Journal of Gerontology: SOCIAL SCIENCES Copyright 2001 by The Gerontological Society of America

2001, Vol. 56B, No. 5, S294–S301

Educational Differences in the Prevalence of Mobility

Disability in Old Age: The Dynamics of Incidence,
Mortality, and Recovery
David Melzer, Grant Izmirlian, Suzanne G. Leveille, and Jack M. Guralnik

Epidemiology, Demography, and Biometry Office, National Institute on Aging, Bethesda, Maryland.

Objectives. Older people with less education have substantially higher prevalence rates of mobility disability. This
study aimed to establish the relative contributions of incidence, recovery rates, and death to prevalence differences in

Downloaded from psychsocgerontology.oxfordjournals.org by guest on January 23, 2011

mobility disability associated with educational status.

Methods. Data were from 3 sites of the Established Populations for Epidemiological Study of the Elderly, covering
8,871 people aged 65–84 years who were followed for up to 7 years. Participants were classified on years of education
received and as disabled if they needed help or were unable to walk up or down stairs or walk half a mile. A Markov
model computed relative risks, adjusting for the effects of repeated observations on the same individuals.

Results. Differences between education groups in person-years lived with disability were large. The relative risk of
incident disability in men with 0–7 years of education (vs. those with 12 or more years) was 1.65 (95% CI 1.37–1.97)
and in women was 1.70 (95% CI 1.15–2.53). Both recovery risks and risks of death in those with disability were not
significantly different across education groups in either gender.

Discussion. Higher incidence of disability is the main contributor to the substantially higher prevalence of disability
in older people of lower socioeconomic status. Efforts to reduce the disparity in disability rates by socioeconomic status
in old age should focus mainly on preventing disability, because differences in the course of mobility disability after on-
set appear to play a limited role in the observed prevalence disparities.

D IFFERENCES in mortality and the prevalence of func-

tional disability in old age across groups with different
levels of education are well established, with higher rates
ing, alcohol abuse, and sedentary lifestyle, may contribute a
proportion of observed socioeconomic differences in mor-
tality (Lantz et al., 1998). Psychosocial factors, including
reported in less educated subgroups in virtually every study stress, appear important in explaining some of the risk not
(Guralnik, Fried, & Salive, 1996). Reducing such socioeco- explained by traditional factors (Baum, Garofalo, & Yali,
nomic health disparities is one of the central goals of the 1999; Landsbergis, Schnall, Warren, Pickering, & Schwartz,
Healthy People 2010 initiative (U.S. Department of Health 1999). In addition, differences in access to health services
and Human Services, 2000). by socioeconomic status have been well documented (An-
Epidemiological analyses of socioeconomic differences drulis, 1998). Finally, reverse causation, with poorer health
in health are broadly based on sociological notions that so- causing a fall in social position, may play a limited role
ciety is stratified into classes or groups, and these groups (Bartley & Plewis, 1997).
have different material circumstances; exposures; behav- Older people often suffer from more than one disease,
iors; and psychosocial, political, and other experiences and disability, usually defined as the inability to carry out
(Lynch & Kaplan, 2000). The traditional individual mea- the usual tasks of daily life, is established as a powerful
sures of education, income, and occupation are seen as indi- measure of health status in old age (Guralnik et al., 1996).
cators of the social and economic factors that dominate the Physical disability is linked to acute illness, chronic disease,
social structure. There is now a large body of evidence and injury. Single diseases making important contributions
based on these markers showing clear trends of poorer to disability include cardiovascular disease, diabetes, arthri-
health with each step down the hierarchy of social position tis, and stroke, but comorbidity is also an important risk for
(Marmot & Wilkinson, 1999). disability. Underlying behavioral risks for disability include
The higher mortality and prevalence of most diseases in low levels of physical activity, few social contacts, and
less privileged socioeconomic groups have been attributed smoking (Stuck et al., 1999)
to a variety of factors (Adler & Ostrove, 1999). Early life The numbers of prevalent cases of disability in a popula-
experiences (including maternal malnutrition) have been tion during any chosen time period is the result of the dy-
implicated in the risk of developing chronic disease in namic balance between the numbers of new cases (disability
adulthood (Barker & Martyn, 1992). Social position may di- incidence), the numbers recovering from disability, and the
rectly result in greater exposure to injury or toxic com- numbers of deaths in those with and without disability. Un-
pounds. Adverse health behaviors, such as cigarette smok- derstanding the relative contribution of each of these factors

S294
EDUCATIONAL DIFFERENCES IN DISABILITY S295

to the observed socioeconomic disability prevalence differ- logical Studies of the Elderly (EPESE). We used Markov
ences could lead to better targeting of efforts to reduce dis- model–based analysis to estimate relative risks, so that we
parities. If most of the excess disability in less privileged could take account of the repeated observations of the same
groups is attributable to greater incidence, then greater ex- study respondents over the follow-up periods.
posure to the underlying causes of disability, less resistance
to disability, or less access to effective preventive services METHODS
are the most important factors. On the other hand, if lower The original three EPESE study populations were included
recovery rates are the main factor, then factors facilitating in these analyses: East Boston, Massachusetts; New Haven,
recovery, including curative and rehabilitation services, Connecticut; and Washington and Iowa counties, Iowa. In
might be more important. Differences in death rates offer East Boston and Iowa, the entire populations aged 65 years
complex potential explanations. The higher prevalence of and older in the identified local communities were eligible,
disability in older people with less education would arise if and response rates were 85% and 80%, respectively. In New
death rates in less educated disabled elderly persons were Haven, a stratified random sample was selected on the basis
lower than in more educated groups. Conversely, higher of gender and residence in private or public housing: The
prevalence in less educated groups would also occur if there response rate was 82% of eligible participants.

Downloaded from psychsocgerontology.oxfordjournals.org by guest on January 23, 2011

were substantially increased death rates in less educated el- Details of the study methods have been published previ-
ders without disability. ously (Cornoni-Huntley et al., 1993). In-home baseline in-
Previous work provides some evidence for higher rates of terviews were conducted between 1981 and 1983 followed
incidence of disability in less privileged subpopulations by seven annual interviews in New Haven and Iowa and six
(Fried & Guralnik, 1997; Stuck et al., 1999). Overall mor- in Boston. Proxy informants were interviewed when partici-
tality rates have been shown to be elevated in those with pants were unable to answer interview questions. Mortality
less education (Feldman, Malone, Kleinman, & Cornoni- was ascertained from obituaries and notification of death by
Huntley, 1989; Lew & Garfinkel, 2000), although an Italian friends and relatives. Death certificates were also obtained
study has suggested that mortality in disabled older people for decedents.
with less education is not raised (Amaducci et al., 1998), Assessment of mobility status was based on responses to
and disability may be the mediator between education and the following questions: “Are you able to walk up and down
mortality, perhaps because of higher severity level of dis- stairs to the second floor without help?” and “Are you able
eases. On the key issue of possible differences in recovery to walk half a mile without help (about 8 blocks)?” Persons
rates in older people by socioeconomic status, little is who responded “no” to either question were classified as
known. having mobility disability. This classification of mobility
A further problem in exploring the dynamics of socioeco- has been shown to have predictive validity for mortality
nomic status differences in disability prevalence is the vari- (Corti, Salive, & Guralnik, 1996). We used responses to
ety of definitions of disability. Definitions based on an in- questions on the number of completed years of education to
ability to perform basic activities of daily living identify a classify people into three groups: 0–7 years, 8–11 years, or
severe form of disability, which is often the end result of a 12 or more years.
progressive disablement process (Ferrucci et al., 1996). Clear trends in the prevalence of mobility disability by
Lower extremity disability is often a precursor to more se- years of education were present at baseline in people aged
vere disability (Dunlop, Hughes, & Manheim, 1997), and 65–84 but not in older persons, and hence the older group
because of its relatively high prevalence and the large dif- was excluded from the analyses. A total of 3,690 men and
ferences in prevalence across educational groups, it pro- 5,618 women aged 65–84 were eligible for this analysis;
vides an important focus for exploring socioeconomic dif- data on education and mobility status at baseline and death
ferences. or mobility status on at least one follow-up interview were
The measurement of socioeconomic differences in older available for 3,554 men (96%) and 5,317 women (95%).
people is also complicated by a number of factors. Most Transitions in functional status were analyzed in two
older people have no current occupation, and the former oc- ways. First, a descriptive analysis treated each full year of
cupations, particularly of many older women, are a poor in- follow-up as a “person-year” of observation. For example,
dicator of their social position. Current incomes may also if data on disability were available on a respondent at base-
not reflect older people’s long-term material circumstances, line and 7 follow-up years, then this respondent contributed
even when these data are obtainable. In addition, both in- 7 separate person-years of observations to the analysis.
come and occupation may have been adversely affected by Where disability or vital status data were missing for the be-
poor health in adult life, rather than the other way around. ginning or end of a study year, that year was excluded from
Years of education was used in our study as a marker of so- the descriptive person-years analysis.
cioeconomic status, because it is closely related to long- In age and sex-specific analyses, four transitions were
term economic position (Smith & Kington, 1997) and is less studied: incidence of mobility disability, recovery from mo-
susceptible to the effects of later health status on employ- bility disability, death in those who had disability at the start
ment and income. of the person-year of observation, and death in those who
In this analysis, we aimed to measure incidence rates of were not disabled at the start of the person-year. Prevalence
mobility disability together with recovery and death rates in of disability was measured as the proportion of observations
data from a large population-based longitudinal study of in which disability was present. Incidence was measured as
older people—the Established Populations for Epidemio- the proportion of those who were nondisabled at the start of
S296 MELZER ET AL.

each 1-year interval (Time 1) who reported being disabled Table 1. Numbers of Respondents and Numbers of Person-Years
at the follow-up interview (Time 2). The recovery rate was of Observation Available for Analysis, by Age Group, Sex, and
the proportion of disabled persons at Time 1 who reported Years of Education
being not disabled at Time 2. Age Group (years)
We conducted analyses to measure education group dif-
ferences in each component of prevalence according to mo- Years of Education 65–69 70–74 75–79 80–84 Total
bility status at baseline. Among the nondisabled, risk of in- Respondents
cident disablement, death, or no change in 1 year was Women
estimated for men and women with 0–7 years and 8–11 0–7 years 260 280 281 231 1,052
8–11 years 693 734 522 350 2,299
years of education, each compared with those with 12 or
12 years 729 605 405 227 1,966
more years, and data are presented within four 5-year age Total 1,682 1,619 1,208 808 5,317
groups. Similarly, in the disabled, likelihood of death, re- Men
covery, or remaining disabled was estimated in the sex and 0–7 years 223 230 205 143 801
education subgroups. 8–11 years 517 476 353 175 1,521
We drew inferences regarding differences in these transi- 12 years 552 380 203 97 1,232

Downloaded from psychsocgerontology.oxfordjournals.org by guest on January 23, 2011

Total 1,292 1,086 761 415 3,554
tion rates using a Markov chain model developed for this
All 2,974 2,705 1,969 1,223 8,871
purpose. Details of this method and examples of its use in
Person-Years of Observation
disability research have been published elsewhere (Beckett
Women
et al., 1996; Mendes de Leon et al., 1995; Muenz & Ruben- 0–7 years 610 1,680 1,733 1,484 5,507
stein, 1985). Briefly, we assumed that each person’s likeli- 8–11 years 1,850 4,575 3,897 2,688 13,010
hood for his or her given series of functional states could be 12 years 2,051 4,276 3,223 1,943 11,493
expressed in terms of annual transition probabilities. Transi- Total 4,511 10,531 8,853 6,115 30,010
tion probabilities were estimated for each group identified Men
by chosen baseline characteristics, but within these groups 0–7 years 580 1,327 1,195 881 3,983
8–11 years 1,316 2,983 2,301 1,415 8,015
the estimated likelihood of change in functional status be- 12 years 1,545 2,906 1,688 783 6,922
tween two interviews was assumed to depend upon the Total 3,441 7,216 5,184 3,079 18,920
functional state at the most recent interview, independent of All 7,952 17,747 14,037 9,194 48,930
the history of preceding functional states.
Note: Data are from Boston, New Haven, and Iowa sites of the Established
We modeled transition probabilities using logistic link Populations for Epidemiological Studies of the Elderly (Cornoni-Huntley et al.,
functions for each of the four dichotomous outcomes: death 1993).
versus survival from the nondisabled state; death versus sur-
vival from the disabled state; and, conditional upon survival
to the next interview, incident disability and recovery. All
participants with disability and mortality information from picture of higher rates of disability in those who had fewer
at least two of the eight interviews were included in the years of education. For example, in men aged 70–74 with
model-based analyses. We treated interviews with missing 0–7 years of education, 20.8% of person-years were lived
disability status flanked by interviews with nonmissing data with disability, compared with 12.8% in those with 12 or
by averaging (under the model) the predicted transition more years of education. Disability rates in women overall
probabilities for all possible paths in between. We ac- were higher, but differences between the education groups
counted for within-person correlation by using a robust in women were similar: The same comparison at age 70–74
variance estimator that treated each person’s data as a clus- was 38.2% and 19.2%.
ter, or primary sampling unit. We made an adjustment for Figure 2 shows incidence rates (as the percentage of ob-
the stratified sampling from the New Haven site by treating served person-years starting free of disability in which mo-
the combined sample using a separate stratum for each of bility disability was reported at 1-year follow-up) by gen-
the other two sites. All analyses were done separately by sex der, age group, and years of education. This figure again
and included main effects of continuous age and education shows differences by years of education within gender, es-
(0–7 years or 8–11 years vs 12 or more years), as well as pecially in the younger age groups. For example, at age 70–
separate intercepts for EPESE site. We tested the Age 74, women with less than 8 years of education had an inci-
Education interaction but found that it did not significantly dence rate of 19.2 per 100 person-years, compared with 8.0
influence incident disability or mortality. in those with 12 or more years of education.
In contrast to incidence rates, rates of recovery showed
RESULTS little consistent difference by years of education (Figure 2)
A total of 8,871 people aged 65–84 were included in these across the age range. Using the same examples again, in the
analyses, of which 5,317 were women and 3,554 were men age group 70–74, women with less than 8 years of education
(40%). Overall, 21% (1,853) of the sample had received 0–7 had a recovery rate of 21.9 cases per 100 person-years, with
years of education, 43% (3,820) had 8–11 years, and 36% a similar rate in those with 12 or more years of education
(3,198) had 12 or more years (see Table 1). (19.9%). In the same age group in men, however, recovery
Figure 1 presents the percentage of person-years lived rates in those with least education appeared slightly higher,
with mobility disability in each age group, including those although in the older age groups there was no evident differ-
already disabled at baseline. The figure shows the familiar ence in rates.
EDUCATIONAL DIFFERENCES IN DISABILITY S297

rates by disability status at the start of the year (Figure 3)

varied, but showed little systematic difference across educa-
tion groups in either those who were disabled at the start of
the studied years or those who were nondisabled.
One limitation of the person-years–based analysis pre-
sented previously is that it treats each person-year of obser-
vation as independent, ignoring the fact that the same indi-
vidual may contribute to up to seven different observations.
Figure 4 presents the overall results of the Markov models
for the four main factors influencing disability prevalence:
incidence, recovery, and death in those with disability and
those without.
The comparison group for all relative risk calculations
was the relevant age and gender subgroup with 12 or more
years of education. In terms of incidence of mobility dis-

Downloaded from psychsocgerontology.oxfordjournals.org by guest on January 23, 2011

ability, risks were significantly raised in both sexes for the
two lower education groups: Men with 0–7 years of educa-
Figure 1. Percentage of person-years lived with mobility disability tion were 1.65 times (95% CI 1.37–1.97) more likely to
by gender, age group, and years of education (n 9,615 years com- develop mobility disability than those with 12 or more years
mencing with disability in 29,500 years observed for women and 3,899 of education. This relative risk for incident disability in
years with disability in 18,681 years observed for men). Data are from
Boston, New Haven, and Iowa sites of the Established Populations for
women was virtually identical: 1.70 (95% CI 1.15–2.53).
Epidemiological Studies of the Elderly (Cornoni-Huntley et al., 1993). In both sexes relative risks of recovery were not signifi-
cantly different by years of education.
In those who were disabled, relative risks of death were
The well-established pattern of higher overall mortality also not significantly different by education. Death rates in
in those with less education was also evident, although ab- those with less education who were not disabled showed
solute numbers of deaths were relatively small, especially complex differences, with odds in the least educated women
for women. For example, deaths occurred in 6.5% (n being significantly lowered and those in the less educated
336) of follow-up years in men with 0–7 years of education, men being higher than in those with 12 or more years of ed-
compared with 4.8% (412) of men with 12 or more years of ucation.
education. For women, these rates were 4.2% (310) in those To understand the influence of age on the transition risks,
with 0–7 years of education, compared with 3.3% (452) for age-specific relative risks (Table 2) are presented for 5-year
those with 12 or more years of education. Person-year death age bands. The incidence and mortality risks that were sig-

Figure 2. Incidence and recovery rates (percentage of person-years) of mobility disability, by gender, age group, and years of education (n 2,736
incident during 19,885 years observed for women and 1,455 incident during 14,782 years in men; n 1,675 recoveries from disability in 9,615 years
observed in women and 867 recoveries during 3,899 years in men). Data are from Boston, New Haven, and Iowa sites of the Established Popula-
tions for Epidemiological Studies of the Elderly (Cornoni-Huntley et al., 1993).
S298 MELZER ET AL.

Downloaded from psychsocgerontology.oxfordjournals.org by guest on January 23, 2011

Figure 3. Deaths in nondisabled and disabled (percentage of person-years observed) by gender, age group, and years of education (n 291
deaths in 19,885 years commencing without disability observed in women and 583 deaths in 14,782 nondisabled years in men; n 738 deaths in
years commencing with disability in women of 9,615 years observed and 593 deaths of 3,899 disabled years in men). Data are from Boston, New
Haven, and Iowa sites of the Established Populations for Epidemiological Studies of the Elderly (Cornoni-Huntley et al., 1993).

nificant are evident, although they tended not to be statisti- Melzer, Izmirlian, and Guralnik (2000). These life table mod-
cally significant at the older ages. Relative risks for recov- els used the four studied transition probabilities applied to a
ery were not significant at any age in men, and for women, theoretical cohort to estimate prevalence of disability at each
recovery rates were significantly lower only in the youngest age. In a model of 1,000 men with 0–7 years of education
and least educated group. Within age groups, among those entering the life table at age 65 with a 15% prevalence of
disabled, relative risks for death with lower education were disability, only 338 would survive to the age of 75, for ex-
not raised in women or men. ample, of whom 92 (27%) would have disability. During their
In women, higher incidence of mobility disability was the 75th year, there would be 33 incident cases of disability, 12
only significant factor available to explain the elevated dis- would recover from disability, 27 disabled men would die,
ability prevalence in less educated groups across the studied and 13 nondisabled men would die.
age range, because lower death rates in the nondisabled (the Substituting the death rates only of men with 12 or more
only other generally significant difference) would tend to years of education into the life table for men with 0–7 years of
reduce rate differences, rather than increase them. education reduced the excess prevalence by a mean of only
We explored the relative importance of the elevated inci- 10% of the difference between the least and most educated
dence and elevated death rates in the nondisabled least edu- male groups, across the years of age in the studied range.
cated men in a multistate life table model, using methods of On the other hand, substitution of the lower incidence rates
Crimmins, Hayward, and Saito (1994) and Leveille, Penninx, only of the most educated men removed the elevation of
disability rates in the least educated (with a mean change of
107% of the difference in prevalence across the age range).

DISCUSSION
Patterns of disability in old age are of considerable impor-
tance, not only because of their impact on the quality of life
of people who find difficulty or need help in doing everyday
tasks, but also because disability is closely related to the
need for health and long-term institutional care (Guralnik et
al., 1996). Markers of socioeconomic status, whether mea-
sured by occupation, income, or years of education, have
been repeatedly shown to be associated with disability prev-
alence (Guralnik, Land, Blazer, Fillenbaum, & Branch,
Figure 4. Relative risks (with 95% confidence intervals) for transi- 1993; House et al., 1994; LaCroix, Guralnik, Berkman,
tions in those with 0–7 and 8–11 versus 12 or more years of education Wallace, & Satterfield, 1993; Stuck et al., 1999). In addi-
(reference line) by gender, based on Markov model. Data are from
Boston, New Haven, and Iowa sites of the Established Populations tion, some studies have also reported higher incidence of
for Epidemiological Studies of the Elderly (Cornoni-Huntley et al., disability in less privileged subgroups (Stuck et al., 1999).
1993). However, in this study we go further in analyzing not only
EDUCATIONAL DIFFERENCES IN DISABILITY S299

Table 2. Relative Risks and 95% Confidence Intervals for Transitions in Those With 0–7 and 8–11 versus 12 or more Years of Education
by Gender for 5-Year Age Groups, Based on Markov Model
Age 65–69 Age 70–74 Age 75–79 Age 80–84

Years of Relative Confidence Relative Confidence Relative Confidence Relative Confidence

Transitions Education Risk Interval Risk Interval Risk Interval Risk Interval
Men
Incidence of disability 0–7 1.91 1.18–3.07* 1.70 1.28–2.27* 1.64 1.21–2.22* 1.33 0.91–1.94
8–11 1.82 1.26–2.61* 1.58 1.25–2.01* 1.25 0.97–1.62 1.13 0.80–1.59
Recovery from disability 0–7 1.26 0.63–2.52 1.47 0.93–2.31 1.07 0.68–1.68 0.75 0.42–1.36
8–11 0.78 0.44–1.38 1.00 0.68–1.48 0.99 0.66–1.48 1.13 0.68–1.91
Death in disabled 0–7 0.39 0.14–1.13 0.78 0.47–1.30 0.87 0.56–1.36 0.86 0.57–1.28
8–11 0.53 0.25–1.10 1.27 0.87–1.86 1.06 0.71–1.57 0.72 0.49–1.04
Death in nondisabled 0–7 2.12 1.22–3.69* 1.46 0.91–2.35 1.13 0.71–1.79 1.74 1.06–2.87*
8–11 1.12 0.66–1.90 1.39 1.00–1.94* 1.39 0.96–2.02 1.15 0.72–1.85
Women

Downloaded from psychsocgerontology.oxfordjournals.org by guest on January 23, 2011

Incidence of disability 0–7 1.91 1.29–2.85* 2.23 1.74–2.87* 1.45 1.12–1.87* 1.38 1.02–1.85*
8–11 1.54 1.14–2.08* 1.39 1.14–1.69* 1.40 1.13–1.73* 1.21 0.96–1.53
Recovery from disability 0–7 0.45 0.26–0.75* 0.96 0.68–1.35 1.07 0.77–1.49 0.93 0.64–1.34
8–11 0.84 0.53–1.33 0.91 0.68–1.21 0.96 0.71–1.30 0.96 0.69–1.33
Death in disabled 0–7 1.26 0.59–2.70 0.86 0.56–1.33 0.89 0.59–1.36 0.94 0.65–1.34
8–11 0.81 0.38–1.76 0.84 0.57–1.22 0.97 0.70–1.34 0.87 0.63–1.22
Death in nondisabled 0–7 1.37 0.53–3.52 0.44 0.19–1.00 0.77 0.40–1.48 0.67 0.34–1.33
8–11 0.60 0.28–1.27 0.94 0.58–1.50 1.09 0.66–1.80 0.69 0.39–1.24

Notes: Data are from Boston, New Haven, and Iowa sites of the Established Populations for Epidemiological Studies of the Elderly (Cornoni-Huntley et al., 1993).
Sample size as in Table 1.
*p .05.

the inflow into the “pool” of prevalent disability in terms of are good markers of lower limb disability, which in turn are
incident cases, but also the outflow in terms of recovery and good predictors of progression of disability, nursing home
mortality. The results clearly show that relative risks of dis- admission, and mortality (Guralnik, Ferrucci, Simonsick,
ability incidence are significantly higher in groups with Salive, & Wallace, 1995). Mobility disability represents an
fewer years of education compared with those with more early stage of disability, and for example, in the Women’s
education, but relative risks of recovery are not significantly Health and Aging study, more than 90% of disabled older
different. people had mobility difficulties (Fried, Bandeen-Roche,
A number of factors need to be considered in evaluating Kasper, & Guralnik, 1999). Therefore, mobility disability is
these results. Years of education has been widely used as a a key form of disability to study, but it is possible that other
marker of socioeconomic status (Smith & Kington, 1997), but measures, for example of severe disability, could produce
it does have some drawbacks. Years of education is treated different results from those we report for mobility disability.
here as a personal variable, although it is possible that effects The Markov model presented has several advantages, in-
of education may be mediated at the household level: The ef- cluding simultaneously taking account of all possible out-
fects of spouses’ education may result in some misclassifica- comes (incidence, recovery, or death); utilizing all the avail-
tion of true socioeconomic status, especially in the study gener- able data, even on those with some missing elements; and
ation of older women (Williams & Collins, 1995). In the dealing with the effect of repeated observations on the same
EPESE data analyzed, data on income were unavailable at individuals. The model is also stratified for differences be-
baseline for 15.5% of cases in the studied age group, and tween study sites, but numbers are too small to yield signifi-
those with missing income data were significantly more likely cant results within each site. The Markov model does make
to have been disabled at baseline, suggesting potentially im- the technical assumption that the transition probabilities are
portant biases in responses. By contrast, education data were dependent only on the immediate state at the time of the in-
missing in only 1.6% of cases. Years of education therefore terview, although this only operates in the model within the
provides the best available marker of socioeconomic status, “risk factor” groups studied, defined by age, sex, and years
not least because of its relative freedom from the effect of poor of education, and should not introduce any obvious bias.
health on economic and employment status over most of the Overall, the Markov model should provide the best possible
lifetime of the older people studied, because education is estimate of the relative risks in which we are interested.
mostly completed early in life. A sex-specific secondary analy- The Markov model results confirm the analysis of the per-
sis indicated that the number of missing interviews was not son-year data in showing that the most significant factor in-
correlated with the level of education, adjusting for number fluencing the prevalence differences in mobility disability
of interviews with or without disability and age, which adds by years of education is incidence of disability. Interest-
some weight to the argument that missing data on education ingly, using the same dataset and similar methods, Leveille
should not have introduced bias. and colleagues (2000) found that incidence was also the most
Mobility disability is measured here by self-report of being important factor in sex differences in those aged less than 90.
unable to walk or climb stairs without help. These measures Rates of recovery from disability are not significantly different
S300 MELZER ET AL.

by years of education, and this may suggest that differences disability and mortality among men and women aged 65 to 84: The Italian
in care received after the onset of disability in different edu- Longitudinal Study on Aging. Journal of Gerontology: Medical Sci-
ences, 53A, M484–M490.
cational group are of limited importance in explaining the Andrulis, D. P. (1998). Access to care is the centerpiece in the elimination of
excess rates of mobility disability in less educated groups. socioeconomic disparities in health. Annals of Internal Medicine, 129,
Overall mortality differences by years of education are well 412–416.
established (U.S. Department of Health and Human Services, Barker, D., & Martyn, C. (1992). The maternal and fetal origins of cardio-
vascular disease. Journal of Epidemiology and Community Health, 46,
1998; Wilkinson & Marmot, 1998) and present in the study 8–11.
data. The results also suggest that the excess mortality rates Bartley, M., & Plewis, I. (1997). Does health-selective mobility account for so-
are partly mediated through the higher rates of disability in cioeconomic differences in health? Evidence from England and Wales,
less educated groups. Death rates in older people with dis- 1971 to 1991. Journal of Health and Social Behavior, 38, 376–386.
ability did not vary significantly by years of education, sup- Baum, A., Garofalo, J. P., & Yali, A. M. (1999). Socioeconomic status and
chronic stress: Does stress account for SES effects on health? Annals of
porting the suggestion that differences in course of illness the New York Academy of Sciences, 896, 131–144.
after the onset of disability are not factors in mortality differ- Beckett, L. A., Brock, D. B., Lemke, J. H., Mendes de Leon, C., Guralnik,
ences by socioeconomic status in old age. The excess risk of J. M., Fillenbaum, G. G., Branch, L. G., Wetle, T. T., & Evans, D. A.
death in less educated men who were not disabled may reflect (1996). Analysis of change in self-reported physical function among

Downloaded from psychsocgerontology.oxfordjournals.org by guest on January 23, 2011

older persons in four population studies. American Journal of Epidemi-
the effects of higher prevalence of cardiovascular and other ology, 143, 766–778.
conditions that can kill suddenly without prior disability. Cornoni-Huntley, J., Ostfeld, A., Taylor, J., Wallace, R. B. D., Berkman, L.,
The multistate life table model for men, however, shows Evans, D. K. F. L. J., Scherr, P., & Korper, S. (1993). Established popula-
that differences in death rates between education groups make tions for epidemiological studies in the elderly: Study design and
a small contribution to prevalence differences overall, com- methodology. Aging—Clinical and Experimental Research, 5, 27–37.
Corti, M. C., Salive, M. E., & Guralnik, J. M. (1996). Serum albumin and
pared with differences in incidence. physical function as predictors of coronary heart disease mortality and
The chief importance of these findings is in targeting efforts incidence in older persons. Journal of Clinical Epidemiology, 49, 519–
to reduce disparities in health, of the sort envisaged in the 526.
Healthy People 2010 initiative (U.S. Department of Health and Crimmins, E. M., Hayward, M. D., & Saito, Y. (1994). Changing mortality
and morbidity rates and the health status and life expectancy of the
Human Services, 2000). Although a great deal is already older population. Demography, 31, 159–175.
known about the risk factors for disability in old age (Stuck Dunlop, D. D., Hughes, S. L., & Manheim, L. M. (1997). Disability in ac-
et al., 1999) and about programs to prevent disability tivities of daily living: Patterns of change and a hierarchy of disability.
(Wagner, 1997), there is a popular tendency to assume that dif- American Journal of Public Health, 87, 378–383.
ferences in care for already disabled elderly persons from less Feldman, J., Malone, J. K., Kleinman, J., & Cornoni-Huntley, J. (1989). Na-
tional trends in educational differentials in mortality. American Journal
privileged socioeconomic groups must account for a sub- of Epidemiology, 129, 919–933.
stantial part of the poorer disability experience of these popu- Ferrucci, L., Guralnik, J. M., Simonsick, E., Salive, M. E., Corti, C., &
lations. This analysis shows, however, that at a population Langlois, J. (1996). Progressive versus catastrophic disability: A longi-
level differences in influences on recovery or survival tudinal view of the disablement process. Journal of Gerontology:
Medical Sciences, 51A, M123–M130.
with disability are of limited importance. It follows, therefore, Fried, L., Bandeen-Roche, K., Kasper, J., & Guralnik, J. (1999). Associa-
that efforts to reduce disability disparities in old age should tar- tion of comorbidity with disability in older women: The Women’s
get prevention, including better management of disabling med- Health and Aging Study. Journal of Clinical Epidemiology, 52, 27–37.
ical conditions before they progress to disability. Fried, L. P., & Guralnik, J. M. (1997). Disability in older adults: Evidence
The most significant influence on the higher prevalence regarding significance, etiology, and risk. Journal of the American Ge-
riatrics Society, 45, 92–100.
rates of mobility disability in older people who had fewer Guralnik, J. M., Ferrucci, L., Simonsick, E. M., Salive, M. E., & Wallace,
years of education is higher incidence rates of disability. Both R. B. (1995). Lower-extremity function in persons over the age of 70
recovery rates and death rates in those with disability do not years as a predictor of subsequent disability. New England Journal of
differ significantly by socioeconomic status. This analysis sug- Medicine, 332, 556–561.
Guralnik, J. M., Fried, L. P., & Salive, M. E. (1996). Disability as a public
gests that programs aiming to reduce socioeconomic disparities health outcome in the aging population. Annual Review of Public
in the prevalence of disability in old age should focus on Health, 17, 25–46.
prevention of disability onset. Guralnik, J. M., Land, K. C., Blazer, D., Fillenbaum, G. G., & Branch, L. G.
(1993). Educational status and active life expectancy among older
Blacks and Whites. New England Journal of Medicine, 329, 110–116.
Acknowledgments
House, J. S., Lepkowski, J. M., Kinney, A. M., Mero, R. P., Kessler, R. C.,
Dr. Melzer was a Harkness Fellow in Health Policy, funded by The & Herzog, A. R. (1994). The social stratification of aging and health.
Commonwealth Fund of New York, at the time this research was con- Journal of Health and Social Behavior, 35, 213–234.
ducted. The views presented here are those of the authors and not neces- LaCroix, A. Z., Guralnik, J. M., Berkman, L. F., Wallace, R. B., & Satter-
sarily those of The Commonwealth Fund, its director, officers, or staff. field, S. (1993). Maintaining mobility in late life. II. Smoking, alcohol
Address correspondence to Dr. David Melzer, Institute of Public Health, consumption, physical activity, and body mass index. American Jour-
Forvie Site, Robinson Way, Cambridge CB2 2SR, United Kingdom. E-mail: nal of Epidemiology, 137, 858–869.
[email protected] Landsbergis, P., Schnall, P., Warren, K., Pickering, T., & Schwartz, J.
(1999). The effect of job strain on ambulatory blood pressure in men:
Does it vary by socioeconomic status? Annals of the New York Acad-
References emy of Sciences, 896, 414–416.
Adler, N. E., & Ostrove, J. M. (1999). Socioeconomic status and health: Lantz, P. M., House, J. S., Lepkowski, J. M., Williams, D. R., Mero, R. P.,
what we know and what we don’t. Annals of the New York Academy of & Chen, J. (1998). Socioeconomic factors, health behaviors, and mor-
Sciences, 896, 3–15. tality: Results from a nationally representative prospective study of U.S.
Amaducci, L., Maggi, S., Langlois, J., Minicuci, N., Baldereschi, M., Di adults. Journal of the American Medical Association, 279, 1703–1708.
Carlo, A., & Grigoletto, F. (1998). Education and the risk of physical Leveille, S. G., Penninx, B. W., Melzer, D., Izmirlian, G., & Guralnik, J. M.
EDUCATIONAL DIFFERENCES IN DISABILITY S301

(2000). Sex differences in the prevalence of mobility disability in old munity-living elderly people: A systematic literature review. Social
age: The dynamics of incidence, recovery, and mortality. Journal of Science and Medicine, 48, 445–469.
Gerontology: Social Sciences, 55B, S41–S50. U.S. Department of Health and Human Services. (1998). Health,
Lew, E., & Garfinkel, L. (2000). Mortality at ages 75 and older in the Can- United States, 1998, with socioeconomic status and health chart-
cer Prevention Study (CPS I). CA: A Cancer Journal for Clinicians, 40, book: The Secretary of the Department of Health and Human
210–224. Services’ annual report on the health status of the nation. Hyatts-
Lynch, G., & Kaplan, G. (2000). Socioeconomic position. In L. Berkman & ville, MD: National Center for Health Statistics.
I. Kawachi (Eds.), Social epidemiology (pp. 13–35). New York: Ox- U.S. Department of Health and Human Services. (2000). Healthy people
ford University Press. 2010 (conference ed.). Washington, DC: Author.
Marmot, M., & Wilkinson, R. (1999). Social determinants of health. Ox- Wagner, E. H. (1997). Preventing decline in function. Evidence from ran-
ford, England: Oxford University Press. domized trials around the world. Western Journal of Medicine, 167,
Mendes de Leon, C., Beckett, L. A., Fillenbaum, G. G., Brock, D. B., Branch, 295–298.
L. G., Evans, D. A., & Berkman, L. F. (1995). Black-White differences in Wilkinson, R., & Marmot, M. (1998). The solid facts: Social determinants of
risk of becoming disabled and recovering from disability in old age: A lon- health. Geneva, Switzerland: World Health Organisation.
gitudinal analysis of two EPESE populations. American Journal of Williams, D., & Collins, C. (1995). U.S. socioeconomic and racial differ-
Epidemiology, 145, 488–497. ences in health: Patterns and explanations. Annual Review of Sociology,
Muenz, L., & Rubenstein, L. (1985). Markov models for covariate depen- 21, 349–386.
dence of binary sequences. Biometrics, 41, 91–101.

Downloaded from psychsocgerontology.oxfordjournals.org by guest on January 23, 2011

Smith, J. P., & Kington, R. (1997). Demographic and economic correlates
of health in old age. Demography, 34, 159–170.
Stuck, A. E., Walthert, J. M., Nikolaus, T., Bula, C. J., Hohmann, C., & Received April 26, 2000
Beck, J. C. (1999). Risk factors for functional status decline in com- Accepted March 20, 2001