Journal of Gerontology: MEDICAL SCIENCES
2004, Vol. 59A, No. 12, 1295–1303
Copyright 2004 by The Gerontological Society of America
Variation in Thresholds for Reporting
Mobility Disability Between National
Population Subgroups and Studies
David Melzer,1 Tzuo-Yun Lan,1 Brian D. M. Tom,2
Dorly J. H. Deeg,3 and Jack M. Guralnik4
1
Department of Public Health and Primary Care, University of Cambridge, U.K.
2
MRC Biostatistics Unit, Institute of Public Health, Cambridge, U.K.
3
Department of Psychiatry and Institute for Research in Extramural Medicine, Vrije Universiteit, Amsterdam, The Netherlands.
4
Laboratory of Epidemiology, Demography, and Biometry, National Institute on Aging, Bethesda, Maryland.
Background. Disability questions require older people to report difficulties with everyday activities, using broad categorical
responses. Relatively little is known about population group differences in the thresholds for reporting difficulty or inability
with medium-distance mobility against tested mobility-related performance. We aimed to estimate the thresholds on tested
performance at which self-reports change from one category to another, across a range of sociodemographic subgroups. We
also aimed to compare reported and tested performance across two national population studies.
Methods. The samples were from the third U.S. National Health and Nutrition Examination Study (NHANES III) and the
Longitudinal Aging Study Amsterdam (LASA). Measures of gait speed, chair stands, and peak expiratory flow rate in both
studies yielded the validated index of mobility-related physical limitations (MOBLI). Latent probit models were used to
estimate cutpoints (thresholds) on the index for reporting difficulty or inability to walk a medium distance.
Results. Thresholds for reporting difficulty or inability were studied by age, sex, race, educational level, and income in
NHANES III. In models adjusting for the other factors, performance thresholds for reporting disability categories varied by
age and income. The younger elderly persons in NHANES III on average reported difficulties or inabilities only when they
reached a more severe level of tested limitation compared with older old persons. A similar pattern exists for those on higher
incomes. For race, differences in threshold were present only for reporting inability, but not difficulty. Significant differences
in thresholds were not present between groups defined by sex or for years of education. Comparisons between the NHANES
and LASA studies show that lower reported mobility difficulty or inability prevalence in the Dutch sample is attributable
both to reporting at higher levels of limitation and to better functioning.
Conclusions. There is evidence of differences in thresholds for reporting mobility disability, especially across age and
income groups in older Americans. Further work is needed to understand the perceptual, attitudinal, or environmental
factors that cause these reporting differences.
A
S THE population ages, accurate monitoring of the
health of older people grows more important. Older
people typically suffer from several diseases simultaneously, and therefore disability, reflecting the impact and
severity of these diseases, provides an important focus for
aging research. Disability is a risk factor for dependence,
institutionalization, and health care utilization (1).
Questions on disability require respondents to report levels
of difficulty (e.g., none, some, much, or unable) without
indicating when these terms should be used. The categorical
responses identify groups with functional decline due, most
commonly, to arthritis, cardiovascular disease, and cognitive
impairment (2). However, the environment also influences
difficulty with everyday tasks: Older people faced with high
ambient temperatures or upward slopes, for example, may
report difficulty walking a quarter of a mile at relatively less
severe levels of physical health problems than those facing
normal temperatures or flat ground. In addition, attitudes
toward reporting difficulty might also influence responses,
with differing thresholds for admitting difficulty. The net
result of these factors is so-called ‘‘response category
cutpoint shifts’’ for self-reported health status (3).
One way of conceptualizing the response categories to
disability questions is to see them as resulting from
a mapping process onto a continuous measure of relevant
aspects of physical health. The transition from reporting ‘‘no
difficulty’’ to ‘‘difficulty’’ occurs at a cutpoint or threshold
on the underlying physical health status measure. If cutpoints
differ systematically across populations, or even across
sociodemographic groups within a population, then the
disability reports are not comparable. A striking example of
shifted thresholds is from Kerala State in India, which has the
lowest mortality and illiteracy rates but consistently has the
highest self-reported morbidity rates (4). For disability
studies, recent analyses by the World Health Organization
have also confirmed that different subpopulations have
significantly different attitudes toward reporting disabilities,
including mobility disability (5,6).
We have recently reported the development of the index of
mobility-related limitations (MOBLI) (7) for epidemiologic
use comparing the physical component of mobility disability
across populations or over time. In logistic regression models,
three measures were identified as being most strongly associated with reported difficulty or inability in walking a
medium distance (a quarter of a mile) in the third U.S. National
Health and Nutrition Examination Study (NHANES III): gait
speed, time to complete five chair stands, and peak expiratory
flow rate. We have since shown that the MOBLI is predictive
of mortality over 4 years in the U.S. Established Populations
for the Epidemiologic Studies of the Elderly Study (8).
1295
1296
MELZER ET AL.
MOBLI also had good sensitivity and specificity in the
Longitudinal Aging Study Amsterdam (LASA) and was
responsive to change over two 3-year periods of follow-up (9).
The MOBLI is thus well suited as a measure of
‘‘mobility-related physical health’’ for studying response
shifts. In this analysis, we aimed to estimate the thresholds
on tested performance at which self-reports change from one
category to another, across a range of sociodemographic
subgroups. We also aimed to compare reported and tested
performance across two national population studies.
METHODS
Study Sample
The NHANES III was a cross-sectional study of the U.S.
population, undertaken between 1988 and 1994 (10,11).
Complex clustered samples of the civilian, noninstitutionalized populations were used. Of 6596 persons aged 60 or
over interviewed in their homes, a total of 5724 took the
further examinations (5302 attended the mobile examination
center, and 422 were examined in their own houses because
they were unable to attend the center). One hundred ninetythree persons whose self-reported walking information was
not available were excluded, and the remaining 5531 were
included in this analysis. Data sets used in this analysis were
from the Public Use Data File (12).
LASA includes a representative sample of The Netherlands (13–15). In brief, it is a longitudinal study of predictors
and consequences of changes in well-being and autonomy
in the older population. Thus far, the LASA study has
conducted three cycles of interviews and performance tests,
approximately 3 years apart. From September 1992 to
September 1993, 3107 persons participated in the LASA
baseline interview. To be comparable with the NHANES III
data set, respondents aged 60 or over, living in the
community, with measured and self-reported mobility
information were included in the analysis. A total of 2115
respondents were eligible for this analysis.
Measurement of Variables
MOBLI.—The MOBLI included measured average gait
speed, time to complete five chair stands, and peak
expiratory flow rate (6). The MOBLI is the calculated
overall (whole NHANES III population) probability of
reporting difficulty (or inability) in walking a quarter mile,
based on the logistic regression models including the three
measures. The MOBLI score varies from 0 to 1: The higher
the score, the higher the probability of mobility-related
limitation and of reporting medium-distance walking
difficulty or inability. (Details of the MOBLI equations
and calculation of scores are available online [16].)
Self-reported mobility disability.—In NHANES III, the
question on difficulty in walking a quarter mile had four
response categories: without, some, much, and unable to do.
In this analysis we dichotomized responses into (1) ‘‘any
difficulty’’ versus none or (2) ‘‘inability’’ versus ability.
In LASA pilot studies (17), questions on ‘‘walking 400
meters’’ and ‘‘walking for 5 minutes’’ were asked. Walking
for 5 minutes at a normal pace (5 km/h) is equivalent to
walking 400 meters or a quarter of a mile, and the questions
gave similar results. Ultimately, the ‘‘walking for 5 minutes’’
question was chosen for LASA, as conveying a clearer
concept to respondents.
Sociodemographic variables that separate subgroups.—
Sociodemographic variables in this analysis include sex,
age, race, and income and educational level. Levels of
income were classified by the poverty income ratio (on the
midpoint of observed family income category as the
numerator and the poverty threshold, the age of the family
reference person, and the calendar year in which the family
was interviewed as the denominator). Tertiles of the
weighted whole poverty income ratio distribution were
used to classify people’s income as low (0.025–1.708),
middle (1.709–3.239), or high (3.240). Levels of education were based on the number of completed years of
education, coded as low (0–7 years), middle (8–11 years), or
high (12 years). Those not answering questions on
educational or income level were coded as missing.
Statistical Analysis
Latent probit model.—Statistically, we assume there is
a latent variable representing an individual’s relevant
physiologic status. With the use of this, we can estimate
and compare difficulty or inability thresholds (cutpoints) on
this scale. In this analysis, the latent variable is measured
using the MOBLI index. To estimate the difficulty or inability
cutpoints, we used the Generalized Linear Latent and Mixed
Models (GLLAMM) Program, developed by Rabe-Hesketh
and colleagues for use in STATA (18,19). In our model, the
binary response yi of person i to the self-reported mobility
question is modeled using a probit model with underlying
response yi*. The latent variable can be described by the
measured mobility level of person i. That is, we assume yi* ;
N(li,1) and li ¼ xib, where xi is the MOBLI score and b is its
associated coefficient. The observed responses yi (0: no
difficulty or no inability; or 1: with difficulty or with inability)
are generated assuming a threshold (response category
cutpoint) model with person-specific thresholds (response
category cutpoint) si. That is, yi ¼ 0 if yi* , si and yi ¼ 1 if yi*
si. The ith person-specific threshold (response category
cutpoint) si is a linear combination of covariates (zi,
individual-level sociodemographic characteristics) and can
be expressed as si ¼ zi9c, where c is the coefficient vector
associated with zi and is estimated by maximum likelihood,
assuming responses are independent across individuals. In
this cutpoint model, the covariates include age, sex, race,
educational level, income level, and national study.
To understand individual effects and the overall adjusted
effect of each variable on people’s threshold for reporting
disability, we built latent probit models for each variable
separately and then together. In each model, the first
subgroup of the variable was fixed as the baseline category
and the estimated coefficient and standard errors for other
subgroups were calculated. The level of statistical significance was set at p , .01.
In the analyses based on each variable, the underlying
mobility function measure was rescaled from 0 to 1, where
0 corresponds to the lowest level of functional limitation (i.e.,
THRESHOLDS FOR MOBILITY DISABILITY BETWEEN STUDIES
no functional limitation) and 1 corresponds to the highest
level of functional limitation. The estimated difficulty or
inability cutpoints were also appropriately rescaled to lie
between 0 and 1. The 95% confidence intervals for these
rescaled cutpoints were also calculated.
Table 1. Numbers and Percentage of Respondents Available for
Analysis, by Basic Characteristics, Self-Reported Walking, and
Physiological Measures (Total N ¼ 5531)
Basic Characteristic
Comparisons of Cutpoints Between U.S. and Dutch Studies
The NHANES and LASA age and sex structures (Table
4) were similar. However, the study sample from The
Netherlands had much less impairment of mobility function
measured either by self-report or by physiologic tests (peak
expiratory flow, five chair stands, average walking speed,
and MOBLI scores) compared with the U.S. sample.
With use of the latent probit models, the difficulty
and inability cutpoints for each study were calculated
N
%
Age (y)
RESULTS
Comparisons of Cutpoints for Different Subgroups
in the NHANES III Study
In the NHANES III sample (Table 1), the proportions of
male and female are nearly equal. Most of the sample were
aged 60–69 (41.7%) and were White (59%). One-third
(34.4%) reported having difficulties in walking a quarter of
a mile, and about one-third of them (706/1901) were unable
to do it.
Table 2 presents the coefficients and standard errors of the
subgroups in each variable in the latent probit model. Males
and females do not have significantly different cutpoints in
reporting difficulty or inability in walking. White people do
have different cutpoints from other groups for reporting
inability but not difficulty. Subgroups by age, income level,
and educational level have significant differences in
cutpoints for both difficulty and inability. For age and
income level, there is a clear gradient for cutpoints, with
younger ages and higher income levels having higher
cutpoints. The direction of the educational level effect is not
clear for either difficulty or inability cutpoints.
The latent probit model coefficients of the subgroups in
each variable (Table 2, Figure 1) show a clear gradient
relationship of cutpoints for subgroups of age and income
level, although differences in difficulty cutpoints between
age groups 70–79 and 80 or over (p ¼ .1482), between
middle- and high-income groups ( p ¼ .0369), and in inability
cutpoints between low- and middle-income groups ( p¼.062)
were not statistically significant. Again, no obvious differences between difficulty or inability cutpoints can be seen by
sex. For race, White Americans had relatively low inability
cutpoints compared with others. By education, the subgroup
of 12 years and over had significantly higher difficulty
cutpoints only compared with the subgroup of 8–11 years
(p , .001), which had significantly lower inability cutpoints than the other two subgroups.
We further put all covariates together in the models to
see the independent effect of each variable on the cutpoints (Table 3). Compared with the results in the singleeffect models, age and income level remained important
independent predictors for both difficulty and inability
cutpoints, and race was significant only for inability
cutpoints. However, educational level was not statistically
significant in the overall model.
1297
60–69
70–79
80
2307
1804
1420
41.7
32.6
25.7
Male
Female
2697
2834
48.8
51.2
3263
1085
1019
164
59.0
19.6
18.4
3.0
2244
1446
1173
668
40.6
26.1
21.2
12.1
1570
1548
2368
45
28.4
28.0
42.8
0.8
3630
1901
65.6
34.4
4825
706
87.2
12.8
Sex
Race
Non-Hispanic white
Non-Hispanic black
Mexican American
Other
Income
Low (0.025–1.708)
Middle (1.709–3.239)
High (3.240–11.889)
Missing
Education (y)
0–7
8–11
12
Missing
Self-reported walking
Difficulty
No
With
Inability
No
With
Physiological measures
Peak expiratory flow (ml/s)
Mean (SD)
Missing
5544.17
802
(2180.72)
14.5
14.19
442
407
(5.36)
8.0
7.4
0.7002
177
417
(0.2311)
3.2
7.5
0.3479
(0.2444)
0.1371
(0.1910)
Five chair stands (s)
Mean (SD)
Unable
Missing
Average walking speed (m/s)
Mean (SD)
Unable
Missing
MOBLI score (difficulty)
Mean (SD)
MOBLI score (inability)
Mean (SD)
Note: SD ¼ standard deviation; MOBLI ¼ mobility-related limitations.
and plotted in Figure 2. The LASA sample had relatively
higher cutpoints (all p , .001 in difficulty and inability) than
the U.S. sample; that is, the threshold for reporting difficulty
or inability was at a more severe level of tested impairment.
These differences remained (all p , .001 in difficulty and
inability) even if only the White participants (n ¼ 3,263 in the
United States and 2,102 in The Netherlands) were included
MELZER ET AL.
1298
Table 2. Unadjusted Results for the Cutpoint Component of the Latent Probit Models, by Sociodemographic
Characteristics and Category of Response to Mobility Disability Questioning.
Coefficient
SE
Z
p Value
70–79
80
Constant
0.1902
0.2624
1.5567
0.0453
0.0500
0.0412
4.20
5.24
,.001
,.001
Female
Constant
0.0320
1.4623
0.0392
0.0391
0.82
.414
0.0576
0.0508
0.1781
1.4511
0.0495
0.0507
0.1125
0.0392
1.16
1.00
1.58
.244
.316
.113
Characteristic
Overall v2
Overall p Value
31.66
,.0001
0.67
.4143
3.86
.2765
27.97
,.0001
27.76*
,.0001*
20.98
.0001
19.80*
.0001*
A. Difficulty
Age (y)
Sex
Race
Non-Hispanic black
Mexican American
Other
Constant
Income
Middle (1.709–3.239)
High (3.240–11.889)
Missing
Constant
0.1515
0.2747
0.1208
1.3328
0.0480
0.0545
0.0611
0.0458
3.15
5.04
1.98
.002
,.001
.048
Education (y)
0.0713
0.1310
0.2016
1.4234
0.0495
0.0468
0.2116
0.0503
1.44
2.80
0.95
.150
.005
.341
70–79
80
Constant
0.1864
0.4658
1.9446
0.0634
0.0627
0.0484
2.94
7.43
.003
,.001
Female
Constant
0.1011
1.8203
0.0499
0.0424
2.03
.043
0.2072
0.3766
0.4596
1.6620
0.0638
0.0729
0.1735
0.0371
3.24
5.17
2.65
.001
,.001
.008
8–11
12
Missing
Constant
B. Inability
Age (y)
Sex
Race
Non-Hispanic black
Mexican American
Other
Constant
Income
Middle (1.709–3.239)
High (3.240–11.889)
Missing
Constant
0.1144
0.4487
0.0461
1.6544
0.0613
0.0790
0.0746
0.0431
1.87
5.68
0.62
56.94
,.0001
4.11
.0426
36.07
,.0001
32.89
,.0001
32.41*
,.0001*
25.43
,.0001
.062
,.001
.536
Education (y)
17.93*
8–11
12
Missing
Constant
0.1806
0.0662
0.6453
1.8027
0.0621
0.0611
0.2255
0.0521
2.91
1.08
2.86
.0001*
.004
.279
.004
Note: SE ¼ standard error.
* Excluding missing data.
or after the models were adjusted for age and sex (data not
shown). Thus, U.S. respondents were more likely to report
difficulty or inability in walking than the Dutch people, for
a given level of measured impairment.
DISCUSSION
Questions to respondents on medium-distance mobility
disability generally require older people to report difficulties
or inability, without defining when these terms should be
THRESHOLDS FOR MOBILITY DISABILITY BETWEEN STUDIES
1299
Figure 1. Diagram showing rescaled estimated mobility index cutpoints (with 95% confidence intervals) for the reporting of difficulty or inability, in different
subgroups (Y axis: index of mobility-related limitations [MOBLI] score; 1 ¼ most severe measured impairments).
used. These categorical responses can be seen as a subjective
mapping onto an underlying ‘‘latent’’ measure of mobilityrelated physical health. In this analysis, we have shown that
there are important variations in thresholds (cutpoints) for
reporting disability categories across different sociodemo-
graphic subgroups and studies. In the U.S. NHANES III
study, against a validated measure of mobility-related
impairments or limitations (the MOBLI), groups defined by
age and income level have significantly different thresholds
for reporting disability categories. In addition, comparing
MELZER ET AL.
1300
Table 3. Adjusted Results for the Cutpoint Component of the Latent Probit Model
Characteristic
Coefficient
SE
Z
p Value
0.1746
0.2400
0.0468
0.0546
3.73
4.40
,.001
,.001
0.0139
0.0400
0.35
.728
0.0625
0.0639
0.1782
0.0546
0.0613
0.1157
1.15
1.04
1.54
.252
.297
.124
Overall v2
Overall p Value
22.53
,.0001
0.12
.7282
3.38
.3362
A. Difficulty
Age (y)
70–79
80
Sex
Female
Race
Non-Hispanic black
Mexican American
Other
Income
Middle (1.709–3.239)
High (3.240–11.889)
Missing
0.1535
0.2599
0.1249
0.0505
0.0602
0.0616
3.04
4.32
2.03
21.05
.0001
20.57*
,.0001*
10.01
.0185
.002
,.001
.043
Education (y)
9.07*
0.0577
0.0888
0.1930
1.3510
0.0546
0.0564
0.2130
0.0722
1.06
1.57
0.91
.291
.115
.365
70–79
80
0.0984
0.3264
0.0662
0.0687
1.49
4.75
.137
,.001
Female
0.1001
0.0518
1.93
.053
0.2360
0.4151
0.5108
0.0714
0.0858
0.1786
3.30
4.84
2.86
.001
,.001
.004
8–11
12
Missing
Constant
.0107
B. Inability
Age (y)
Sex
Race
Non-Hispanic black
Mexican American
Other
Income
Middle (1.709–3.239)
High (3.240–11.889)
Missing
0.1529
0.5207
0.0914
0.0654
0.0873
0.0767
2.34
5.97
1.19
24.90
,.0001
3.74
.0532
30.44
,.0001
35.81
,.0001
35.74*
,.0001*
14.31
.0025
.019
,.001
.234
Education (y)
6.94*
8–11
12
Missing
Constant
0.0548
0.1112
0.5950
1.6475
0.0689
0.0728
0.2284
0.0884
0.79
1.53
2.61
.0311
.427
.127
.009
Note: SE ¼ standard error.
*Excluding missing data.
self-reports in the U.S. study and the LASA sample from The
Netherlands suggests that the very large differences in
reported mobility disability are, in fact, substantially smaller
when differences in thresholds for reporting disability are
accounted for in this cross-study comparison.
In similar analyses of the NHANES III data set, Iburg and
colleagues (5) compared self- and physician reports with
a latent variable based on eight available physical tests,
namely, shoulder external rotation—right and left, hip and
knee flexion—right and left, timed 8-ft walk, timed tandem
stand, and five timed chair stands. The analysis assumed that
this latent variable represented an individual’s true mobility
level. That analysis similarly showed that thresholds for
reporting difficulties occurred at more severe levels of tested
impairment in men compared with women, in non-White
respondents compared with White respondents, and in those
on higher incomes compared with those on lower incomes.
Two limitations of this earlier work have been addressed in
our analyses: namely, the use of an unvalidated statistical
latent variable in contrast to our use of the MOBLI and the
analysis of age and educational level in addition to sex, race,
and income level as grouping variables. A key difference
between the Iburg latent variable and the MOBLI score is
that the latter included expiratory flow rate, which was
THRESHOLDS FOR MOBILITY DISABILITY BETWEEN STUDIES
1301
Table 4. Comparison of Basic Demography, Self-Reported and Objective Measures of Walking
Between the United States and The Netherlands
United States (n ¼ 5531)
Basic Characteristic
The Netherlands (n ¼ 2115)
N
%
N
%
72.44
(8.33)
72.21
(7.27)
v2
t
p Value
1.16
.24
Age (y)
Mean (SD)
Sex
0.04
Male
Female
.84
2697
2834
48.8
51.2
1037
1078
49.0
51.0
1901
706
34.4
12.8
292
76
13.8
3.6
20.87
15.17
,.0001
,.0001
Self-reported walking
With difficulty
With inability
Physiological measures
Peak expiratory flow (ml/sec)
Mean (SD)
Missing
5544.17
802
(2180.72)
14.5
6571.82
27
(2107.35)
1.3
18.12
,.0001
14.19
442
407
(5.36)
8.0
7.4
12.72
174
74
(4.41)
8.2
3.5
11.41
,.0001
0.7002
177
417
(0.2311)
3.2
7.5
0.8126
14
226
(0.2812)
0.7
10.7
15.44
,.0001
0.3479
(0.2444)
0.2393
(0.2207)
18.68
,.0001
0.1371
(0.1910)
0.0755
(0.1286)
16.21
,.0001
Five chair stands (sec)
Mean (SD)
Unable
Missing
Average walking speed (m/sec)
Mean (SD)
Unable
Missing
MOBLI score (difficulty)
Mean (SD)
MOBLI score (inability)
Mean (SD)
Notes: SD ¼ standard deviation; MOBLI ¼ mobility-related limitations.
empirically found to be more closely associated with
medium-distance mobility than the upper-extremity and
balance tests. In addition, Iburg and colleagues (5) did not
explore cross-study comparisons.
NHANES Analyses
A number of limitations in this analysis need to be
considered in assessing the results. First, the NHANES III
study included community-living older people only. One
common hurdle in analyzing performance test data is
missing data, but here we have used the MOBLI, which
includes ‘‘missing’’ as a category in score calculation. In
addition, ‘‘missing’’ categories were included for grouping
variables. The use of the latent probit model, although less
familiar, is conceptually very similar to logistic regression
models: In logistic regression, group differences are
measured by relative odds, whereas in probit models,
thresholds are reported. The traditional probit model is
widely used in calculation of doses required to achieve
target effects (20,21). The use of this statistical technique
should therefore not be problematic.
Our analyses also have a number of strengths. The
NHANES III study is a high-quality national study with
a relatively large sample of older people, combining selfreport and tested performance relevant to medium-distance
mobility. Our use of the MOBLI score as a continuous
measure of the physical component of mobility disability
provides a strong foundation for the analysis, given the
construct validity of the MOBLI, plus the evidence for its
good test characteristics, predictive validity, and sensitivity
to change.
The results of the analysis of NHANES III are clear-cut:
Thresholds for reporting disability categories vary by age
and income level when compared with tested performance
in models adjusted for the other studied factors. Thus,
younger elderly individuals on average report difficulties or
inabilities only when they reach a more severe level of
tested impairment compared with older old individuals. A
similar pattern exists for higher income level. At the same
time, significant differences were not present between
groups defined by sex or for the main categories of
educational level. For race, differences in threshold were
present only for reporting ‘‘inability’’ but not difficulty.
This analysis is also relevant to understanding causes of
disability. For sex, it is well documented that women have
both higher disability prevalence and incidence rates than
men (22,23), even after controlling for age (24), and worse
measured mobility function (25). Our results have shown
that after adjusting for measured mobility function (the
‘‘physiologic’’ component of mobility disability), women
and men report disability with the same thresholds. The sex
difference in disability rate appears, therefore, to reflect
physiologic differences. The lack of significant threshold
differences by educational level similarly suggests that the
1302
MELZER ET AL.
Figure 2. Diagram showing rescaled estimated mobility index cutpoints (with 95% confidence intervals) for the reporting of difficulty or inability, in the United
States and The Netherlands (Y axis: index of mobility-related limitations [MOBLI] score; 1 ¼ most severe measured impairment).
higher prevalence of mobility disability in less educated
groups (26–28) reflects real physical health differences (25).
Clearly, the mechanism for reporting disability across
different age or income groups is somewhat different. With
advancing age, self-reported disability and measured functional limitations both become more common. In our analysis,
the younger old subjects were shown to be relatively less
likely to report mobility disabilities than the older old subjects
with similar measured physical health status. Possible
explanations for this finding include differences in attitudes
toward reporting or differences in environment. How precisely attitudes toward disability reporting are formed in
different age groups and cohorts needs to be further explored.
Evidence from some studies has shown that older people
with higher income level have both better self-reported
(27,28) and measured (25) physical function. Differences in
reporting of disability across income groups evident in our
analysis may perhaps arise from real differences in the help
and facilities available to older people with higher incomes,
or it may reflect only differences in attitudes. Again, further
research is needed.
It is interesting that White Americans are more likely than
non-Whites to report inability. However, results from
previous studies about racial differences in self-reported
disability are inconsistent (29,30). The possible directions of
racial differences are still unclear, especially if socioeconomic status differences are accounted for in analyses (31).
Cross-Study Comparison
Comparing studies of older people is seldom easy, as
detailed differences in design and instruments abound.
Although the measures used in both the LASA and the
NHANES III studies are very similar, there are some
differences that could make comparison across these two
studies difficult to interpret. First, the comparable question on
walking a quarter of a mile in LASA related to ‘‘walking for
five minutes,’’ which was designed to cover roughly the same
medium-distance walk but was found to convey a clearer
concept to Dutch respondents. Reported prevalence rates in
the LASA study for mobility disability were far lower than
those in the U.S. study, and these differences could be
dismissed as uninterpretable because of the differences in the
disability question asked. However, recent evidence from
another cross-country study, when all questions were
standardized, also suggests that Dutch people have a lower
prevalence of self-reported disability than do people in other
countries (32).
By using the MOBLI score in both studies, we can assess
whether part of this difference in reported disability is due to
differences in reporting, and we can also assess how much
of the difference in mobility-related performance remains to
be explained. Our analyses show that the thresholds for
reporting difficulty and inability in the LASA study were
indeed substantially different from those in the U.S. study,
with Dutch respondents reporting disability only at more
severe levels of tested performance. Nevertheless, differences in MOBLI score and in performance on the individual
tests are also present.
Clearly, much more work has to be done to understand
why differences in reporting of disability exist between some
groups but not others in comparison with measured physical
performance. Important differences in, for example, home
and outside environment may play important roles and need
further exploration. In any event, the analyses presented
support the case for the use of relevant and validated testbased measures in epidemiologic studies of the causes of the
physical health component of disability.
THRESHOLDS FOR MOBILITY DISABILITY BETWEEN STUDIES
Conclusion
Traditional disability questions require older people to
report difficulties with everyday activities including mobility, using broad and largely undefined categories. The core
component of disability is physiologic. When it is measured
by a valid continuous measure of mobility-related physical
impairment or limitation and is compared with disability,
there is evidence of differences in thresholds for reporting
mobility disability across age and income groups in
older Americans. Furthermore, comparisons between the
NHANES III and LASA studies suggest that both reporting
thresholds and measured impairment or limitation contribute
to reported differences in medium-distance mobility disability. Further work is needed to understand the causes of
attitude or environmental factors that may contribute to
these reporting differences.
ACKNOWLEDGMENTS
This study was conducted as a part of the Longitudinal Aging Study
Amsterdam (LASA) and was supported by funding from The Netherlands
Ministry of Health, Welfare, and Sports to D. J. H. Deeg and grants from
the U.K. NHS Research and Development Programme to D. Melzer.
The authors thank Prof. Gary King and Dr. Sophia Rabe-Hesketh for
their statistical advice.
Address correspondence to Dr. David Melzer, Department of Public
Health and Primary Care, University of Cambridge, Forvie Site, Robinson
Way, Cambridge CB2 2SR, U.K. E-mail: [email protected]
REFERENCES
1. Fried LP, Guralnik JM. Disability in older adults: evidence regarding significance, etiology, and risk. J Am Geriatr Soc. 1997;45:
92–100.
2. Stuck AE, Walthert JM, Nikolaus T, Bula CJ, Hohmann C, Beck JC.
Risk factors for functional status decline in community-living elderly
people: a systematic literature review. Soc Sci Med. 1999;48:445–469.
3. Murray CJL, Tandon A, Salomon J, Mathers CD, Sadana R. CrossPopulation Comparability of Evidence for Health Policy. Global
Programme on Evidence for Health Policy Series, No. 46. Geneva:
World Health Organization; 2002.
4. Sen A. Health: perception versus observation. Br Med J. 2002;324:
860–861.
5. Iburg K, Salomon J, Tandon A, Murray CJL. Cross-Population
Comparability of Self-Reported and Physician-Assessed Mobility
Levels: Evidence from the Third National Health and Nutrition
Examination Survey. Global Programme on Evidence for Health
Policy Series, No. 14. Geneva: World Health Organization; 2001.
6. Murray CJL, Tandon A, Salomon J, Mathers CD. Enhancing CrossPopulation Comparability of Survey Results. Global Programme on
Evidence for Health Policy Series, No. 35. Geneva: World Health
Organization; 2001.
7. Lan T-Y, Melzer D, Tom BD, Guralnik JM. Performance tests and
disability: developing an objective index of mobility-related limitation
in older populations. J Gerontol Med Sci. 2002;57A:M294–M301.
8. Melzer D, Lan T-Y, Guralnik JM. The predictive validity for mortality
of the index of mobility-related limitation. Result from the EPESE
Study. Age Ageing. 2003;32(6):619–625.
9. Lan T-Y, Deeg DJH, Guralnik JM, Melzer D. Responsiveness of the
index of mobility limitation: comparison with gait speed alone in the
Longitudinal Aging Study Amsterdam. J Gerontol A Biol Sci Med Sci.
2003;58(8):721–727.
10. Burt VL, Harris T. The third National Health and Nutrition
Examination Survey: contributing data on aging and health. The
Gerontologist. 1994;34:486–490.
11. Marwick C. NHANES III health data relevant for aging nation. JAMA.
1997;277:100–102.
12. National Center for Health Statistics, U.S. Department of Health and
Human Services. National Health and Nutrition Examination Survey
III, 1988–1994. CD-ROM Series 11, Nos. 1A & 2A; 1997.
1303
13. Deeg DJH, Westendorp-de Serière M. Autonomy and Well-Being in the
Older Population I. Report from the Longitudinal Aging Study
Amsterdam, 1992–1993. Amsterdam: VU University Press; 1994.
14. Broese van Groenou MI, van Tilburg TG, de Leeuw ED, Liefbroer AC.
Data collection. In Knipscheer CPM, de Jong Gierveld J, van Tilburg
TG, Dykstra PA, eds. Living Arrangements and Social Networks of
Older Adults in the Netherlands. First Results. Amsterdam: VU
University Press; 1995:185–197.
15. Aartsen MJ, Smits CH, van Tilburg T, Knipscheer KC, Deeg DJH.
Activity in older adults: cause or consequence of cognitive
functioning? A longitudinal study on everyday activities and cognitive
performance in older adults. J Gerontol Psychol Sci. 2002;57B:
P153–P162.
16. Epidemiology for Policy Group, Department of Public Health and
Primary Care University of Cambridge. MOBLI Index. Available at
www.phpc.cam.ac.uk/epg/mobli.htm 2002.
17. Smits CH, Deeg DJ, Jonker C. Cognitive and emotional predictors of
disablement in older adults. J Aging Health. 1997;9:204–221.
18. Rabe-Hesketh S, Pickles A, Skrondal A. GLLAMM Manual. Technical Report 2001/01. London: Department of Biostatistics and Computing, Institute of Psychiatry, King’s College, University of London;
2001.
19. Rabe-Hesketh S, Skrondal A. Estimating Chopit Models in GLLAMM.
London: Department of Biostatistics and Computing, Institute of
Psychiatry, King’s College, University of London; 2002.
20. Finney JD. Probit Analysis. Cambridge, U.K.: Cambridge University
Press; 1971.
21. Aldrich JH, Nelson FD. Linear Probability, Logit, and Probit Models.
Quantitative Applications in the Social Sciences; No. 45. Beverly Hills:
Sage Publications; 1984.
22. Strawbridge WJ, Kaplan GA, Camacho T, Cohen RD. The dynamics of
disability and functional change in an elderly cohort: results from the
Alameda County Study. J Am Geriatr Soc. 1992;40:799–806.
23. Leveille SG, Penninx BW, Melzer D, Izmirlian G, Guralnik JM. Sex
differences in the prevalence of mobility disability in old age: the
dynamics of incidence, recovery, and mortality. J Gerontol Soc Sci.
2000;55B:S41–S50.
24. Crimmins EM, Saito Y, Reynolds SL. Further evidence on recent trends
in the prevalence and incidence of disability among older Americans
from two sources: the LSOA and the NHIS. J Gerontol Soc Sci.
1997;52B:S59–S71.
25. Seeman TE, Charpentier PA, Berkman LF, et al. Predicting changes in
physical performance in a high-functioning elderly cohort: MacArthur
Studies of Successful Aging. J Gerontol Med Sci. 1994;49A:M97–M108.
26. Snowdon DA, Ostwald SK, Kane RL. Education, survival, and
independence in elderly Catholic sisters, 1936–1988. Am J Epidemiol.
1989;130:999–1012.
27. Boult C, Kane RL, Louis TA, Boult L, McCaffrey D. Chronic
conditions that lead to functional limitation in the elderly. J Gerontol
Med Sci. 1994;49A:M28–M36.
28. Mor V, Murphy J, Masterson-Allen S, et al. Risk of functional decline
among well elders. J Clin Epidemiol. 1989;42:895–904.
29. Mendes de Leon CF,, Beckett LA, Fillenbaum GG, et al. Black–white
differences in risk of becoming disabled and recovering from disability
in old age: a longitudinal analysis of two EPESE populations. Am
J Epidemiol. 1997;145:488–497.
30. Mendes de Leon CF, Fillenbaum GG, Williams CS, Brock DB, Beckett
LA, Berkman LF. Functional disability among elderly blacks and
whites in two diverse areas: the New Haven and North Carolina
EPESE. Established Populations for the Epidemiologic Studies of the
Elderly. Am J Public Health. 1995;85:994–998.
31. Guralnik JM, Leveille SG. Race, ethnicity, and health outcomes—
unraveling the mediating role of socioeconomic status. Am J Public
Health. 1997;87:728–730.
32. Pluijm S, Deeg DJ. Harmonizing a measure of ADL across six
countries. Presented at the Symposium: European Longitudinal Studies
of Health in Older Age, annual meeting of the Gerontological Society
of America, Chicago, IL, 2002.
Received April 23, 2003
Accepted July 31, 2003