neighborhood, or community.

Previous research has found mixed results on the agreement between performance-based and self-reported measures of functioning. Some findings demonstrate high levels of agreement. For example, gait speed accurately discriminated self-reported mobility limitations among older adults (4,5). In a separate study, authors found that cut points for performance-based assessments of hearing and vision accurately identified older adults with self-reported sensory limitations (6). Conversely, other studies have found modest to low levels of agreement between performance-based and self-reported measures of functioning among older adults. Several studies have concluded that there is discordance between performance-based and self-reported mobility limitations (7,8). Furthermore, in a nationally representative sample, only 72% of adults ages 50 and older exhibited concordance between performance-based and self-reported measures of hearing limitations, without factoring in random agreement (9). Similar discordance has been found among studies of self-report and performance-based vision (10,11) and cognition (12,13).

Disagreement within the literature may be driven by differences between domains (eg, vision, hearing, mobility, and cognition), differences in study populations (eg, living in different social, physical, and attitudinal environments), or differences in aspects of functioning captured by the measures (eg, impairments in body function vs activity limitations). While much of the previous research has focused on evaluating the agreement within a single domain, such as mobility, additional research is needed to examine agreement among multiple domains in the same population. Moreover, prior work has not always included a representative sample of older adults in their study population, limiting the external validity of findings. Therefore, additional research is needed to understand and quantify the relationship between performance-based and self-reported functioning measures among older adults. This study examined whether cut-points for performance-based functioning measures accurately identify self-reported functioning among older adults. The current study uniquely contributes to the current literature by examining multiple domains of functioning (ie, vision, hearing, mobility, and memory) and evaluating discordance among a representative sample of adults aged 71 and older in the United States.

Method

Study Sample

The National Health and Aging Trends Study (NHATS) is a national panel study designed to support investigations of late-life disability. NHATS gathers annual information through in-home interviews from a nationally representative sample of Medicare beneficiaries ages 65 and older. The study oversamples Black individuals and older age groups. NHATS interviews are completed within the respondent’s home. Performance-based measures of mobility and memory have been collected annually since the first round of NHATS in 2011 (14,15). In 2021, NHATS added performance-based hearing and vision assessment through a mobile tablet and calibrated headphones (16). Additional details on NHATS study design, assessment techniques, and annual response rates have been previously published (17,18). In this study, we focus on respondents from Round 11 of NHATS aged 71 and older at the time of data collection with a completed sample person interview (n = 3 388). We further limited our analytic sample to self-responding individuals living in the community who were not missing any functioning measures of interest (n = 997 excluded), leaving 2 391 in the final analytical population.

Self-reported Measures

Self-reported measures focus on 4 domains: vision, hearing, mobility, and memory (Table 1). Vision limitations were self-reported through questions about difficulty (with glasses or vision aids, if used) seeing someone across the street, reading newspaper print, or if the respondent was blind. Hearing limitations were self-reported through questions about difficulty (when wearing aids, if used) hearing well enough to converse, use a telephone, or if the respondent was deaf. Mobility limitations were self-reported through questions about the ability to walk at least 3 blocks or climb 10 stairs without devices or assistance from another person. Sensitivity analyses were completed using additional questions about carrying, bending, reaching, and grasping. Memory limitations were self-reported through questions about self-rated memory and how often memory problems interfere with daily activities (17).

Table 1.

NHATS Variables Used to Define Self-report and Performance Based Measures of Functioning Among 2021 Respondents (n = 2 391)

Self-report measures of functioning
Variable	NHATS item	Question text	Variable description
Vision	ss11seewellst	{When {you use/SP uses} glasses or contacts, {do you/does {he/she}}/{Do you/Does SP}} see well enough to recognize someone across the street?	Respondents who report not seeing well enough to recognize someone across the street, not seeing well enough to read newspaper print, or being blind were coded as having vision limitation.
	ss11glrednewp	{When {you use/SP uses} glasses or contacts/When {you use/SP uses} vision aids/When {you use/SP uses} glasses or contacts and vision aids}, {{do you/does {he/she}}/{Do you/Does SP}} see well enough to read newspaper print?
	fl11blind	Flag variable to indicate if a respondent is blind.
Hearing	ss11convwradi	{When {you use/SP uses} a hearing aid, {do you/does {he/she}}/{Do you/Does SP}} hear well enough to carry on a conversation in a room with a radio or TV playing?	Respondents who report not hearing well enough to carry on a conversation in a room with a radio or TV playing, not hearing well enough to use the telephone, or being deaf were coded as having hearing limitation.
	ss11hearphone	{When {you use/SP uses} a hearing aid, {do you/does {he/she}}/{Do you/Does SP}} hear well enough to use the telephone?
	fl11deaf	Flag variable to indicate if a respondent is deaf.
Mobility	pc11walk3blks	[In the last month, {were you/was SP} able to . . . ] walk 3 blocks by {yourself/himself/herself} {and without {your/his/her} {cane/walker/cane or walker}}?	Respondents who report not being able to walk 3 blocks by themselves or up 10 stairs without a mobility aid were coded as having mobility limitation.
	pc11up10stair	[In the last month, {were you/was SP} able to . . . ] walk up 10 stairs by {yourself/himself/herself} {and without {your/his/her} cane}?
Memory	cg11ratememry	First, how would you rate your memory at the present time? Would you say it is excellent, very good, good, fair, or poor?	Respondents who report memory at the present time as fair or poor and indicated that memory problems interfere with their daily activities at least some days of the week were coded as having memory limitation.
	cg11ofmemprob	In the last month, how often did memory problems interfere with your daily activities? Would you say every day, most days, some days, rarely, or never?
Performance based measures of functioning
Vision	vb11ddistance	Distance visual acuity was assessed at 5 feet, and respondents were asked to read 5 letters per screen. A score was calculated using a sum of correct letters identified during the assessment. The variable vb11ddistance was calculated by taking the logarithm of the score.	A binary variable of vision impairment was created by assigning respondents a value of 1 if distance visual acuity was greater than 0.3, near visual acuity was greater than 0.3, or contrast sensitivity was less than or equal to 1.55.
	vb11dnear	Near visual acuity was assessed at usual reading distance, and respondents were asked to read 5 letters per screen. Size of the letters decreased with each successful attempt. A score was calculated by integrating the number of correct letters and reading distance. The variable vb11dnear was calculated by taking the logarithm of the score.
	vb11dcontrast	Contrast sensitivity was assessed at 5 feet, and respondents were asked to read two letters per screen. Letters became lighter with each successive screen. A score was calculated using a sum of correctly identified lettered during the assessment. The variable vb11dcontrast was calculated by taking the logarithm of the score.
Hearing	hb11dbpta	Hearing was assessed using a portable audiometer combined with noise-attenuated headphones. An automated algorithm presents tones to identify the lowest volume at which a respondent can respond to a sound. Pure tone average of the better ear was used to categorize hearing loss.	A binary variable of hearing impairment was created by assigning respondents a value of 1 if pure tone average of the better ear was greater than or equal to 26 dB.
Mobility	r11dnhatssppb	Mobility was assessed using the NHATS expanded SPPB. The expanded battery includes functional tests of balance, walking, repeated chair stands, grip strength, and peak air flow. A summary score was generated based on quartiles of the NHATS sample distribution.	A binary variable of mobility impairment was created by assigning respondents a value of 1 when NHATS expanded SPPB score was less than 8
Memory	cg11dwrdimmrc	Memory was assessed by immediate and delayed 10-word recall. A list of 10 nouns were read to respondents and the respondent is asked to recall as many words as possible, in any order.	A binary variable of memory impairment was created by assigning respondents a value of 1 when respondent’s score was less than or equal to 1.5 standard deviations below the mean.

Note: NHATS = National Health and Aging Trends Study; SPPB = Short Physical Performance Battery.

Performance-based Measures

Performance-based measures also focus on 4 domains: vision, hearing, mobility, and memory (Table 1). Visual function was measured using a series of 3 tests, which assessed distance visual acuity, near visual acuity, and contrast sensitivity (16,19). Respondents read letters from a mobile tablet, with the task becoming more challenging (eg, smaller letters and lighter contrast) as the activity proceeded (16,19). Interviewers stopped the activity once the respondent gave a specified number of incorrect answers or when the respondent completed all tasks within the activity (16,19). A binary indicator of vision impairment was created using the WHO definition and previously established cut-points (20). Unaided hearing function was measured using a portable audiometer with noise-attenuated headphones (16,19). A binary indicator of hearing impairment was created using the WHO 25 dB cut-point based on the pure-tone average of the better ear (3). Mobility was measured using the expanded NHATS Short Physical Performance Battery (SPPB), completed in the respondent’s home (14,21). The expanded SPPB (range 0–12) includes tests of balance, walking, repeated chair stands, grip strength, and peak airflow (14,21). Respondents were classified as having a mobility impairment when the SPPB score was less than 8 (22). Sensitivity analyses were completed using SPPB cut-off values of 6 and 10 (23–25). Cognitive function was measured using a memory test to align with the self-reported memory measure. Interviewers read a list of 10 nouns to respondents, and the respondent recalled as many words as possible in any order. Word recall occurred immediately after the interviewer read them and later in the interview (ie, delayed recall) (17). Respondents with a score less than or equal to 1.5 standard deviations below the mean (score ≤ 3) were classified as having a memory impairment (15). A sensitivity analysis was completed using the classification criteria for dementia developed by NHATS investigators, which differentiates participants with probable, possible, and no dementia (15). Respondents with probable or possible dementia were classified as having a cognitive impairment (Supplementary Table 1).

Analytic Strategy

Descriptive statistics of the study sample were generated using analytic survey weights. Survey weights account for the differential probability of selection into NHATS and nonresponse bias. For purposes of assessing agreement, self-reported measures are treated as the gold standard in this analysis. This decision was informed by prior research on ethnicity, which has revealed that self-identification is one of the most valid measures of group membership (26). Furthermore, self-identification aligns with disability justice goals to “recognize wholeness” of the individual and their life experience (1,27). The agreement of performance-based with self-reported measures was examined using sensitivity, specificity, positive predictive value, negative predictive value, percent agreement, and receiver operating characteristics (ROC) curves. For each domain, both weighted and unweighted estimates were calculated.

Sensitivity, or the ability of a performance-based measure to correctly identify those with a self-reported limitation, is the proportion of respondents identified as having an impairment according to the performance tests and a self-reported limitation divided by all respondents with a self-reported limitation. Specificity, or the ability of a performance-based measure to correctly identify those who do not have a self-reported limitation, is the proportion of respondents identified as not having an impairment in performance-based functioning and no self-reported limitation divided by all respondents without a self-reported limitation. Positive predictive value is the proportion of respondents correctly identified as having an impairment according to the performance test and a self-reported limitation divided by all respondents with a performance-based impairment. Negative predictive value is the proportion of respondents correctly identified as not having a performance-based impairment and no self-reported limitation divided by all respondents without a performance-based impairment. Percent agreement is the proportion of respondents correctly classified divided by all respondents in the analytic sample. ROC curves range from 0.5 to 1.0 and summarize the sensitivity and specificity of each measure in one value. Greater values indicate better classification when using performance-based measures; a value of 0.5 indicates that performance-based measures are no better at classifying disability than chance alone. In a sensitivity analysis, we examined the agreement between self-reported and performance-based hearing measures, excluding respondents who use hearing aids.

In an exploratory analysis, multinomial logistic regression models were used to estimate the log odds of disagreement between self-reported and performance-based limitations for each domain. The primary outcome was categorized into 3 groups, including: (1) agreement, respondents correctly identified with a self-reported limitation and performance-based impairment in functioning and respondents correctly identified as having no self-reported limitation and no performance-based impairment in functioning, (2) false negative, respondents with self-reported limitation classified as having no performance-based impairment, and (3) false positive, respondents without a self-reported limitation who are classified as having a performance-based impairment in functioning. Age, sex, race/ethnicity, education, and marital status were included as independent variables within these analyses. Univariable and multivariable models were used to estimate the association between each sociodemographic factor and the odds of being a false negative or a false positive. Odds ratios and 95% confidence intervals are presented using correctly classified respondents as the reference outcome category. An overall adjusted Wald F-test was used to assess the significance of the global association of risk factors.

Results

A total of 2 391 respondents were included in the final analytic sample (Figure 1). Respondents with a proxy respondent (n = 252), self-respondents living in residential care settings (n = 210), and self-respondents in the community with missing information (n = 535) were excluded.

Figure 1.

Flow chart of analytic study sample from 2021 National Health and Aging Trends Study Respondents (NHATS; n = 2 391).

The mean age in the final analytic sample was 78.5 (SD = 5.3) years; 56% were women, 81% were White, and 7% were Black (Table 2). Regarding self-reported limitations, 6% of participants reported vision, 10% hearing, 26% mobility, and 11% memory limitations. Using performance-based measures, 32% had vision, 66% hearing, 51% mobility, and 13% memory limitations in functioning. Compared to the excluded respondents, the analytic population had a larger proportion in younger age groups, was more likely to be non-Hispanic White, had fewer self-reported limitations, and had fewer objectively measured impairments (Table 2).

Table 2.

Respondent Characteristics for the 2021 NHATS Study Sample (n = 2 391)

	Included (n = 2 391)		Excluded (n = 997)
	n	Weighted %	n	Weighted %	Difference	p Value
Age (mean, SD)		78.5 (5.3)		81.1 (7.4)	2.6	<.01
Age groups						<.01
71–74	348	30.1	93	22.6	7.5
75–79	756	34.2	226	26.9	7.3
80–84	622	20.1	216	20.2	0.1
85–89	428	11.0	200	14.5	3.5
90+	237	4.7	262	15.9	11.2
Sex						.67
Male	1 000	43.9	410	45.0	1.1
Female	1 391	56.1	587	55.0	1.1
Race/ethnicity						<.01
Non-Hispanic White	1 778	81.4	640	71.9	9.5
Non-Hispanic Black	437	7.2	244	9.9	2.7
Hispanic and others*	176	11.4	113	18.2	6.8
Self-report
Vision limitation						<.01
No	2 229	93.7	780	80.4	13.3
Yes	162	6.4	213	19.6	13.2
Hearing limitation						<.01
No	2 145	90.5	754	77.0	13.5
Yes	246	9.5	240	23.0	13.5
Mobility limitation						<.01
No	1 617	74.5	388	49.4	25.1
Yes	774	25.5	606	50.6	25.1
Memory limitation						<.01
No	2 103	89.4	606	83.7	5.7
Yes	288	10.6	136	16.4	5.7
Performance-based
Vision impairment						<.01
No	1 527	68.3	302	56.3	12.0
Yes	864	31.7	362	43.7	12.0
Hearing impairment						.04
No	678	33.8	46	22.3	11.5
Yes	1 713	66.2	243	77.7	11.5
Mobility impairment						<.01
No	947	49.3	111	30.4	18.9
Yes	1 444	50.7	449	69.6	18.9
Memory impairment						<.01
No	1 982	86.8	535	77.5	9.3
Yes	409	13.2	227	22.5	9.3

Note: ^*Hispanic ethnicity includes Mexican American, Chicano, Puerto Rican, Cuban American, and others. Others include American Indian, Asian, Native Hawaiian, Pacific Islander, and others.

Table 3 shows the agreement between performance-based and self-reported measures of functioning. Vision had a 71% sensitivity and 71% specificity. Hearing and mobility had high sensitivity (>85% both) and low specificity (36% and 63%, respectively). Finally, memory had a 28% sensitivity and 89% specificity. ROC areas for vision, hearing, mobility, and memory were 0.70, 0.63, 0.78, and 0.59, respectively. In sensitivity analyses excluding respondents who use hearing aids yielded very similar findings to the full sample results. When changing the cut-off for limitation in mobility from 8 to a value of 6 and 10, we found that the 6 cut-offs yielded the highest ROC area (0.80 with SPPB < 6, 0.78 with SPPB < 8, and 0.67 with SPPB < 10). Comparing performance-based limitations in memory (using the memory test alone) to the NHATS dementia classification algorithm yielded similar results.

Table 3.

Agreement Between Self-reported and Performance Based Functioning Among 2021 NHATS Respondents (n = 2 391)

	Vision	Hearing	Mobility	Memory
Weighted
Sensitivity	70.5%	88.7%	90.5%	28.2%
Specificity	70.9%	36.1%	63.0%	88.6%
Positive predictive value	14.1%	12.7%	45.5%	22.7%
Negative predictive value	97.3%	96.8%	95.1%	91.2%
Percent agreement	70.9%	41.1%	70.0%	82.2%
ROC area (95% CI)	0.70 (0.66, 0.74)	0.63 (0.60, 0.66)	0.78 (0.76, 0.79)	0.59 (0.56, 0.62)
Unweighted
Sensitivity	75.3%	91.1%	93.2%	31.6%
Specificity	66.7%	30.6%	55.3%	84.9%
Positive predictive value	14.1%	13.1%	49.9%	22.2%
Negative predictive value	97.4%	96.8%	94.4%	90.1%
Percent agreement	67.3%	36.8%	67.5%	78.5%
ROC area (95% CI)	0.71 (0.68, 0.74)	0.61 (0.59, 0.63)	0.74 (0.73, 0.76)	0.58 (0.55, 0.61)

Note: ROC = receiver operating characteristics.

In regression analyses (Table 4), we compared the odds of being correctly classified by the performance-based measures for all domains across age, sex, race/ethnicity, educational attainment, and marital status.

Table 4.

Association of Demographic Characteristics with Discrepancies in Self-reported and Performance-based Functioning

	Only self-reported limitations* (false negative)		Only performance-based impairments* (false positive)
	Univariable	Multivariable	Univariable	Multivariable
	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)	p Value†	p Value‡
Vision
Age					<.01	<.01
71–74	—	—	—	—
75–79	0.86 (0.30, 2.44)	0.77 (0.28, 2.09)	1.08 (0.78, 1.49)	1.08 (0.78, 1.50)
80–84	1.20 (0.42, 3.37)	0.88 (0.31, 2.51)	1.58 (1.15, 2.16)	1.49 (1.07, 2.06)
85+	0.43 (0.13, 1.40)	0.27 (0.08, 0.95)	2.65 (1.84, 3.83)	2.40 (1.63, 3.54)
Gender					.36	.03
Male	—	—	—	—
Female	2.21 (0.70, 6.97)	1.53 (0.48, 4.80)	0.92 (0.74, 1.14)	0.72 (0.56, 0.91)
Race/ethnicity					.03	.03
White, non-Hispanic	—	—	—	—
Black, non-Hispanic	0.66 (0.24, 1.86)	0.36 (0.11, 1.12)	1.72 (1.22, 2.43)	1.42 (0.99, 2.05)
Other	1.27 (0.22, 7.46)	1.13 (0.21, 6.01)	1.67 (1.04, 2.68)	1.83 (1.18, 2.84)
Education					<.01	.06
College graduate	—	—	—	—
Some college	3.02 (0.96, 9.50)	2.68 (0.88, 8.20)	1.08 (0.83, 1.41)	1.02 (0.77, 1.35)
HS graduate	2.95 (0.94, 9.28)	2.77 (0.92, 8.36)	1.12 (0.83, 1.51)	0.98 (0.73, 1.34)
Less than HS	6.79 (2.61, 17.67)	6.47 (2.06, 20.36)	2.28 (1.49, 3.48)	1.68 (1.13, 1.51)
Marital status					<.01	<.01
Married	—	—	—	—
Separated/divorced	3.77 (1.76, 8.07)	3.34 (1.48, 7.56)	1.97 (1.36, 2.86)	2.01 (1.32, 3.08)
Widowed	2.78 (1.14, 6.79)	2.39 (0.84, 6.82)	1.89 (1.41, 2.54)	1.68 (1.21, 2.32)
Never married	0.93 (0.11, 7.92)	0.88 (0.09, 8.21)	1.67 (0.96, 2.91)	1.42 (0.83, 2.42)
Hearing
Age					<.01	<.01
71–74	—	—	—	—
75–79	0.89 (0.22, 3.54)	0.87 (0.23, 3.30)	1.28 (0.98, 1.66)	1.32 (1.03, 1.69)
80–84	0.67 (0.19, 2.32)	0.82 (0.24, 2.83)	2.29 (1.78, 2.94)	2.23 (1.76, 2.83)
85+	0.04 (0.005, 0.42)	0.05 (0.005, 0.44)	3.28 (2.42, 4.46)	3.09 (2.24, 4.26)
Gender					.19	.02
Male	—	—	—	—
Female	2.00 (0.56, 7.17)	2.36 (0.61, 9.22)	0.80 (0.61, 1.05)	0.69 (0.52, 0.92)
Race/ethnicity					<.01	<.01
White, non-Hispanic	—	—	—	—
Black, non-Hispanic	1.62 (0.48, 5.51)	1.52 (0.42, 5.49)	0.69 (0.53, 0.89)	0.62 (0.46, 0.83)
Other	5.60 (1.67, 18.74)	6.58 (2.00, 21.69)	0.64 (0.44, 0.94)	0.66 (0.43, 1.01)
Education					.08	.19
College graduate	—	—	—	—
Some college	0.92 (0.18, 4.66)	0.79 (0.16, 3.99)	0.97 (0.76, 1.24)	1.04 (0.81, 1.32)
HS graduate	0.30 (0.08, 1.16)	0.26 (0.07, 1.00)	1.25 (0.93, 1.67)	1.19 (0.87, 1.62)
Less than HS	3.00 (0.63, 14.29)	1.53 (0.36, 6.53)	1.28 (0.88, 1.85)	1.35 (0.96, 1.91)
Marital status					.14	.23
Married	—	—	—	—
Separated/divorced	1.37 (0.31, 6.07)	1.11 (0.24, 5.06)	1.27 (0.89, 1.81)	1.38 (0.95, 2.01)
Widowed	0.86 (0.21, 3.48)	0.74 (0.20, 2.69)	1.36 (1.07, 1.74)	1.21 (0.89, 1.63)
Never married	0.20 (0.02, 1.83)	0.10 (0.01, 1.12)	0.97 (0.48, 1.97)	1.13 (0.53, 2.42)
Mobility
Age					<.01	<.01
71–74	—	—	—	—
75–79	1.22 (0.49, 3.09)	1.19 (0.46, 3.06)	2.13 (1.52, 2.98)	2.14 (1.53, 2.98)
80–84	1.22 (0.40, 3.76)	1.17 (0.37, 3.73)	2.83 (2.07, 3.87)	2.77 (2.01, 3.82)
85+	0.55 (0.13, 2.30)	0.57 (0.13, 2.43)	2.92 (2.24, 3.82)	2.85 (2.10, 3.87)
Gender					.01	<.01
Male	—	—	—	—
Female	1.50 (0.69, 3.27)	1.44 (0.66, 3.11)	0.73 (0.57, 0.93)	0.63 (0.48, 0.82)
Race/ethnicity					.34	.46
White, non-Hispanic	—	—	—	—
Black, non-Hispanic	1.36 (0.60, 3.08)	1.05 (0.43, 2.56)	1.35 (1.01, 1.80)	1.35 (0.99, 1.85)
Other	1.05 (0.22, 4.99)	1.22 (0.20, 7.48)	1.07 (0.70, 1.63)	1.03 (0.69, 1.55)
Education					.43	.83
College graduate	—	—	—	—
Some college	1.68 (0.62, 4.55)	1.66 (0.60, 4.55)	1.00 (0.77, 1.29)	1.08 (0.82, 1.41)
HS graduate	2.10 (0.85, 5.21)	2.16 (0.81, 5.76)	1.06 (0.83, 1.36)	1.01 (0.77, 1.32)
Less than HS	1.54 (0.53, 4.46)	1.42 (0.40, 5.03)	1.26 (0.89, 1.79)	1.13 (0.79, 1.63)
Marital status					.24	.39
Married	—	—	—	—
Separated/divorced	1.72 (0.74, 4.00)	1.53 (0.65, 3.62)	0.92 (0.66, 1.27)	0.97 (0.68, 1.37)
Widowed	0.84 (0.37, 1.88)	0.71 (0.32, 1.60)	1.27 (0.97, 1.67)	1.18 (0.86, 1.62)
Never married	1.99 (0.30, 13.17)	1.96 (0.25, 15.14)	1.63 (0.90, 2.98)	1.68 (0.86, 3.29)
Memory
Age					<.01	<.01
71–74	—	—	—	—
75–79	1.16 (0.66, 2.05)	1.23 (0.67, 2.25)	1.77 (0.90, 3.50)	1.65 (0.87, 3.15)
80–84	0.86 (0.50, 1.47)	0.78 (0.44, 1.38)	3.45 (1.83, 6.51)	3.13 (1.67, 5.87)
85+	1.20 (0.69, 2.10)	1.14 (0.61, 2.15)	5.84 (3.21, 10.62)	5.11 (2.83, 9.22)
Gender					.09	<.01
Male	—	—	—	—
Female	1.09 (0.72, 1.66)	0.96 (0.64, 1.44)	0.75 (0.58, 0.98)	0.55 (0.39, 0.77)
Race/ethnicity					<.01	.04
White, non-Hispanic	—	—	—	—
Black, non-Hispanic	2.06 (1.44, 2.95)	1.58 (1.10, 2.27)	1.57 (1.03, 2.38)	1.32 (0.85, 2.04)
Other	1.75 (1.09, 2.80)	1.15 (0.67, 1.99)	1.55 (0.98, 2.46)	1.40 (0.85, 2.33)
Education					<.01	<.01
College graduate	—	—	—	—
Some college	1.52 (0.79, 2.90)	1.47 (0.77, 2.86)	1.57 (0.93, 2.64)	1.67 (0.95, 2.95)
HS graduate	2.11 (1.18, 3.80)	2.07 (1.10, 3.89)	3.05 (1.91, 4.89)	2.84 (1.64, 4.91)
Less than HS	3.98 (2.11, 7.52)	3.53 (1.88, 6.63)	4.10 (2.58, 6.53)	3.43 (2.03, 5.80)
Marital status					.04	.85
Married	—	—	—	—
Separated/divorced	1.40 (0.72, 2.70)	1.17 (0.59, 2.34)	1.17 (0.71, 1.94)	1.23 (0.70, 2.16)
Widowed	1.32 (0.85, 2.06)	1.21 (0.74, 1.99)	1.79 (1.31, 2.45)	1.33 (0.90, 1.97)
Never married	1.59 (0.76, 3.34)	1.22 (0.49, 3.04)	1.27 (0.62, 2.56)	1.24 (0.59, 2.62)

Notes: Bold values indicate a statistically significant association of <.05; CI = confidence interval; HS = High school.

^*Reference category: participants with agreement between self-report and performance-based measures of functioning (ie, self-report no limitation and no performance-based impairment, self-report limitation, and performance-based impairment).

^†Overall adjusted Wald test for the univariable model.

^‡Overall adjusted Wald test for the multivariable model.

Only Self-reported Limitations (False Negatives)

Adults aged 85 years or older, compared to those aged 71–74 years, had lower odds of only self-reported (but not performance-based) limitations in vision and hearing (eg, OR for aged 85 or older vs 71–74 = 0.27 [95% CI: 0.08, 0.95] for vision). Gender was not significantly associated with this discrepancy type. For race/ethnicity, participants who self-identified as other race/ethnicity compared to non-Hispanic Whites were more likely to be in the false negative group for hearing. Participants who self-identified as non-Hispanic Black compared to non-Hispanic Whites were more likely to be in the false negative group for the memory domain. For the vision and memory domains, the group of participants with the lowest (<high school) versus highest (college graduates) educational attainment were more like have this type of discrepancy. Divorced participants, were more likely than married participants to be in the false negative group for the vision domain.

Only Performance-based Impairments (False Positives)

In the same models, false positives occurred when only a performance-based impairment (but no self-reported limitation) was observed. For all domains across sociodemographic characteristics, being older and male was associated with higher odds of being in the false positive group. Lower educational attainment was associated with higher odds of being in the false positive group for both the vision and memory domains.

Discussion

We observed substantial discordance between performance-based and self-reported functioning. Performance-based impairments in functioning were more prevalent than self-reported limitations for the same domains. Our observation agrees with a hierarchical model of the biological and pathological aging trajectories, suggesting that people maintain functioning in daily life (self-reported) for many years, even in the presence of phenotypic changes that performance-based tests may capture (28). Additionally, self-reported limitations are likely influenced by a broad range of factors, including social and environmental factors and self-perceptions about one’s ability, whereas socioenvironmental factors have less influence on performance-based functioning measures (29). The effect of social and environmental factors on measures capturing body functions compared with questions that allude to specific tasks or activities (walking upstairs, seeing across the room) may account for some of the observed discordance between measures.

This study confirms findings from previous research, which found a wide discordance in the classification between performance-based and self-reported functioning measures (30). Furthermore, we found discordance between performance-based impairments and self-reported limitations in functioning varied by demographic characteristics. Some have hypothesized that health pessimism or optimism could partially explain discordance (10,11). However, to our knowledge, this hypothesis has yet to be formally tested, and additional research is needed to elucidate the pathways through which discordance could arise. An alternative hypothesis that could explain discordance is that people with greater flexible resources (eg, money, power, and prestige) may better navigate and accommodate functioning declines and thus do not experience limitations in daily life (31). Within this paradigm, the discordance between self-reported and performance-based functioning measures could highlight the need for modifiable environmental intervention, such as support or services among individuals or communities.

In contrast to the current study findings, previous research has found performance-based measures of functioning to be reasonable estimators of self-reported limitations (4–6,32). It is possible that these observations are accurate within unique study populations (eg, clinical settings), for a singular disability subgroup (eg, mobility), or within a specific geographic region. Differences in findings between studies could also be explained by variations in data collection procedures. By design, NHATS is more accessible to older adults because all interviews occur within the respondent’s homes. Removal of transportation barriers may lend NHATS to a more heterogeneous study population than studies that rely on clinic visits for enrollment.

Intentional conceptualization, measurement, and interpretation in disability-related research are essential to developing inclusive policies and practices. By selecting self-reported limitations as the frame of reference for assessing performance-based measures we aimed to challenge frameworks that prioritize objective, performance-based measures over subjective measures that reflect an older adult’s lived experiences. Shifting how medical science defines and uses language around disability and function can raise awareness about the critical roles researchers and practitioners can play in remedying inequities that stem from environmental and other barriers in later life (33).

Implications of Findings

The results of this work have important implications for public health and policy. Social support and service programs that rely on performance-based measures of functioning to determine eligibility may be misclassifying large segments of the population. The current study demonstrates that performance-based impairments and self-reported limitations provide complementary information about functioning, and performance-based impairments in functioning alone do not accurately classify perceived limitations among older adults. Furthermore, discordance between performance-based and self-report measures varied by demographic characteristics. Previous research has illuminated the disjointed nature of disability identity and how social programs define disability using limitations in functioning (34). Dorfman found that when social services exclusively depend on performance-based functioning, people with disabilities are often forced to embody a “sick role,” presenting as weak or dependent, which may negatively affect their self-perception and self-worth (34,35). Previous research has highlighted the problematic nature of only using performance-based measures of functioning for social service eligibility, namely excluding people with disabilities from society and endangering their legal disability status (34). Although performance-based measures of functioning provide valuable information over a broad range of functions, using only these measures for eligibility determination may omit important information, particularly for those unable to participate in performance measures. Kasper and colleagues demonstrated, for instance, that performance measures of physical function provided information over a broader range but that self-reports of function better identified those with very low levels of function (36). As previously recommended, social policies should include people with disabilities in the assessment team (34). In addition, the fluidity of disability should be integrated into public policy to allow people to enter and leave disability status throughout their life course (37).

There are also significant clinical implications for these findings. There is growing momentum toward integrating self-report measures into eligibility determination to qualify for medical devices. For example, a previous study suggests that hearing loss self-assessment should be used to qualify for hearing aid device use among older adults (38). Within his analysis, Humes found that only 44% of the variance in self-reported scores was explained by a performance-based measure of hearing (38). Humes argues that older adults would be better served by a system that incorporates auditory wellness assessed by older adults rather than a system embedded in medical diagnoses alone (38). Clinicians should also be aware that discordance between performance-based and self-reported measures of functioning varies by demographic group. Our findings suggest that specific groups of older adults (eg, 85+ years of age) may require different clinical cut points to identify limitations in daily life. Society can shape differential living conditions, surrounding environments, health beliefs, and perceptions which may lead to differences in performance gaps. Considering these differences, complementary information obtained from performance-based and self-reported functioning may be necessary to provide a comprehensive assessment of a patient’s health care needs. Observed differences in performance gaps may highlight the need for intervention on modifiable factors (eg, equipment, therapy, and social services).

Strengths/limitations and Future Research Directions

This study has several strengths. Drawing on a national sample, we evaluated the ability of performance-based functioning measures to estimate self-reports within a nationally representative population-based panel study. Additionally, multiple domains were evaluated to highlight differences in accuracy within vision, hearing, mobility, and memory. Last, the findings were robust to numerous sensitivity analyses using different cut points or definitions to classify functioning. This study also has limitations. Eligible respondents excluded from the study (n = 997; 29%) did not contribute to this analysis and may be systematically different from those we observed within our final analytic sample. Additionally, cut-points can influence the accuracy of performance-based measures in assessing self-reported activities. Notably, there are currently no standards for what constitutes impairment in functioning for contrast sensitivity. However, we completed several sensitivity analyses to assess the impact of our decisions and yielded similar results.

Conclusion

Performance-based measures of functioning were poor predictors of self-reported functioning among older adults. We hypothesize that individual and environmental characteristics may explain observed differences. Additional research is needed to identify the critical factors at play. Discordance between performance-based and self-reported functioning may elucidate barriers and gaps in support needed to facilitate participation in society by older adults.

Funding

Research reported in this publication was supported by the National Institute on Aging of the National Institutes of Health [Award Numbers: K99AG081563, T32AG000247, U01AG032947]. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Conflict of Interest

None declared.

Author Contributions

E. Twardzik conducted the data analysis, interpreted the results, and drafted the initial manuscript. J.A. Schrack supervised the data collection process, provided critical revisions to the manuscript, and ensured the accuracy of the intellectual content. V.A. Freedman conceived and conceptualized the study, designed the research methodology, supervised the data collection process, provided critical revisions to the manuscript, and ensured the accuracy of the intellectual content. N.S. Reed and J.R. Ehrlich provided critical revisions to the manuscript and ensured the accuracy of the intellectual content. P. Martinez-Amezcua supervised the data analysis, drafted the initial manuscript, provided critical revisions to the manuscript, and ensured the accuracy of the intellectual content. All authors reviewed and approved the final version of the manuscript.

Ethics Approval

Informed consent was obtained from all participants. This study was approved by the Johns Hopkins Bloomberg School of Public Health Institutional Review Board.