American Journal of Epidemiology
Copyright © 1998 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol. 148, No. 1
Printed in U.S.A.
Predictors of Onset of and Recovery from Mobility Difficulty among Adults
Aged 51-61 Years
Daniel O. Clark, 12 Timothy E. Stump,2 and Fredric D. Wolinsky3
Relative to information on activities of daily living, information regarding the onset of and recovery from
mobility difficulty has been limited. Drawing upon data gathered from 6,376 self-respondents aged 51-61
years at baseline (1992) who were successfully reinterviewed in 1994 as part of the Health and Retirement
Survey, the authors were able to build upon and add to knowledge gained from previous studies of the onset
of and recovery from mobility difficulty. Hierarchical logistic regression was used to separate the direct and
indirect effects of predictors of mobility difficulty onset and recovery at 2-year follow-up. To separate direct
and indirect effects, the authors categorized various predictors as being related to sociodemographic factors,
economic factors, health behavior, chronic disease, or physical impairment, and the categories were sequentially incorporated into a series of equations. The order in which the predictors were incorporated into the
equations followed from a theoretical model of the disability process. In this study of mobility difficulty, the
strongest direct predictors of recovery were having little baseline difficulty and the absence of diabetes
mellitus, lung disease, and frequent pain. The strongest direct predictors of onset were female sex, less
education, low net worth, lack of private health insurance, obesity, and frequent pain. Few indirect predictors
for either onset or recovery were identified. Predictors of recovery were few and differed from predictors of
onset. Further efforts are needed to identify modifiable predictors among females, persons with few economic
resources, and those with frequent pain. Am J Epidemiol 1998; 148:63-71.
aging; chronic disease; disability evaluation; risk factors
Recent research indicates that mobility difficulty is
a primary risk factor for the onset of disability. Studies
have presented considerable evidence that mobility
problems have a strong negative impact on the ability
to perform basic and household activities of daily
living, as well as on perceived health and the incidence
of depression (1-4). This suggests that identification
of predictors of the onset of mobility difficulty represents an important step in the development of effective
disability prevention programs and policies. Thus, our
objective in this analysis was to create separate models
for prediction of the onset of and recovery from mobility difficulty at 2-year follow-up (1994) among the
6,376 self-respondents to the 1992 Health and Retirement Survey.
In the several published reports on predictors of
change in mobility status that do exist (2, 5, 6), socio-
demographic characteristics and the prevalence of
chronic health conditions received considerable attention. Race did not independently predict change in
mobility status among respondents in the Longitudinal
Study on Aging, but greater age and female sex did
(6). Among respondents to the EPESE survey [Established Populations for Epidemiologic Study of the
Elderly], age and female sex also predicted the onset
of mobility limitations (2) and were associated with
greater rates of decline and a lower probability of
recovery (5). Lower income was a significant predictor of mobility loss among both men and women,
while a lower educational level predicted loss among
men only (2). In both the Wolinsky et al. (6) and
Guralnik et al. (2) reports, the presence of diabetes
mellitus, hypertension, stroke, and arthritis at baseline
were predictive of onset of or increase in mobility
difficulty. Predictors unique to one or the other study
included leg pain, dyspnea, heart attack, Alzheimer's
disease, and greater body mass index (weight (kg)/
height2 (m2)).
None of the above studies included behavioral predictors, but in a companion study to that of Guralnik
et al. (2), LaCroix et al. (7) focused specifically on
Received for publication April 17,1997, and in final form January
12, 1998.
1
Department of Medicine, Indiana University Center for Aging
Research, Indianapolis, IN.
2
Regenstrief Institute for Health Care, Indiana University School
of Medicine, Indianapolis, IN.
3
School of Public Health, St. Louis University Health Sciences
Center, St. Louis, MO.
63
64
Clark et al.
behavioral predictors. They found current smoking
among men, a body mass index at the 80th percentile
or higher among women, and low levels of physical
activity among both sexes to be predictors for the
onset of mobility limitations. Relative to no alcohol
consumption, low to moderate alcohol consumption
was protective. The authors reported that the effects of
the behavioral predictors were essentially unchanged
when baseline age, education, and number of chronic
conditions were controlled (7).
The findings of the above studies represent important steps toward the identification of risk factors for
mobility difficulty or loss. We incorporated each of
the predictors identified in these studies into our analysis, and, using data from the Health and Retirement
Survey, we were able to extend the findings of these
studies by incorporating important additional predictor
variables (e.g., comorbidity and physical impairments
(8, 9)). In addition, we modeled difficulty onset and
recovery separately, focused on a slightly younger age
group, and estimated both indirect and direct effects.
Separate models of onset and recovery may be important, given the fact that more than one fourth of persons who report having difficulty with activities of
daily living or limitations in lower body function improve or recover by 2-year follow-up (6, 10). A focus
on a younger age group may also prove valuable,
because primary prevention is often most effective
when the intervention population has low prevalence
but elevated risk. Thus, a focus on persons aged 51-61
years, the majority of whom have not yet experienced
declines in lower body function, may provide important clues for the prevention of mobility difficulty and, ultimately, disability in activities of daily
living.
The identification of both direct and indirect effects
has been noted to be a much overlooked but very
important step in risk factor modeling (11, 12). Because the strongest proximate or direct causes are
often not modifiable (e.g., age, sex), it is of interest to
identify indirect effects, which may be modifiable.
Age, for example, may have an effect on the onset of
mobility difficulty via physical activity level, because
quadriceps muscle activity is negatively correlated
with age. Identifying this indirect effect allows for a
more thorough understanding of predictors and thus
greater potential for intervention (e.g., promotion of
physical activity). In this study, we used the hierarchical regression approach to attempt to identify direct
and indirect effects of sociodemographic factors, economic factors, health behavior, chronic disease, and
physical impairment on mobility difficulty onset and
recovery.
MATERIALS AND METHODS
Data collection
The data for these analyses were obtained from
self-responses to questionnaires administered during
the 1992 (baseline) and 1994 (2-year follow-up)
phases of the Health and Retirement Survey. The Health
and Retirement Survey was funded by the National
Institute of Aging and was coordinated and managed
by the Institute for Survey Research at the University
of Michigan (Ann Arbor, Michigan). The study was
designed to provide in-depth data for the investigation
of analytical and policy issues surrounding retirement;
hence, it focused on persons aged 51-61 years. Over
93 percent of the respondents were interviewed in
person; the remainder were interviewed via telephone.
The survey lasted approximately 1 hour for each respondent. The substantial breadth of data collected on
work experiences, financial status, and health allowed
the creation of more fully specified models than was
previously possible.
To identify age-eligible persons, the Health and
Retirement Survey screened approximately 70,000
households obtained from an area probability sample.
Census tracts containing a high density of African
Americans or Mexican Americans were oversampled
2 : 1 , as were census tracts in Florida. All spouses were
interviewed regardless of age, because of the frequency of dual-earner couples and the influence of
spouses in the retirement decision process. Thus, the
Health and Retirement Survey also includes data on
2,912 persons either younger than 51 years or older
than 61 years. These persons do not constitute a representative sample of persons in those age ranges, and
they were excluded from the analyses presented here.
The analyses were further restricted to baseline selfrespondents, because of the subjective nature of many
of the baseline measures. The baseline proxy response rate was 5 percent, and among baseline
self-respondents, the follow-up proxy response rate
was 2 percent.
The overall baseline response rate was 82 percent,
and comparisons with 1990 US Census data provided
no indication of sample bias (13). Among the 51- to
61-year-old self-respondents interviewed at baseline,
97 percent were reinterviewed in 1994. There were no
significant demographic, economic, or chronic disease
differences between those who were reinterviewed
and those who were not. There were 6,437 selfrespondents aged 51-61 years with complete mobility
data at baseline and follow-up (table 1) and 6,376
self-respondents who had complete data on all measures used in the analyses (tables 2 and 3).
Am J Epidemiol
Vol. 148, No. 1, 1998
Onset of and Recovery from Mobility Difficulty
Measures considered
The sociodemographic characteristics considered
included a continuous measure for age; indicator variables for female sex, marital status, residence in the
Southeast, and non-US birth; and a set of three dummy
variables for ethnicity (African-American, MexicanAmerican, and white; white was the reference category). Economic measures consisted of indicator variables for an income (respondent and spouse combined)
of $10,000 or less in the past year, $10,000 or less in
total net worth, private health insurance, and Medicaid
insurance. The questions regarding net worth covered
real estate, business ownership, pensions, stocks and
bonds, checking accounts, savings accounts, certificates of deposit, and all other sources of income. For
missing information on income and net worth, values
imputed by the Institute for Survey Research were
used. By dichotomizing income and net worth at
$10,000 or less, we minimized the potential implications of the imputation of missing values. Education
was represented by a continuous measure for years of
schooling completed, and an indicator variable identified persons currently working for pay. For the latter
respondents, a nine-point occupational physical demands score (not shown) was created from responses
to three statements: 1) "My job requires lots of physical effort," 2) "My job requires lifting heavy loads,"
and 3) "My job requires stooping, kneeling, or crouching." There were four possible responses, ranging
from "all or almost all of the time" (coded as a 1) to
"none or almost none of the time" (coded as a 4). The
three items formed one factor, with factor loadings all
above 0.75 and Cronbach's alpha at 0.83. This variable was only used in subgroup analyses of persons
who were currently working.
The behavioral predictors studied consisted of alcohol consumption, smoking status, overweight, and
physical inactivity.
Alcohol consumption. The CAGE scale is a fouritem alcohol use scale (14) designed to determine
whether an individual has ever abused alcohol. The
four CAGE items are: 1) "Have you ever felt you
should cut down on your drinking?"; 2) "Have people
ever annoyed you by criticizing your drinking?";
3) "Have you ever felt bad or guilty about drinking?";
and 4) "Have you ever taken a drink first thing in the
morning to steady your nerves (an eye-opener)!". A
score of 2 or more is indicative of alcohol abuse.
Smoking. Markers were created for current smokers and for former smokers who had quit smoking
within the past 10 years. Nonsmokers and those who
had quit more than 10 years previously formed the
reference group.
Am J Epidemiol
Vol. 148, No. 1, 1998
65
Overweight. Respondents at or below 90 percent
of their ideal body mass were considered underweight,
while those at or above 140 percent of their ideal body
mass were considered obese. Those between 91 and
139 percent of their ideal body mass formed the reference group.
Physical activity. A set of three dummy variables
was used to capture physical activity levels. A high
level of physical activity was represented by a selfreport of vigorous physical activity (e.g., swimming,
aerobics, biking) at least once per week. A medium
level of physical activity was represented by a selfreport of moderate physical activity (e.g., gardening)
at least once per week but no weekly vigorous physical
activity. Reports of no weekly vigorous or moderate
physical activity formed the reference category.
Disease was represented by a series of indicator
variables for self-reports of chronic disease. In the
Health and Retirement Survey, it is possible to identify
those who report having been told by a doctor that they
have high blood pressure or hypertension, diabetes or
high blood sugar, cancer (other than skin cancer), or
chronic lung disease (excluding asthma) such as
chronic bronchitis or emphysema. Respondents were
also asked, in a single question, whether they had ever
been told by a doctor that they had angina, coronary
heart disease, congestive heart failure, heart attack, or
another heart problem. In a separate question, respondents were asked whether they had ever been told by
a doctor that they had had a stroke. Finally, respondents were asked whether they had ever had or had
ever been told by a doctor that they had arthritis or
rheumatism, and whether they had seen a doctor for
psychiatric problems in the past year.
Measures for physical impairments captured information on sensory deficits and pain. Vision impairment has important consequences for difficulty and
functional limitation, while the role of hearing impairment appears to be less significant (15, 16). Vision and
hearing were self-rated by the use of glasses, contact
lenses, or hearing aids, with responses dichotomized
as fair/poor eyesight or hearing versus other. Pain was
defined as a positive response to the question, "Are
you often troubled with pain?" Experimental studies
have demonstrated that such self-reports of pain are
reliable (17).
Dependent variable
The lower body mobility scale consisted of reports
of any difficulty in performing three mobility tasks:
1) walking one block, 2) walking several blocks, and
3) climbing one flight of stairs without resting. At
baseline, the questions read, "How difficult is it for
you to ... ?", and the possible responses were "not at
66
Clark et al.
all," "a little," "somewhat," "very," and "don't do." At
follow-up, the questions read, "Do you have any difficulty with ... ?", and the response set consisted of
"yes," "no," and "don't do." We coded the baseline
responses of "a little," "somewhat," and "very" and
the follow-up response of "yes" as l's, indicating
difficulty. Accordingly, possible scores on the threeitem scale ranged from 0 (no difficulty on any item) to
3 (difficulty on every item). Persons who reported that
they did not do the activity (less than 2 percent of the
sample) were not included in our analyses. Exploratory factor analysis models yielded minimum factor
loadings of 0.79; Cronbach's alpha was 0.71 at baseline and 0.72 at follow-up.
Analyses
In hierarchical logistic regression, variables are entered sequentially into successive models, allowing the
estimation of both direct and indirect effects of independent variables. In the analyses described here, we
estimated mobility difficulty onset and recovery in
separate models and sequentially incorporated categories of variables one at a time. We first incorporated
sociodemographic characteristics and then economic
measures, followed by health behavior measures,
chronic disease indicators, and finally physical impairments. In this way, we were able to separate the
indirect effects and direct effects of each of the categories of variables. Model fit was assessed through the
use of the receiver operating characteristic curve and
the Hosmer-Lemeshow goodness-of-fit-statistic (18).
RESULTS
The incidence of recovery from mobility difficulty
was substantially greater than the incidence of onset
(table 1). In fact, 56 percent of the 1,420 respondents
who reported having difficulty at baseline no longer
reported difficulty at follow-up, while slightly more
TABLE 1. Baseline prevalence (proportion) and 2-year
follow-up reports of mobility difficulty onset and recovery
among 6,437 self-respondents to the Health and Retirement
Survey, 1992 and 1994
Baseline
(1992)
Difficulty
No difficulty
Total
Follow-up
(1994)
Difficulty
No
difficulty
Total
0.44
(n = 621)
0.06
(n = 303)
0.94
(n = 4,714)
0.22
(n= 1,420)
0.78
(n = 5,017)
0.14
(n = 924)
0.86
(n = 5,513)
1.00
(n = 6,437)
0.56
(n = 799)
than 6 percent of the 5,017 respondents with no difficulty at baseline reported having difficulty at followup. In addition, 53 percent of those who had recovered
by the time of the 2-year follow-up had recovered
from a self-report of "a little" difficulty on just one
lower body item at baseline (data not shown).
To conserve space, we have not shown comparisons
between the measured characteristics of persons with
and without difficulty at baseline. Because of the large
sample size, most differences were statistically significant. With the exception of sex, however, sociodemographic characteristics did not vary substantially by
baseline difficulty status; 52 percent of respondents
without difficulty were female versus 66 percent of
those with difficulty. On the other hand, several of the
economic indicators varied substantially by difficulty
status. The percentages of respondents with low family income, low net worth, and Medicaid insurance
were over two times higher among those with baseline
mobility difficulty. Similarly, those with difficulty
were twice as likely to be at 140 percent or more of
their ideal body mass and to report low physical activity, and they were 50 percent more likely to currently smoke. Chronic diseases and physical impairments were all 50-150 percent more prevalent among
persons with difficulty.
Adjusted odds ratios from the hierarchical analysis
of difficulty onset are shown in table 2. Model 1
indicated that Mexican Americans were 2.8 times and
African Americans 1.7 times as likely to experience
difficulty onset as were Caucasians, while females
were nearly 1.7 times as likely to experience onset as
males. Each year of age was associated with a 1.06
greater risk of difficulty onset. In a separate model
(data not shown), we tested the assumption of linearity
by replacing the continuous age variable with a set of
five 2-year dummy variables (51-53 years was the
reference age group). Those analyses showed that the
risk of difficulty onset was equivalent for persons aged
51-55 years but greater for persons aged 56-61 years.
Model 2 incorporated economic indicators. The excess risk of mobility difficulty among Mexican Americans and African Americans was an entirely indirect
risk that operated through economic status. The effects
of female sex and age, on the other hand, remained and
did not appear to indirectly affect difficulty onset via
economic status. Each year of education was associated with an 11 percent lower likelihood of onset,
while a net worth of $10,000 or less was associated
with a 96 percent increase in the likelihood of onset.
Persons with private health insurance were 0.56 times
as likely to experience onset as those without it. In
analyses limited to those working for pay (data not
shown), we tested whether physical demands at work
Am J Epidemiol
Vol. 148, No. 1, 1998
Onset of and Recovery from Mobility Difficulty
67
TABLE 2. Adjusted odds ratios obtained from hierarchical logistic regression models of onset of lower
body difficulty among 4,974 self-respondents to the Health and Retirement Survey, 1992 and 1994
Independent
variable
Sociodemographic (actors
Age (years)
Female sex
Married (vs. not married)
Not bom in the United States
Living in the South
Mexican American (vs. white)
African American (vs. white)
Model!
Model 2
Model 3
Model 4
Model 5
1.06**
1.66****
0.72*
0.92
1.02
2.81****
1.70***
1.06**
1.54**
0.84
0.65
0.94
1.13
1.21
1.07***
1.78****
0.87
0.76
0.94
1.07
1.06
1.06**
1.64***
0.89
0.79
0.96
1.18
1.12
1.06**
1.64***
0.86
0.78
0.94
1.16
1.12
0.92
1.96***
0.89****
0.76
0.87
1.90***
0.90****
0.73
0.85
1.86***
0.91*•**
0.63
0.82
1.81**
0.91****
0.61
0.56***
0.71*
0.57***
0.69*
0.55***
0.70*
0.57***
0.72
0.38
0.39
0.42
2.33****
1.64**
1.11
1.43*
2.13****
1.65**
1.07
1.31
2.14****
1.65**
1.10
1.22
0.629**
0.646*
0.649*
1.17
1.18
1.18
1.25
1.27
1.50
1.41
1.51*
1.51**
1.93
1.49
1.26
1.25
1.46
1.26
1.40
1.28
1.78
1.29
Economic factors
Family income below $10,000/year
Net worth below $10,000
Years of education
Medicaid insurance (vs. no Medicaid)
Private health insurance (vs. no private
insurance)
Working for pay
Health behaviors
Body mass indext £90% of ideal body
weighty
Body mass index £140% of ideal body
weighty
Current smoker (vs. never smoker)
Former smoker (vs. never smoker)
CAGE scale§ score Z2
High level of physical activity (vs.
medium level)
Low level of physical activity (vs.
medium level)
Chronic diseases
High blood pressure
Current diabetes mellitus
Cancer
Lung disease
Heart disease
Arthritis
Stroke
Psychiatric problems in past year
Physical impairments
Fair or poor eyesight
Fair or poor hearing
Bothered by pain
Model statistics
Receiver operating characteristic
Hosmer-Lemeshow p value
1.64**
1.05
2.11****
0.63
0.04
0.70
0.84
0.74
0.32
0.75
0.16
0.77
0.32
* p<, 0.05; *• p <, 0.01; *** p <, 0.001; * • • • p <, 0.0001.
t Weight (kg)/height* (m*).
$ Reference group: body mass index of 91-139.
§ See text (14).
were responsible for the modest association between
work status at baseline and difficulty onset. We found
no association between scores on the occupational
physical demands scale and risk of difficulty onset.
The direct effects of model 2 were essentially unAm J Epidemiol
Vol. 148, No. 1, 1998
changed by the incorporation of health behaviors
(model 3). Persons with a baseline body mass index of
140 percent or more of their ideal weight were 2.3
times as likely to experience difficulty onset as those
with a body mass index in the 91st—139th percentile,
68
Clark et al.
and those who smoked at baseline were 1.6 times as
likely to experience onset as nonsmokers. Respondents who scored 2 or more on the CAGE alcohol
abuse scale were 1.4 times as likely to experience
onset as those who scored less than 2, while those who
reported participating in vigorous physical activities at
least once per week at baseline were 0.63 times as
likely to experience onset as those who engaged in no
physical activity.
Although there is some indication that a small portion of the effects of obesity may have operated indirectly via chronic disease, the majority of the effects of
model 3 were unchanged in model 4. Persons who
reported having been diagnosed with arthritis or heart
disease were 1.5 times as likely to experience difficulty onset as those who did not.
Physical impairments were incorporated in model 5.
Persons who reported being often bothered by pain at
baseline were 2.1 times as likely to experience difficulty onset as those who did not, and those reporting
fair or poor eyesight were 1.6 times as likely. The
effects of arthritis and heart disease were no longer
significant with the addition of physical impairments,
and separate analyses (not shown) indicated that this
modest drop was the result of an indirect effect via
pain. The receiver operating characteristic (0.766) and
Hosmer-Lemeshow (p = 0.322) statistics suggested
that this final model had an acceptable fit to the data.
We incorporated a few select interaction terms one
at a time into model 5. Following the findings and
speculations of other authors, we tested for interactions between indicators for arthritis and obesity
(19, 20), arthritis and heart disease (16), obesity and
African-American ethnicity (21), diabetes and AfricanAmerican ethnicity (22), and non-US birth and
Mexican-American ethnicity (23). None of these terms
were statistically significant.
Table 3 shows results from logistic regression analysis for recovery of lower body function. The five
models created were identical to those for difficulty
onset, with the exceptions that no interaction terms
were estimated (there are no existing reports or hypotheses with which to guide tests of interaction terms
for recovery from mobihty difficulty) and an indicator
variable was included in each model for persons who
recovered from a report of "a little" difficulty on just
one item at baseline. (These recoveries were modest
and would be the most likely to be influenced by
instrumentation changes between baseline and followup.) The estimates shown at the top of table 3 under
"Model 1" indicate that persons with "a little" difficulty on just one item had a 3.2 times' greater likelihood of recovery than those with greater levels of
baseline difficulty. In addition, with all variables con-
trolled (i.e., model 5), persons with "a little" difficulty
on just one item at baseline had a 2.3 times' greater
likelihood of recovery.
Although model 1 of table 2 showed that Mexicanand African-American respondents were more likely
to experience difficulty onset, the estimates shown in
model 1 of table 3 do not indicate any effects of
ethnicity on recovery; neither are any sex effects apparent. Age had a modest negative association with
recovery in model 1, and the incorporation of economic indicators in model 2 did not alter that effect.
Each year of education (model 2) was associated with
a 1.06-fold greater likelihood of recovery. Respondents with a net worth of less than $10,000 were 0.69
times as likely to recover as those with a greater net
worth, and those with Medicaid insurance were 0.41
times as likely to recover.
Baseline health behaviors were incorporated in
model 3, and none of them affected the odds of recovery. The associations of age and Medicaid insurance
shown in model 3 were reduced somewhat by the
incorporation of chronic disease in model 4, suggesting that older respondents and those on Medicaid were
less likely to recover in part because of greater comorbidity. Having high blood pressure, diabetes, lung
disease, or arthritis lowered the likelihood of recovery
30-55 percent.
Model 5 indicated that persons who reported at
baseline that they were often bothered by pain were
0.55 times as likely to experience recovery as those
who were not often bothered by pain. As with difficulty onset, model 5 suggested an indirect effect of
arthritis via pain. A CAGE score of 2 or more had a
marginal positive association with the odds of recovery. The model fit statistics indicate an acceptable fit
for this final model (receiver operating characteristic = 0.746; p = 0.693).
DISCUSSION
In these analyses, we attempted to improve understanding of the predictors of mobility difficulty by
creating separate models for onset and recovery and by
utilizing a theoretically guided hierarchical approach
to identify indirect effects. The strongest predictors of
difficulty onset were female sex, less education, lack
of private health insurance, high body mass index, and
frequent pain. Age, low net worth, current smoking,
lack of physical activity, and fair or poor eyesight
were also significant predictors of onset. Predictors of
2-year recovery from difficulty were few, and with the
exception of being bothered by pain, they differed
from predictors of onset in this late-middle-aged
group. The strongest predictors of recovery in the final
model were mild baseline difficulty and an absence of
Am J Epidemiol
Vol. 148, No. 1, 1998
Onset of and Recovery from Mobility Difficulty
TABLE 3. Adjusted odds ratios obtained from hierarchical logistic regression models of recovery from
lower body difficulty among 1,402 self-respondents to the Health and Retirement Survey, 1992 and 1994
Independent
variable
Lower body difficulty at baseline
Self-report of "a little" lower body
difficulty on one item at baseline
Sociodemographic factors
Age (years)
Female sex
Married (vs. not married)
Not born in the United States
Living in the South
Mexican American (vs. white)
African American (vs. white)
Model 1
Model 2
Model 3
Model 4
Model 5
3.16****
2.84****
2.80****
2.53****
2.32****
0.96*
1.00
0.96*
0.99
0.85
1.55
1.28
0.95
0.90
0.96*
1.00
0.84
1.49
1.24
0.89
0.89
0.98
0.97
0.79
1.42
1.27
0.85
0.90
0.97
0.95
0.81
1.42
1.29
0.80
0.87
1.04
0.69*
1.06**
0.41**
1.05
0.69*
1.06*
0.40**
1.04
0.71
1.06*
0.49*
1.03
0.74
1.05*
0.54*
1.35
1.06
1.32
1.06
1.32
0.96
1.33
0.96
0.76
0.68
0.69
0.83
0.77
0.89
1.30
0.91
0.77
0.98
1.41
0.91
0.77
0.99
1.44*
1.27
1.20
1.20
1.00
0.97
0.99
0.70**
0.45****
1.18
0.49****
0.80
0.69**
0.91
0.81
0.69**
0,44****
1.15
0.48****
0.80
0.83
0.92
0.91
1.17
1.34
1.16
0.60
0.76
Economic factors
Family income below $10,000/year
Net worth below $10,000
Years of education
Medicaid insurance (vs. no Medicaid)
Private health insurance (vs. no private
insurance)
Working for pay
Health behaviors
Body mass indexf <90% of ideal body
weighty
Body mass index >140% of ideal body
weight*
Current smoker (vs. never smoker)
Former smoker (vs. never smoker)
CAGE scale§ score £2
High level of physical activity (vs.
medium level)
Low level of physical activity (vs.
medium level)
Chronic diseases
High blood pressure
Current diabetes mellitus
Cancer
Lung disease
Heart disease
Arthritis
Stroke
Psychiatric problems in past year
Physical impairments
Fair or poor eyesight
Fair or poor hearing
Bothered by pain
Model statistics
Receiver operating characteristic
Hosmer-Lemeshow p value
*
t
i
§
Am J Epidemiol
1.04
0.85
0.55****
0.67
0.78
p <, 0.05; ** p <, 0.01; *** p £ 0.001; **** p £ 0.0001.
Weight (kg)/heighP (m*).
Reference group: body mass index of 91-139.
See text (14).
Vol. 148, No. 1, 1998
0.70
0.20
0.70
0.75
0.74
0.58
0.75
0.69
69
70
Clark et al.
diabetes, lung disease, and frequent pain. The presence
of some unique predictors and the apparently considerable differences in the effects of the various economic status indicators suggest that separate theoretical frameworks of onset and recovery may be
necessary. At the very least, further investigation of
the unique effects of the various economic indicators
seems warranted.
Few indirect mechanisms for onset and recovery
were identified. Arthritis operated via pain, and
Mexican- and African-American ethnicity operated
via economic status. The considerable and persistent
direct effects of economic indicators on recovery and
onset are perplexing. With baseline sociodemographic
characteristics, health behaviors, chronic disease, and
physical impairments controlled, respondents with low
educational attainment, low net worth, or no private
insurance were at substantially greater risk of difficulty onset. Persons with Medicaid insurance were
nearly one half as likely to recover as persons without
Medicaid. While health insurance status may change
frequently, net worth and education are relatively stable and are likely to capture a person's economic
status over the course of adulthood. This supports the
view that poor economic status represents a cumulative health risk that accrues over many years.
Females were more likely than males to have mobility difficulty at baseline and to experience onset of
mobility difficulty, but they were no more likely to
recover. This is consistent with the findings of other
studies (24-26) which have shown that females have
poorer functional status scores at baseline and experience greater average rates of decline over periods
ranging from 2 years to 10 years. What is not fully
consistent with existing literature is our finding that
the effects of female sex remained essentially unchanged after the incorporation of economic, health
behavior, chronic disease, and physical impairment
measures. Other investigators have shown that these
greater rates of difficulty among women are in large
part determined by economic indicators. Maddox and
Clark (24) found that income and education accounted
for a majority of the greater mean rates of and 10-year
declines in functional status among women relative to
men aged 58-63 years. Verbrugge (25) and Ross and
Bird (26) had access to data on a wide range of social
and behavioral predictors, and they found that among
adults aged 18 years and over, female sex disadvantages in difficulty, morbidity, and perceived health
were statistically accounted for by these predictors.
There are three potential explanations for the differences in the findings associated with female sex in this
report and those of other researchers (24-26). First,
each study used different measures of mobility diffi-
culty, and direct and indirect predictors for difficulty
are likely to depend on the particular measures employed. In fact, Wolinsky et al. (6) used a mobility
difficulty measure very similar to that used here, and
their results showed that females aged S70 years were
more likely than males to experience decline even
after data were controlled for sociodemographic, economic, and disease indicators. Second, the study by
Verbrugge (25) and Ross and Bird (26) used crosssectional data; the predictors may account for sex
differences at a particular point in time but not account
for differences in prospective risk. This hypothesis,
however, is not consistent with the panel data results
reported by Maddox and Clark (24). Third, Verbrugge
(25) and Ross and Bird (26) had access to information
on psychosocial measures such as stress, sense of
control, and health attitudes, and these psychosocial
factors may be the primary source of sex differences in
difficulty and/or perceived health. Again, however,
this hypothesis is not fully consistent with the findings
of other studies. Thus, the hypothesis that would seem
most consistent with the findings of each of these
studies is that different measures of physical function
and health status have very different predictors.
The primary limitation of these analyses is the modest change in wording between the baseline and
follow-up questions on mobility difficulty. It has been
shown that differences in the wording of questions can
produce quite different rates of difficulty (27). Thus,
we were concerned that the modest change in instrumentation might have biased the results. Fortunately,
several questionnaire modules were tested in baseline
subsamples, and one of those modules (n = 684)
included a "difficulty" question with the same stem
and response set as those used at follow-up (i.e., "Do
you have any difficulty ... ?"/["yes," "no," "don't
do"]). However, "Do you have any difficulty walking?" was the only module question that was similar to
the lower body items included in our scale. Comparison of the distribution of responses to this module
question with the distribution of responses on the
lower body scale item, "How much difficulty do you
have walking several blocks?", yielded an expected
pattern. Ninety-two percent of those who responded
affirmatively to the module question also reported
having at least "a little" difficulty walking several
blocks. Eight-five percent of those who responded
negatively to the module question also reported no
difficulty in walking several blocks. Nonetheless, as
an added safeguard, we also implemented the models
with alternative coding schemes. Persons who reported having "a little" difficulty at baseline, for example, were coded zero along with those who reported
having no difficulty. Results from analyses based on
Am J Epidemiol
Vol. 148, No. 1, 1998
Onset of and Recovery from Mobility Difficulty
this and other alternative coding schemes did not differ
significantly from those shown. Therefore, it is unlikely that the modest change in instrumentation affected these findings.
The identification of predictors of difficulty through
survey data is a necessary but insufficient step toward
the identification of modifiable risk factors. Predictors
were categorized into blocks, and we incorporated
these blocks into the models one at a time in hopes of
isolating the impact of modifiable predictors. This
approach had limited success. Female sex is clearly
immutable, and net worth and education are also essentially immutable. Whether or not the remaining
significant predictors are modifiable depends in large
part on the social and cultural context in which the
predictors exist (28). The perception of pain, for example, might be modifiable through pain management
techniques, but these techniques may not be acceptable in some subcultures. Improvements in insurance
coverage do not appear possible in the current political
context of the United States. Similarly, rates of obesity, smoking, and physical activity, and thus of lung
disease, diabetes, and several other chronic diseases, will
be more modifiable in some contexts than in others.
Mobility difficulty appears to play a very large role
in health and disability in later life, and the older adult
population is projected to increase substantially in the
coming decades. Identification of modifiable predictors and their social contexts will play an important
role in reducing the dependency rates and care costs
associated with such disability in future cohorts of
older adults.
ACKNOWLEDGMENTS
This work was supported in part by National Institutes of
Health grants R29-AG-12987 to Dr. Daniel Clark and R37AG-09692 to Dr. Fredric Wolinsky, and an American Association of Retired Persons Andrus Foundation Grant to
Drs. Clark and Wolinsky.
REFERENCES
1. Guralnik JM, Ferrucci L, Simonsick RM, et al. Lowerextremity function in persons over the age of 70 years as a predictor of subsequent disability. N Engl J Med 1995;332:556-61.
2. Guralnik JM, LaCroix AZ, Abbott RD, et al. Maintaining
mobility in late life. I. Demographic characteristics and
chronic conditions. Am J Epidemiol 1993; 137:845-57.
3. Johnson RJ, Wolinsky FD. The structure of health status
among older adults: disease, disability, functional limitation,
and perceived health. J Health Soc Behav 1993;34:105-21.
4. Stump TE, Clark DO, Johnson RJ, et al. The structure of
health status among Hispanic, African American, and white
older adults. J Gerontol B Psychol Sci Soc Sci 1997;52:4960.
5. Beckett LA, Brock DB, Lemke JH, et al. Analysis of change
in self-reported physical function among older persons in four
population studies. Am J Epidemiol 1996;143:766-78.
Am J Epidemiol
Vol. 148, No. 1, 1998
71
6. Wolinsky FD, Stump TE, Callahan CM, et al. Consistency and
change in functional status among older adults over time. J
Aging Health 1996,8:155-82.
7. LaCroix AZ, Guralnik JM, Berkman LF, et al. Maintaining
mobility in late life. II. Smoking, alcohol consumption, physical activity, and body mass index. Am J Epidemiol 1993;137:
858-69.
8. Wallace RB, Lemke JH. The compression of comorbidity. J
Aging Health 1991;3:237-46.
9. Wilson IB, Cleary PD. Linking clinical variables with healthrelated quality of life: a conceptual model of patient outcomes.
JAMA 1995;273:59-65.
10. Manton KG, Corder LS, Stallard E. Estimates of change in
chronic disability and institutional incidence and prevalence
rates in the U.S. elderly population from the 1982, 1984, and
1989 National Long Term Care Survey. J Gerontol 1993;48:
S153-66.
11. Albrecht GL. Plausible explanations of health behavior
change. Adv Med Soc 1994;4:1-17.
12. Kaplan GA, Haan MN, Cohen RD. Predictors and the study of
prevention in the elderly: methodological issues. In: Wallace
RB, Woolson RF, eds. The Epidemiologic Study of the Elderly. New York, NY: Oxford University Press, 1992.
13. Juster FT. The Health and Retirement Study. Inter-Univ Consort Polit Soc Res Bull 1993;14(2):l-4. [Published by the
Inter-University Consortium for Political and Social Research,
Ann Arbor, ML]
14. Mayfield D, McLeod G, Hall P. The CAGE questionnaire:
validation of a new alcoholism screening instrument. Am J
Psychiatry 1974;131:1121-3.
15. Rudberg MA, Furner SE, Dunn JE, et al. The relationship
of visual and hearing impairments to disability: an analysis
using the Longitudinal Study of Aging. J Gerontol 1993;48:
M261-5.
16. Verbrugge LM, Lepkowski JM, Imanaka Y. Comorbidity and
its impact on disability. Milbank Q 1989;67:450-84.
17. Salovey P, Sieber WJ, Smith AF, et al. Reporting chronic pain
episodes on health surveys. (Vital and health statistics, series
6, no. 6). Hyattsville, MD: National Center for Health Statistics, 1992:1-71. (DHHS publication no. (PHS) 92-108).
18. Hosmer DW Jr, Lemeshow S. Applied logistic regression.
(Wiley series in probability and mathematical statistics). New
York, NY: John Wiley and Sons, Inc, 1989.
19. Hartz AJ, Fischer ME, Bril G, et al. The association of obesity
with joint pain and osteoarthritis in the HANES data.
J Chronic Dis 1986;39:311-19.
20. Verbrugge LM, Gates DM, Ike RW. Risk factors for disability
among U.S. adults with arthritis. J Clin Epidemiol 1991;44:
167-82.
21. Kumanyika S. Obesity in black women. Epidemiol Rev 1987;
9:31-50.
22. Goldschmid MG, Domin WS, Ziemer DC, et al. Diabetes in
urban African-Americans. II. High prevalence of microalbuminuria and nephropathy in African-Americans with diabetes.
Diabetes Care 1995; 18:955-61.
23. Clark DO, Mungai SM, Stump TE, et al. Prevalence and
impact of risk factors for lower body difficulty among Mexican Americans, African Americans, and whites. J Gerontol A
Biol Sci Med Sci 1997;52:M97-105.
24. Maddox GL, Clark DO. Trajectories of functional impairment
in later life. J Health Soc Behav 1992;33:114-25.
25. Verbrugge LM. The twain meet: empirical explanations of sex
differences in health and mortality. J Health Soc Behav 1989;
30:282-304.
26. Ross CE, Bird CE. Sex stratification and health lifestyle:
consequences for men's and women's perceived health.
J Health Soc Behav 1994;35:161-78.
27. Wiener JM, Hanley RJ, Clark R, et al. Measuring the activities
of daily living: comparison across national surveys. J Gerontol
1990;45:S229-37.
28. Link BG, Phelan J. Social conditions as fundamental causes of
disease. J Health Soc Behav 1995;36(suppl):80-94.