obesity | VOLUME 19 NUMBER 7 | jULy 2011 1451
nature publishing group
Life expectancy (LE) has been increasing for decades (1), but
whether and to what extent disability-free LE will increase is
not entirely clear (2,3). In the past, advances in medical tech-
nologies and their increased availability have contributed to
the independence of older people (4). However, any optimism
may be diminished by the impact of obesity (5). The worldwide
epidemic has already resulted in a doubling of the prevalence
of obesity in Western and Westernizing countries (6).
Early studies of US populations found large effects of over-
weight and obesity (referred to the summary term “overweight
status” below) on both premature death risk and on disability
prevalence (7,8). However, these effects reflect life histories
of older cohorts and there is evidence that the excess risk of
higher BMI on mortality has diminished over time (9,10).
Studies using more recent data from the United States showed
smaller impacts on LE, but still large effects on disability-free
Results for Europe might be different from those of the
United States though, due to different epidemiological profiles
between these regions. In Southern Europe for example, lower
cardiovascular mortality rates have been recorded than in the
United States, despite the relatively high prevalence of classi-
cal risk factors (13). Consequently, one might hypothesize that
the diminishing effect of high BMI on mortality over time may
vary across the Atlantic.
Unfortunately, the evidence on the impact of overweight
status on LE and the burden of disability in Europe is largely
incomplete. Previous studies were based on small sample sizes
and were restricted to single countries (14,15). Consequently,
a comprehensive picture on the population health associated
with overweight status in Europe is still missing.
The aim of this study was to assess how much the overweight
status is associated with longevity and the burden of disability
in Western-Europe. To measure this relationship, LE and LE
with disability (LwD) were estimated by overweight status and
Life Expectancy and Life Expectancy With
Disability of Normal Weight, Overweight, and
Obese Smokers and Nonsmokers in Europe
Istvan M. Majer1, Wilma J. Nusselder1, Johan P. Mackenbach1 and Anton E. Kunst2
The goal of this study was to estimate life expectancy (LE) and LE with disability (LwD) among normal weight,
overweight, and obese smokers and nonsmokers in Western Europe. Data from four waves (1998–2001) of the
European Community Household Panel (ECHP) were used; a standardized multipurpose annual longitudinal survey.
Self-reported health and socioeconomic information was collected repeatedly using uniform questionnaires for
66,331 individuals in nine countries. Health status was measured in terms of disability in daily activities. Multistate
Markov (MSM) models were applied to obtain hazard ratios (HRs) and age-specific transition rates according
to BMI and smoking status. Multistate life tables were computed using the predicted transition probabilities to
estimate LE and LwD. Significant associations were observed between disability incidence and BMI (HR = 1.15 for
overweight, HR = 1.64 for obese, compared to normal weight). The risk of mortality was negatively associated with
overweight status among disabled (HR = 0.77). Overweight people had higher LE than people with normal-weight
and obesity. Among women, overweight and obese nonsmokers expect 3.6 and 6.1 more years of LwD than normal
weight persons, respectively. In contrast, daily smokers expect lower LE but a similar LwD. The same patterns were
observed among people with high education and those with low education. To conclude, daily smoking is associated
with mortality more than with disability, whereas obesity is associated with disability more than with mortality. The
findings suggest that further tobacco control would contribute to increasing LE, while tackling the obesity epidemic is
necessary to prevent an expansion of disability.
Obesity (2011) 19, 1451–1459. doi:10.1038/oby.2011.46
1Department of Public Health, Erasmus MC, University Medical Center Rotterdam, Rotterdam, The Netherlands; 2Department of Public Health, Academic Medical Centre
(AMC), University of Amsterdam, Amsterdam, The Netherlands. Correspondence: Istvan M. Majer (firstname.lastname@example.org)
Received 13 July 2010; accepted 25 January 2011; published online 17 March 2011. doi:10.1038/oby.2011.46
VOLUME 19 NUMBER 7 | jULy 2011 | www.obesityjournal.org
smoking status. Furthermore, it was investigated in subgroups
of men and women as well as of low and high educated.
LE and LwD are both aggregate measures of population
health referring to a certain period of time. LwD has simi-
lar interpretation to (total) LE but it refers to the number of
years that people expect to live with disability. LE and LwD
(and their difference, “disability-free LE”) are routinely used
aggregate indicators of population health for a certain calendar
The European Community Household Panel (ECHP) was
used as data source. The main advantages of ECHP were that it
provided comparable information on health and mortality for
several European countries, and that it gave sufficient statisti-
cal power to obtain precise estimates for disability incidence
and recovery rates, and for life table calculations for specific
risk factor groups.
Methods and Procedures
The ECHP is a standardized multipurpose annual longitudinal social
survey carried out at the level of the European Union between 1994 and
2001. It is centrally designed and coordinated by the Statistical Office
of the European Communities (Eurostat), and includes demographics,
labor force behavior, income, health, education, training, housing, and
migration. The data were collected by the National Statistical Institutes
or research centers of the participating countries using uniform ran-
dom sampling design and common blueprint questionnaires. Although
Eurostat left the National Institutes of Statistics free to organize the data
collection, and national reporting on the survey organization is lacking
(making assessment of data quality difficult), data checks, imputation,
and weighting were done centrally to maximize data comparability. The
ECHP is intended to be both cross-sectionally and longitudinally repre-
sentative with respect to the national household populations. For more
information about the design and the data procedures of the ECHP we
refer to an extensive review of Peracchi (17).
Data from the fifth wave of 1998 onward until 2001 were used for the
current study, as 1998 was the first year that height, weight, and smoking
status were asked from the participants. Countries for which mortal-
ity data (the Netherlands), disability data (Luxembourg), or BMI data
(France) was not available were omitted from the study. Similarly, data
from Germany and the UK had to be left out because the original ECHP
data were replaced by data from national surveys (German Social Eco-
nomic Panel and British Household Panel Survey) which did not contain
information on smoking and BMI.
Information on nonresponse rates at baseline and cumulative retention
percentages, i.e., the cumulative percentages of individuals retained until
the fourth wave of the panel are presented in Supplementary Table S1
online. There were differences between countries, with high retention
percentages in Italy (83%) and Portugal (87%), whereas samples in Ire-
land, Denmark, and Spain suffered somewhat higher attrition. Respond-
ents were followed over a maximum of 4 years. Since respondents could
die, while new individuals could also join the survey in calendar years
later than 1998, the average follow-up time was less than the maximum
achievable. However, most people (85%) participated in all four waves.
Detailed information on the distribution of age at entry, number of deaths
as well as average follow-up time is presented in Table 1.
All individuals were asked if they were hampered in their daily activities
by any physical or mental health problem, illness, or disability in each
wave. Positive response were categorized in two classes of severity: “Yes,
to some extent” or “Yes, severely” (18). For this study, we considered
them as a single disabled category. Persons who were lost to the ECHP
because of moving to an institution were also considered as “disabled”
for the remainder of the study period.
The BMI variable was recoded into four categories: (i) underweight:
BMI ≤18.5, (ii) normal weight: 18.5 < BMI ≤ 25, (iii) overweight: 25 <
BMI ≤ 30, and (iv) obese: BMI >30. After preliminary analysis, under-
weight persons were excluded from further analysis because of the small
number of respondents and the high prevalence of disability in this cat-
egory. Moreover, this group is less relevant for the purpose of our study as
their prevalence of disability is commonly explained as a result of weight
loss caused by ill-health, rather than a causal effect of underweight on
Smoking status was classified into three categories: (i) current daily
smokers, (ii) former (daily) smokers, and (iii) never (daily) smokers. The
latter included those who smoke occasionally or who used to smoke occa-
sionally. With regard to former smokers, health problems may increase
the chances of smoking cessation; therefore the association of health
with former smoking may in part reflect a selection effect. Additionally,
the group of former smokers is a heterogeneous group in terms of time
since quitting. As a result, the interpretation of results for this group is
less straightforward. Therefore, former smokers were excluded from the
presentation of the detailed results.
The distribution of overweight and smoking status in the study popula-
tion is presented in Table 1.
The level of completed education at the fifth wave was used to measure
education status. Individuals were divided into three groups according to
their level of educational attainment based on the International Stand-
ard Classification of Education (19): (i) lower secondary education or
lower; (ii) upper secondary education; and (iii) tertiary education, which
includes higher vocational and university education. For this study, we
considered people with upper secondary or tertiary education as those
with high education.
We employed a multistate Markov (MSM) model, which is often used
to describe the process in which individuals move through health states
over time (20). By fitting MSM models to longitudinal data one is able
to estimate hazard or transition rates. The association between variables
and a particular transition rate was modeled assuming proportional
We defined three health states, nondisabled, disabled, and dead, and
four possible transitions between the health states: incidence (from non-
disabled to disabled), recovery (from disabled to nondisabled), and state-
specific mortality (from both nondisabled and disabled). We merged
data from the countries after preliminary analyses showed only minor
cross-national differences in associations with overweight and smoking.
Transition rates were estimated on the pooled dataset adjusted for age,
sex, overweight status, and smoking status. The advantage of preparing
estimates using the data of all nine countries together was the very large
number of observations that ensured sufficient power in estimating the
age profiles of disability incidence and recovery.
To detect significant and meaningful interactions and to obtain esti-
mates of transition rates that accurately describe the data, various mod-
els were considered and evaluated. Models including two- or three-way
interaction terms between sex, overweight, and smoking status were
tested. No significant interactions were found between smoking and
overweight status. Akaike Information Criteria values indicated that
the best model fit was achieved by including two-way sex-interactions.
In additional analyses according to educational subgroup, we included
interaction terms between education and overweight and smoking
status. Age–sex interaction terms were included in both models.
Because mortality cases were under-registered in the ECHP in most
of the countries, the mortality rates were adjusted to the pooled level
(pooled across country and time) in four steps. First, using national
statistics, mortality rates were calculated over the period 1998–2001
for each age and sex. Second, they were transformed into mortality
rates of nondisabled and disabled assuming that (i) the age- and sex-
specific mortality rates in the overall (i.e., mixed nondisabled–disabled)
obesity | VOLUME 19 NUMBER 7 | jULy 2011 1453
population are the weighted average of mortality rates of nondisabled
and disabled populations, with the proportion of nondisabled and
disabled, respectively as weights estimated from the ECHP, and (ii)
that the ratio between the mortality rate of disabled and nondisabled
populations is equal to the hazard ratio (HR) as estimated from the
ECHP data (21). Third, rescaling factors were calculated to specify how
much age- and sex-specific mortality rates estimated from the ECHP
had to be scaled to make them consistent with the estimates from step
two. Fourth, all predicted mortality rates were multiplied by the cor-
responding rescaling factors, assuming that under-representation of
mortality was the same for each overweight and smoking status com-
bination. A more formal explanation of the decomposition is shown
in Supplementary Data online. Furthermore, we provide a figure in
Supplementary Figure S1 online with typical age profiles to help visu-
alize transition rates.
Multistate life tables were used to estimate LE and LwD at the age of
16. The empirical input for the multistate life tables were the transition
rates described above, after converting them into probabilities (assum-
ing that hazard rates were constant over a life year). Once a life table was
set-up, probabilistic sensitivity analysis (22) was performed to estimate
confidence intervals (CIs) around the LEs and LwD. For this, random
log-HRs were drawn from a multivariate normal distribution, which
was defined by the natural logarithm of the regression coefficients and
their variance–covariance structure. The corresponding transition rates
and life expectancies were calculated for each of the 1,000 draws. The
25th and 975th of the latter ordered values indicated the 2.5% and 97.5%
boundaries of the CIs.
All MSM analyses were performed in R (23), whereas life table
calculations were carried out in Microsoft Excel. Due to technical
limitations of R taking into account ECHP sampling weights was not
The HRs of transitions by overweight status and smoking status
of the main effect model are given in Table 2. The risk of dis-
ability onset was moderately higher for overweight (HR = 1.15,
CI: 1.10–1.20) and considerably higher for obese (HR = 1.64,
CI: 1.54–1.74) than for normal weight persons. Conversely,
overweight showed to be protective for dying among those
who were disabled (HR = 0.77, CI: 0.66–0.90). Daily smoking
was strongly associated with early death among nondisabled
(HR = 1.68, CI: 1.29–2.19), weakly associated with disability
onset, and not associated with recovery.
The combined effect of two risk factors can be obtained as
well by multiplying the corresponding HRs. For example, the
HR of overweight and smoking on disability incidence could
be calculated as follows: 1.09 × 1.15 = 1.25 (compared to a per-
son who has normal weight and is nonsmoker).
The HRs of incidence and recovery of the models assessing
interaction with sex and educational level are given in Table 3.
For men, disability incidence was not associated with over-
weight (HR = 1.00, CI: 0.94–1.07), whereas it was associated
with obesity (HR = 1.35, CI: 1.23–1.48). Among women both
table 1 characteristics of the study population
Number of individualsa
Mean age (range)Mean follow-up timeOverweight (%)Obese (%)Smokerb (%)
Finland 3,17944.7 (16.8–89.8) 2.4 42.210.924.8
Denmark1,69446.5 (16.8–89.6) 2.8 38.69.635.8
Ireland 2,412 45.2 (16.7–90.2)2.5 41.77.825.9
Austria2,72145.4 (15.3–89.6) 2.841.911.329.5
Belgium2,07246.2 (16.3–88.7) 2.722.214.171.124
Greece3,985 49.1 (17.3–90.3)2.051.49.944.2
Spain5,21244.9 (15.9–89.7) 2.843.013.237.9
Portugal4,62846.8 (16.9–89.8)2.843.5 9.528.6
Sum32,638 45.9 (15.3–90.3)2.642.31032.5
Finland3,112 45.8 (16.8–89.5)2.428.713.214.9
Ireland2,37546.2 (16.7–89.6)2.528.1 8.523.9
Austria2,814 47.9 (15.3–89.8)2.9 30.210.815.9
Belgium 2,27847.5 (16.4–89.4)2.724.611.120.7
Italy6,742 47.1 (16.8–89.8)2.624.87.413.2
Spain 5,26247.7 (15.9–89.7) 2.828.9 13.6 20.2
Portugal5,015 49.6 (16.9–89.9) 2.832.5 11.14.8
Sum33,69347.9 (15.3–90.4) 2.628.9 10.4 16.0
ECHP, European Community Household Panel.
aThe number of individuals refers to those who participated in at least two waves of ECHP survey between 1998 and 2001. bThose who are not daily smoker could be
never smokers and former smoker, treating these two groups separately.
VOLUME 19 NUMBER 7 | jULy 2011 | www.obesityjournal.org
overweight and obesity were related to an increased risk of
disability onset (overweight: HR = 1.28, CI: 1.20–1.36; obese:
HR = 1.87, CI: 1.72–2.03). With regard to recovery from dis-
ability no substantial differences between the risk factors were
found among men or women; however, the negative associa-
tion between obese women and recovery, while marginal, is
noteworthy (HR = 0.92, CI: 0.85–0.99).
Educational level was strongly associated with the transition
rates. Low educated people faced higher risk of disability inci-
dence, high state-specific mortality, and lower chance of recov-
ery from disability. However, the relationship between the risk
factors and disability incidence and recovery were similar for
high and low educational level people (results not shown).
Table 4 shows estimates of LE, LwD, and the ratio of LwD
to LE by overweight status and smoking status. A positive
association between overweight and LE was found for both
nonsmokers and smokers, as well as for men and women. For
example, among nonsmokers, the difference in LE relative to
normal weight was 2.0 years for men and 2.9 years for women.
Conversely, obesity was negatively associated with LE, although
much more clearly for women (−1.6 years) than for men (−0.2
years). Daily smoking was inversely related to LE. Smokers had
3.5 and 2.3 years lower LE than nonsmokers, among normal
weight men and women, respectively.
The relationship between overweight status and LwD was dif-
ferent: the higher the BMI category the more years are expected
to be spent with disability. Normal weight, overweight, and
obese nonsmoker men can expect to live 9.5, 10.1, and 11.8
years with disability, respectively. Smokers expect lower LE
but the same number of years with disability. In general, over-
weight status was the main driver of the proportion of LE spent
with disability. For example, normal weight, overweight, and
obese nonsmoker women expect to live 11.9 (18.1% of LE),
15.5 (22.6%), and 18.0 (28.2%) years with disability. These gen-
eral patterns were observed for both sexes despite some varia-
tion between men and women.
Figure 1 presents LE and LwD estimates by overweight and
smoking status among low and high educated. The positive
association between overweight and LE was observed among
both high educated and among low educated. For both high
and low educated people, the previously observed patterns were
found: in higher BMI categories people expect to live more
years in disability, whereas smoking is not related to the pro-
portion of life spent with disability. The magnitude of LwD and
its proportions of LE were considerably higher for low educated
than for high educated persons due to the association between
educational level and both mortality and disability burden.
This study quantified the relationship of overweight status
with longevity and the burden of disability in Western-Europe
in terms of LE and LwD. Overweight people can expect to
live slightly longer than those with normal weight, which—
suggested by other studies—might be a consequence of the
protective effect of overweight on mortality in disabled popu-
lations. In contrast, overweight and obese people can expect
to live 3.6 and 6.1 more years with disability though, respec-
tively. Smoking had a different relationship with LE and LwD,
as it was associated with lower LE but with an unchanged
LwD. Similar patterns were observed both among men and
women, and among low and high educated populations.
A number of sensitivity analyses were performed to further
explore the relationship between overweight status, smoking
status, and the burden of disability.
table 2 hazard ratios associated with disability incidence, recovery from disability, and state-specific mortality by overweight
status and smoking status
Hazard ratiosa (CI) (number of transitions)
Normal weight +
Normal weight +
Disability incidence1.001.09*1.15* 1.64*
(1.04, 1.15)(1.10, 1.20)(1.54, 1.74)
Recovery from disability 1.00 1.011.03 0.96
(0.95, 1.07)(0.98, 1.08)(0.90, 1.02)
Mortality of nondisabled1.001.68* 0.81 1.06
(1.29, 2.19)(0.63, 1.03)(0.73, 1.52)
Mortality of disabled1.00 1.040.77* 0.82
(0.81, 1.34) (0.66, 0.90)(0.65, 1.03)
Hazard ratios are interpreted at average age.
aDerived from a multistate Markov (MSM) model that included age, sex, age × sex, overweight, obese, past daily smoking, and daily smoking.
*P < 0.05.
obesity | VOLUME 19 NUMBER 7 | jULy 2011 1455
table 3 hazard ratios associated with disability incidence and recovery from disability by overweight status and smoking status
according to sex and educational level
Normal weight +
Normal weight +
Hazard ratio of incidence
(1.04, 1.15)(1.10, 1.20) (1.54, 1.74)
4,6021.00 1.15*1 1.35*
(1.07, 1.24)(0.94, 1.07) (1.23, 1.48)
5,642 1.00 1.051.28* 1.87*
(0.97, 1.16)(1.20, 1.36)(1.72, 2.03)
7,3071.00 1.11* 1.13*1.54*
(1.04, 1.18) (1.07, 1.19) (1.43, 1.66)
2,9371.00 1.13*1.12* 1.72*
(1.03, 1.24)(1.03, 1.22) (1.53, 1.94)
Hazard ratio of recovery
9,171 1.00 1.011.030.96
(0.95, 1.07)(0.98, 1.08) (0.90, 1.02)
4,0571.00 1.081.04 1.01
(0.99, 1.17)(0.97, 1.11)(0.92, 1.12)
5,114 1.00 0.931.03 0.92*
(0.85, 1.02)(0.97, 1.10)(0.85, 1.00)
6,5241.00 1.02 1.07*1.00
(0.95, 1.10)(1.01, 1.13) (0.92, 1.07)
2,647 1.00 1.010.96 0.88*
(0.92, 1.11)(0.88, 1.05)(0.78, 0.99)
Hazard ratios are interpreted at average age.
aDerived from a multistate Markov (MSM) model that included age, sex, age × sex, overweight, obese, past daily smoking, and daily smoking. bDerived from a multistate
Markov (MSM) regression model that included: age, sex, age × sex, overweight, obese, past smoking, daily smoking, overweight × sex, obese × sex, past smoker × sex,
and daily smoker × sex. cDerived from a multistate Markov (MSM) regression model that included: age, sex, age × sex, education, overweight, obese, past smoking, daily
smoking, overweight × education, obese × education, past smoker × education, and daily smoker × education.
*P < 0.05.
table 4 total life expectancy and life expectancy with disability by smoking and overweight status at the age of 16
Normal weightOverweightObeseNormal weightOverweightObese
LE60.462.4 60.2 56.958.8 56.7
(60.2, 60.5)(62.2, 62.5)(60.0, 60.4)(56.7, 57.0)(58.7, 59.0)(56.6, 56.8)
LwD 9.5 10.111.8 8.99.511.1
(9.1, 9.9)(9.7, 10.5)(11.1, 12.6)(8.4,9.4) (9.0, 10.0)(10.3, 11.9)
LwD/LE 15.70%16.20%19.60%15.70% 16.10%19.60%
LE65.568.4 63.963.2 66.4 61.9
(65.4, 65.7)(68.2, 68.6) (63.7, 64.1)(62.9, 63.4) (66.2, 66.7)(61.6, 62.2)
LwD11.9 15.518.0 12.015.7 18.3
(11.5, 12.3) (15.0, 15.9)(17.3, 18.8) (11.3, 12.7)(14.8, 16.7) (17.2, 19.5)
LwD/LE 18.10%22.60% 28.20%19.00% 23.70%29.60%
Derived from an multistate Markov regression model that included: age, sex, age × sex, overweight, obese, past daily smoking, daily smoking, overweight × sex, obese × sex,
past daily smoker × sex, and daily smoker × sex.
LE, total life expectancy; LwD, life expectancy with disability.
VOLUME 19 NUMBER 7 | jULy 2011 | www.obesityjournal.org
First, an additional analysis was carried out to assess the sen-
sitivity of the results to differences in the ways in which surveys
in the participating countries were carried out. To this end,
dummy variables for each country were added to the “main
effect” MSM model. Such model can be seen as a fixed effects
model. Differences between the countries were found in terms
of the general level of transition rates. However, these inter-
country differences did not substantially alter our key findings.
For example the HR of disability incidence among obese were
1.49 (CI: 1.40–1.59) vs. 1.64 (CI: 1.54–1.74) with and without
controlling for counties, respectively. Similar (but generally
smaller) effects were found for all risk factors.
Second, the original obese category (BMI >30) was split into
two subcategories: BMI between 30 and 35 (mild obese), and
>35 (severe obese). Our results indicated heterogeneity within
the group of obese people, with the severe obese being worse
off. The incidence of disability was associated with mild obes-
ity (HR = 1.60, CI: 1.50–1.70) and slightly stronger with severe
obesity (HR = 1.91, CI: 1.69–2.16). Besides, the relative risk
of recovering from disability was lower in the severely obese
group (HR = 0.79, CI: 0.69–0.90) than in the mild obese group
(HR = 1.00, CI: 0.94–1.07). No statistically significant differ-
ences were found in mortality risk. HRs of the other risk fac-
tors remained stable after recoding the obese category.
NW OW OBNW OW OBNWOWOB NWOW OB
NW = Normal weight
OW = Overweight
OB = Obese
NW OWOBNW OW OBNWOW OB NWOWOB
NW = Normal weight
OW = Overweight
OB = Obese
Figure 1 Life expectancy estimates according to smoking and overweight status at age 16 by educational level. DFLE, disability-free life expectancy;
LwD, life expectancy with disability.
obesity | VOLUME 19 NUMBER 7 | jULy 2011 1457
Third, the influence of age on the association was explored.
It is often argued that the size of the association between a risk
factor and outcome decreases with age. Therefore two-way
interaction variables between age and the risk factors were
added to the “main effects” MSM model. We found significant
interaction terms between age and obesity (for disability inci-
dence), and between age and daily smoking (for both disability
incidence and recovery). The age-interaction term was fairly
small indicating an approximately a 0.4% decrease with every
year of increasing age.
data strength and limitations
The ECHP data has a number of strengths of note. The most
evident of which is the availability of a large number of observa-
tions on disability incidence or recovery, and moreover the use
of an identical survey design and survey questionnaire for all
participating countries. Although the data collection was car-
ried out by the National Institutes of Statistics separately, and
hence national versions of the ECHP are not perfectly compa-
rable, these common questions ensured a much higher degree
of comparability between countries than would have been
possible using national data sources. The key variables used in
this article, on health status and risk factors, are comparable
across all ECHP countries. Our study was the first to assess the
impact of overweight for large number of Western-European
countries simultaneously. In contrast, previous European stud-
ies focused exclusively on single countries, and used relatively
much smaller data sets.
This study also has a number of limitations, however. First,
disability data were self-reported, which can result in either
under- or over-reporting of disability. If the reporting of dis-
ability differs by risk factor group then our estimates of impact
of LE and LwD will also be biased. Studies have shown that
measurements of functional limitations by self-report or objec-
tive measures are consistently associated, and that they reflect
similar assessment of function (24,25). It is therefore likely that
objective measures of disability would show similar associa-
tions with smoking, overweight status, and education.
A second potential limitation is related to the use of self-
reports on smoking and on weight and height. It has been found
that smoking prevalence rates are underestimated if estimates
are based on self-report (26). Similarly, people tended to under-
report weight and over-report height (27). These findings were
confirmed by recent studies (28,29). If such under-reporting of
daily smoking and BMI levels is nondifferential, it may lead to
an underestimation of the association with disability, and thus
an underestimation of the differences in LwD.
A third drawback relates to nonresponse and attrition, which
might be a problem in our study if they are related to disability
or risk factor groups. Some studies have explored the attrition
in the ECHP. For example, analyses on attrition in the ECHP
showed a positive relationship between attrition and worsen-
ing health in all countries (30) and only a weak relationship
between attrition and educational level (31). Additional to
these studies, we assessed the likelihood of attrition in the cur-
rent study population in relation to characteristics at the last
wave in which the respondent participated (32). We found that
the risk of loss to follow-up for reasons other than death or
institutionalization was hardly related to disability status, sex,
overweight status, or educational level (relative differences
of ~10% or less). We found slightly higher risk of attrition in
former and daily smokers only (relative difference 15%). These
findings imply that the relative differences are unlikely to have
a major effect on differential retention, and therefore attrition
could not have strongly biased our estimates of LE and LwD by
When we adjusted the estimates for underestimation of
mortality in the ECHP data, we had to assume that the degree
of underestimation was identical in each risk factor group. If,
for a specific risk factor group the under-registration of mor-
tality would in fact be larger than what we assumed, the LE for
this group would be overestimated. It is difficult to assess to
what extent this is a problem. Even though the causes of under-
registration of deaths in the ECHP data are not known, we see
no reason to expect a strong relationship with overweight sta-
tus or other risk factors. Furthermore, we would like to point
out that even though this problem might affect LE estimates, it
could not explain the large differences between BMI groups in
estimates of LwD.
Data from previous studies are not directly comparable with
our data for several reasons: different ages for which life expect-
ancies were calculated; many of the previous studies reflect life
histories of older cohorts; and different measures and clas-
sifications of disability were used. Nonetheless, comparisons
could be made with regard to the general patterns observed.
Peeters et al. (7,33) found large effects of overweight and
obesity on premature death. According to their results non-
smoking men and women lost 5.8 and 7.1 years of LE at age 40,
respectively, due to obesity. Because of both higher disability
prevalence and higher mortality in the obese population, the
authors found no significant difference in LwD (measured by
activities of daily living scores) between the obese and those of
normal weight. However, these effects reflect life histories of
older cohorts. There is evidence that the effects of high BMI on
mortality have diminished over time (10,34).
Studies using more recent data from the United States have
already justified smaller impacts on LE but large effects on LwD.
For example, Reuser et al. (11) estimated the burden of mortal-
ity of obesity among middle- and old-age adults in the Health
and Retirement Survey. LE and LE with activities of daily liv-
ing disability were calculated in relation to self-reported BMI,
smoking, and education at age 55. Obesity was found to have
only a limited effect on mortality, and overweight was esti-
mated to be protective for dying. Both overweight and obesity
increased LwD for both sexes. These results are closely consist-
ent with our findings.
Walter et al. estimated the influence of overweight and obes-
ity on mortality and disability by quantifying its effect in terms
of disability-free LE and years lost to disability for a suburban
elderly population in the Netherlands (35). Similar to our
VOLUME 19 NUMBER 7 | jULy 2011 | www.obesityjournal.org
conclusions, they did not find that increased BMI reduces LE
because of a protective effect of overweight on death (HR = 0.81
compared to normal weight). As in our study, both overweight
and obesity were found to be associated with a larger number
of years lived with disability.
The relation between overweight and mortality has been a con-
troversial topic. We showed that overweight is associated with
protection for dying among the disabled, whereas the relation-
ship between obesity and mortality is modest, especially among
men. Our results reaffirm recent doubts about overstated con-
cerns of overweight and obesity in terms of excess mortality
(36). Possible explanations for the small and even protective
mortality effect of overweight include improved survival of
overweight persons from major diseases of developed socie-
ties, e.g., heart failure (37) or cardiovascular disease (38), and
a better nutritional status providing necessary reserves during
chronic disease (39). Recent studies reaffirmed the protective
effect of overweight as well documenting that increased BMI
protects against mortality after hospitalization (40).
The steadily increasing LE in the developed countries is a
great achievement but at the same time a major concern.
It raises the question of whether living longer lives will be
accompanied by a decrease or an increase of disability during
old age. The main concern is that the obesity epidemic may,
in the long run, increase the prevalence of disability in ageing
Our results show that tobacco control is still highly relevant
to the prevention of premature death. Continued success in
this area may contribute to a further increase in LE, but with-
out substantially affecting the burden of disability over the life
course (i.e., in terms of LwD). This stresses the importance of
further public health aimed to address the obesity epidemic.
Given the large impact of overweight and obesity on the bur-
den of disability over the life course, halting the obesity epi-
demic is essential if an increase of LwD is to be stopped as LE
continues to grow.
Supplementary material is linked to the online version of the paper at http://
This work was supported by MicMac, an international research project
funded by the European Commission in the context of the Sixth
Framework Programme (grant number: SP23-CT-2005-006637). The
funding organization did not participate in the design and conduct of the
study, collection, management, analysis, and interpretation of the data;
and preparation, review, or approval of the manuscript. MicMac has been
developed by a team of researchers from the Netherlands Interdisciplinary
Demographic Institute (NIDI), The Hague, the Vienna Institute of Demography
(VID), the Institut National d’Etudes Demographiques (INED), Paris, the
Bocconi University, Milan, the Erasmus Medical Centre, Rotterdam, the
Max Planck Institute for Demographic Research, Rostock, the International
Institute for Applied System Analysis (IIASA), Laxenburg, and the University
of Rostock. This study was also part of the project “Living longer in good
health,” which was financially supported by Netspar (grant number:
The authors declared no conflict of interest.
© 2011 The Obesity Society
1. Vaupel JW, Carey JR, Christensen K et al. Biodemographic trajectories of
longevity. Science 1998;280:855–860.
2. Hubert HB, Bloch DA, Oehlert JW, Fries JF. Lifestyle habits and compression
of morbidity. J Gerontol A Biol Sci Med Sci 2002;57:M347–M351.
3. Khaw KT. Healthy aging. BMJ 1997;315:1090–1096.
4. Mor V. The compression of morbidity hypothesis: a review of research and
prospects for the future. J Am Geriatr Soc 2005;53:S308–S309.
5. Olshansky SJ, Passaro DJ, Hershow RC et al. A potential decline in
life expectancy in the United States in the 21st century. N Engl J Med
6. Waxman A, Norum KR. Why a global strategy on diet, physical activity and
health? The growing burden of non-communicable diseases. Public Health
7. Peeters A, Bonneux L, Nusselder WJ, De Laet C, Barendregt JJ.
Adult obesity and the burden of disability throughout life. Obes Res
8. Calle EE, Thun MJ, Petrelli JM, Rodriguez C, Heath CW Jr. Body-mass
index and mortality in a prospective cohort of U.S. adults. N Engl J Med
9. Adams KF, Schatzkin A, Harris TB et al. Overweight, obesity, and mortality
in a large prospective cohort of persons 50 to 71 years old. N Engl J Med
10. McGee DL; Diverse Populations Collaboration. Body mass index and
mortality: a meta-analysis based on person-level data from twenty-six
observational studies. Ann Epidemiol 2005;15:87–97.
11. Reuser M, Bonneux LG, Willekens FJ. Smoking kills, obesity disables: a
multistate approach of the US Health and Retirement Survey. Obesity (Silver
12. Reynolds SL, Saito Y, Crimmins EM. The impact of obesity on active
life expectancy in older American men and women. Gerontologist
13. Lahoz-Rallo B, Blanco-Gonzalez M, Casas-Ciria I et al. Cardiovascular
disease risk in subjects with type 2 diabetes mellitus in a population in
southern Spain. Diabetes Res Clin Pract 2007;76:436–444.
14. Brønnum-Hansen H, Juel K, Davidsen M, Sørensen J. Impact of selected
risk factors on expected lifetime without long-standing, limiting illness in
Denmark. Prev Med 2007;45:49–53.
15. Visscher TL, Rissanen A, Seidell JC et al. Obesity and unhealthy life-years in
adult Finns: an empirical approach. Arch Intern Med 2004;164:1413–1420.
16. Robine JM, Jagger C, Mathers CD, Crimmins EM, Suzman RM. Determining
Health Expectancies. Wiley: Chichester, England, 2003.
17. Peracchi F. The European Community Household Panel: A review. Empir
18. Eurostat. ECHP UDB - Description of Variables. Data Dictionnary, Codebook
and Differences Between Countries and Waves. Luxemburg, 2003.
19. UNESCO. International Standard Classification of Education 1997. 1997.
20. Jackson CE, Snyder PJ. Electroencephalography and event-related
potentials as biomarkers of mild cognitive impairment and mild Alzheimer’s
disease. Alzheimers Dement 2008;4:S137–S143.
21. Nusselder WJ, Looman CW, Marang-van de Mheen PJ, van de Mheen H,
Mackenbach JP. Smoking and the compression of morbidity. J Epidemiol
Community Health 2000;54:566–574.
22. Briggs A, Schulpter M, Claxton K. Decision Modelling for Health Economic
Evaluation. Oxford University Press: Oxford, 2006.
23. Team RDC. R: A Language and Environment for Statistical Computing.
R Foundation for Statistical Computing: Vienna, Austria, 2008.
24. Coman L, Richardson J. Relationship between Self-Report and Performance
Measures of Function: A Systematic Review. Can J Aging 2006;25:253–270.
25. van den Brink CL, Tijhuis M, Kalmijn S et al. Self-reported disability and its
association with performance-based limitation in elderly men: a comparison
of three European countries. J Am Geriatr Soc 2003;51:782–788.
26. Gorber SC, Schofield-Hurwitz S, Hardt J, Levasseur G, Tremblay M. The
accuracy of self-reported smoking: a systematic review of the relationship
between self-reported and cotinine-assessed smoking status. Nicotine Tob
obesity | VOLUME 19 NUMBER 7 | jULy 2011 1459 Download full-text
27. Gorber SC, Tremblay M, Moher D, Gorber B. A comparison of direct vs.
self-report measures for assessing height, weight and body mass index:
a systematic review. Obes Rev 2007;8:307–326.
28. Keith SW, Fontaine KR, Pajewski NM, Mehta T, Allison DB. Use of
self-reported height and weight biases the body mass index-mortality
association. Int J Obes (Lond) 2010.
29. Stommel M, Schoenborn CA. Accuracy and usefulness of BMI measures
based on self-reported weight and height: findings from the NHANES &
NHIS 2001-2006. BMC Public Health 2009;9:421.
30. Eurostat. Attrition in the ECHP. Eurostat: Luxemburg, 2002.
31. Eurostat. Sample Attrition Between Wave 1 and 4 in the European
Community Household Panel. Eurostat: Luxemburg, 2002.
32. Majer IM, Nusselder W, Kunst AE. Age Profiles of Mortality and Disability-
Related Transitions in European Countries. Estimates According to Smoking
Status, and Overweight Status, Educational Level and Marital Status. In:
MicMac reports; Erasmus MC: Rotterdam: 58, 2008.
33. Peeters A, Barendregt JJ, Willekens F et al.; NEDCOM, the Netherlands
Epidemiology and Demography Compression of Morbidity Research Group.
Obesity in adulthood and its consequences for life expectancy: a life-table
analysis. Ann Intern Med 2003;138:24–32.
34. Flegal KM, Graubard BI, Williamson DF, Gail MH. Excess deaths
associated with underweight, overweight, and obesity. JAMA
35. Walter S, Kunst A, Mackenbach J, Hofman A, Tiemeier H. Mortality
and disability: the effect of overweight and obesity. Int J Obes (Lond)
36. Basham P, Luik J. Is the obesity epidemic exaggerated? Yes. BMJ
37. Oreopoulos A, Padwal R, Kalantar-Zadeh K et al. Body mass index
and mortality in heart failure: a meta-analysis. Am Heart J 2008;156:
38. Nusselder WJ, Franco OH, Peeters A, Mackenbach JP. Living healthier
for longer: comparative effects of three heart-healthy behaviors on life
expectancy with and without cardiovascular disease. BMC Public Health
39. Baik I, Ascherio A, Rimm EB et al. Adiposity and mortality in men. Am J
40. Peake SL, Moran JL, Ghelani DR, Lloyd AJ, Walker MJ. The effect of obesity
on 12-month survival following admission to intensive care: a prospective
study. Crit Care Med 2006;34:2929–2939.