Summary of results
The current paper is among the first to compare the effect of different life style factors on years lived with disability. Compared with smoking and drinking alcohol, obesity is more strongly associated with an increased risk of spending many years in disability during life. Using information on time to death in the Sullivan life table does not lead to substantively different estimates of the relative importance of the risk factors.
Evaluation of data and methods
Among all subjects approached in the baseline survey, the non-response was 42%. We could not directly evaluate the effect of selective response on our estimates of years lived with disability, but the total mortality in our study sample (3.15% per year) was comparable to the total mortality in the Dutch population (1997-2006) aged 55 and older (3.21% per year) [
26]. Although non-response is likely to lead to biased estimates of the prevalence of smoking and alcohol intake, it may not substantially affect associations between these risk factors and health outcomes [
27]. None the less, we cannot exclude that selective non-response may have biased our estimates of effects of risk factors on years lived with disability.
The use of self reported measures of disability may have caused some reporting bias, particularly if risk factor exposure was related to reporting behaviour, independently from status of disability. However, in previous studies it was shown that self report of ADL disabilities correlates strongly with performance based measurement of disability [
28], suggesting that it is unlikely that reporting bias could have substantially biased our results.
Self-reported BMI tends to be underreported, particularly by those who have a high BMI [
29,
30]. Therefore, the risk of disability among persons who had a high BMI may have been overestimated. However, the potential bias has been shown to be acceptable for correlation analyses like in our study [
30]. Underreporting in self-reported alcohol consumption may affect prevalence rates, but does not appear to have substantially bias effect on the association between heavy drinking and harmful consequences [
31].
The institutionalized population, an old population with a high disability prevalence, was not included in the baseline survey [
32,
33]. It cannot be excluded that this exclusion may have led to some bias in our estimates of risk factors in relation to years lived with disability. However, in the Netherlands, only a minor part of elderly people live in an institution, e.g. 90% of those aged 80-85 still live at home. Hence, the bias because of excluding the institutionalized population will probably be small [
32,
33].
Our choice of disability cut-off level was arbitrary. Using a more stringent cut-off level of having "with major difficulty" or "only with help" for at least two items resulted in (non-significantly) higher odds ratios for disability for BMI (OR obesity = 2.96) and alcohol consumption (OR >14 alc cons/wk = 1.52), but in a lower odds ratio for smoking (OR current smoker = 1.48). Years lived with disability differed by 2.0 years according to BMI, 0.2 years by smoking and 1.4 by alcohol consumption. Including time to death to the calculation yielded similar results. It can be concluded that the cut-off level to define disability has had no major influence on our substantive conclusions.
The association of BMI, smoking and alcohol with disability is likely to be mediated by the occurrence of specific diseases or other risk factors such as physical activity. If our aim had been to gain insight into causal chains that relate risk factor exposure with disability, it would have been useful to include more covariates. However, as our analysis had a descriptive purpose, and adjusting for co-morbidities or physical activities would take out part of the effect of lifestyle on disability, we used simple and transparent univariate regression models, adjusted for age, sex and marital status only, to estimate the years lived with and without disability.
Comparison with previous studies
A few other studies compared lifestyle factors with respect to their effect on years lived with disability. Most studies used other outcome measures [
14,
34‐
36]. Comparable with our results, these studies found that, compared to heavy drinking or regular smoking, obesity had a much greater effect on the number of years that people could expect to live with long-standing illness, with reduced quality of life, or with less than good self-assessed health [
34‐
36]. The only study that compared effects of different lifestyle factors on years lived with disability confirmed our key finding that obesity is more important than smoking [
14].
The obesity paradox
We found that smoking is associated with shorter life, whereas obesity is associated with spending more life years with disability. This difference is likely to be related to the fact that a high BMI is more strongly associated with non-lethal disabling diseases, such as osteoarthritis and chronic back pain, whereas smoking is more strongly associated with a series of fatal diseases with a relatively short period of disablement, such as lung cancer and other types of cancer [
37‐
39].
Compared with other risk groups, obese persons on average spend a larger part of their last years of life with disability (Figure
1). This observation represents another side of the 'obesity paradox', which refers to the fact that increased BMI is an independent risk factor for heart failure, but that among patients with established heart failure, those who are overweight or obese are at decreased risk for death [
40‐
44]. This suggests that obesity is a 'stretcher of disease and disability', which results in a high prevalence of disability prior to death among obese persons.
Time to death in the Sullivan life table
The occurrence of ADL disability is not only associated with age (time since birth), but even more strongly with time to death [
21,
22]. A substantial part of disability occurs in relation with end-of-life processes [
21,
22]. As a result, a longer life is likely to be associated with a shift of the burden of disability towards older ages [
21,
22]. Conventional Sullivan life table methods do not account for possible shifts of disability towards older ages, but use one age schedule of disability, irrespective of the length of life (i.e. age of death) of individual people. Using an innovative approach, we accounted for the association of disability with length of life by defining schedules of disability not only as a function of age, but also as a function of age of death (Figure
1). The new estimates of the number of years lived with disability differed substantially from the original estimates. As expected, the expected years lived with disability were lower according to the new estimates. However, the relative importance of risk factors remained unchanged.
Therefore, this new methodology may be useful for obtaining more precise estimates of the occurrence of disability across the life cycle. It may be especially useful to assess the effect of increasing life expectancies on years with disability, which may have been overestimated in conventional methodologies. On the other hand, conventional methods appear to have yielded valid estimates of the relative importance of different risk factors.