Results

Figure 1 shows that at each age the prevalence of disability is lower in the non-smoking than in the mixed smoking-non-smoking population. The reason is that non-smokers have a lower incidence of disability and a higher recovery from disability (table 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Prevalence of disability by age in the non-smoking and mixed (smoking/non-smoking) population based on the multistate life table.

To assess whether non-smoking reduces the number of disability years, the lower mortality risks and thus the longer period in which non-smokers will be exposed to disability should be considered as well. By looking at (changes in) the number of person years with and without disability (by age) and life expectancy with and without disability (summarised across ages)—both outcomes of the multistate life table—it is possible to take into account the effects of changes in disability and mortality at the same time. Figure 2 shows that at younger ages the number of person years with disability is smaller in the non-smoking population than in the current mixed population, whereas at older ages the opposite is true. On balance, life expectancy with disability at age 30 is lower in the non-smoking than in the mixed population (table 3). In the mixed population, total life expectancy at age 30 is 44.8 years (men) and 50.8 years (women), of which about 38.5 years without disability and respectively 6.4 and 12.4 years with disability. In the non-smoking population total life expectancy is 46.4 (men) and 51.6 years (women), of which 41.0 and 40.3 years, respectively, without disability and 5.5 and 11.3 years, respectively, with disability. These results indicate that elimination of smoking will produce a substantially larger gain in disability free life expectancy (2.5 years in men and 1.9 year women) than in total life expectancy (1.6 and 0.8 years respectively). Consequently, smoking elimination will reduce the number of years with disability (−0.9 and −1.1 years respectively). This reduction in the number of years with disability indicates that absolute compression of morbidity will occur. At age 70, the effects of elimination of smoking were in the same direction, but smaller.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Number of person years with disability by age in the non-smoking and mixed (smoking-non-smoking) population based on the multistate life table.

Table 4 shows that the conclusion on compression of morbidity was not affected by the probable underrepresentation of the institutionalised population, nor by using lower incidence rates in men based on the GLOBE study. In addition, the outcomes were not very sensitive to different values in the rate ratios associated with smoking. Using 50% lower or higher excess risks did not change the conclusion. Finally, using rate ratios of one for recovery (ceteris paribus), for mortality among non-disabled and disabled persons (ceteris paribus), or excluding in the estimation of the rate ratios persons who reported to have (had) a serious chronic disease at baseline, did not change the conclusion. Nor did using 50% lower excess risks for all transitions at age 70. Only in the unlikely situation that the association between smoking and incidence of disability is substantially weaker in men, while the other associations are not, will virtually no compression of morbidity above age 70 occur (data not shown for age 70).

Discussion

This study evaluated whether eliminating smoking would produce compression of morbidity into a shorter period, by comparing life expectancy with disability in the current mixed smoking-non-smoking population (baseline) and in the non-smoking population (after smoking elimination). Our results show that on balance non-smokers spend fewer years with disability than the mixed smoking-non-smoking population. Although non-smokers have lower mortality risks, and are thus exposed to disability over a longer period, their higher ability to maintain health and to restore health losses results in a net reduction of the length of time spent with disability and in an increase in the life time free of disability. These outcomes indicate that eliminating smoking will extend the length of life and the length of disability free life, and will compress disability into a shorter period.

Previous studies on the effects of smoking have reported that smoking increases the risks of several chronic diseases, such as (lung) cancer, heart disease, stroke, and chronic obstructive lung disease.3 These diseases are significantly associated with mortality2 3 and most of them with disability.22-24 This implies that an effect of smoking on disability and mortality is biologically plausible. Nevertheless, there are still important gaps in our understanding of the mechanisms through which smoking affects disability and mortality. Smoking may have an effect on disability and mortality by increasing disease incidence, decreasing disease recovery, increasing disease severity or by increasing the incidence of comorbid conditions. In addition, physiological losses and symptomatology not operating through specific diseases may have an effect on frailty8 and in turn on disability and mortality. In evaluating the effect of smoking elimination, we used information on the association of smoking with disability and mortality, rather than information on the different diseases and pathways through which smoking affects disability and mortality. This enabled us to take into account the overall impact of smoking without needing information on the exact mechanisms. To better understand the association of smoking with disability and mortality further research is needed to unravel these underlying mechanisms.

The limitations of our study regarding the data should be noted. Firstly, we used data from two different countries to estimate the transition rates between the different health states and to death. Moreover, persons living in institutions were underrepresented in these data because of the sampling design, especially in the LSOA. At different stages in the research process, we examined the consequences of pooling the GLOBE and LSOA dataset. Firstly, we examined whether significant discontinuities existed between the two datasets in the proportions of persons who died and the proportions of persons with disability, and whether the two datasets described the Dutch situation.14 16 We found that the two datasets showed no substantial discontinuities, except the smaller proportion of disabled men in GLOBE. With the exception of the smaller proportions of persons dying at older ages in LSOA (especially in women, most probably reflecting the underrepresentation of the institutionalised population) the datasets described the Dutch situation. Next, we examined whether the transition rates estimated from these datasets differed between the two studies regarding the level (α) and increase with age (β). We found a difference in the level of incidence in men, reflecting the difference in the proportion disabled, and a difference in the level of mortality among disabled women, reflecting the underestimation of mortality in elderly women. Finally, we examined whether the outcomes of the multistate life table model based on these transition rates reproduced the Dutch situation. As could be expected, age specific mortality rates derived from the multistate model were underestimated in women above age 85. Nevertheless, survival curves were very similar to those based on national mortality from Statistics Netherlands and total life expectancy at age 30 and age 70 differed less than 0.3 year in both sexes. The agreement between the prevalence of disability by age based on the multistate model and the external data sources was also rather close. Thus, apart from the underestimation of the institutionalised population in the LSOA and the possible overestimation of incidence in men, we found that the two datasets could be joined and together described the Dutch situation. As, in addition, sensitivity analyses showed that our conclusion was robust to these two factors, we considered the outcomes to be valid.

Secondly, we used the same rate ratios at all ages. Although evidence exists that rate ratios of total mortality for smokers as compared with non-smokers decrease with increasing age,2 3we do not know studies showing a decline in the rate ratios of incidence, recovery and mortality by disability status with increasing age. We examined whether interactions were present between age and smoking within the limited age range of the GLOBE study (30–74 years), but did not find any significant interactions. Moreover, using the same rate ratios of incidence, recovery, and mortality among non-disabled and disabled persons for all ages produced declining rate ratios on total mortality with increasing age as observed in literature.2 3 However, when in reality rate ratios of incidence, recovery and mortality by disability status do decrease with increasing age, our outcomes might still be biased. Such a bias is to be expected when the excess risks for disability but not for mortality are lower at older ages. We consider it more probable that if excess risks associated with smoking are lower at older ages, this decline would affect all transitions, as we expect that the same pathophysiological mechanisms that cause excess risks of disability for smokers, also cause higher mortality risks. Sensitivity analysis, exploring the effect of lower excess risks for all transitions above age 70, showed that even a reduction of the excess risks by 50% for all transitions would not change our conclusion.

Two other methodological issues affect the interpretation of our outcomes. Firstly, we assumed that elimination of smoking would produce a population that experiences the same incidence, recovery and mortality rates as observed for non-smokers in the current mixed smoking-non-smoking population. This assumption is only valid if the differences between the current mixed smoking-non-smoking and smoking population were solely attributable to smoking. This would imply that the association between smoking and incidence, recovery and mortality rates among non-disabled and disabled persons, reflects a causal relation, and is not influenced by variation in the prevalence of other risk factors or confounded by sociodemographic characteristics. Although more research is needed to validate the magnitude of the causal association between smoking and incidence, recovery and mortality among non-disabled and disabled persons, uncertainty about the exact magnitude of the rate ratios for smoking is not expected to have biased our conclusion. Firstly, sensitivity analyses showed that the conclusion is not very sensitive to different values in the rate ratios associated with smoking within the margin of 50% lower and higher excess risks and in some other situations specified in table 4. Secondly, additional logistic regression analyses indicated that a correction for sociodemographic characteristics and for other life style factors than smoking would change the rate ratios only slightly and within the margins of the sensitivity analyses (results not shown). Thirdly, sensitivity analyses removing the possible confounding of the association between smoking and incidence, recovery and mortality among disabled and non-disabled persons, when smokers have quit smoking because of a chronic condition, showed no change in our conclusion. Finally, rate ratios which could be validated, such as those for the incidence of disability and total mortality, were in line with results from previous studies (see Methods section).

Secondly, the population that will emerge after elimination of smoking is assumed to have the same distribution of persons who never smoked and former smokers, and the same average duration since smoking cessation as the current non-smoking population. That is, some persons never smoked, some quitted smoking long ago and some quit smoking recently. Although even after a long period since smoking cessation, mortality and disability risks will still be higher among former smokers than among never smokers, because only part of the biological damage attributable to smoking is irreversible, persons who quitted smoking longer ago, will have lower excess risks as compared with never smokers than persons who quitted recently.3 25 If all smokers were to quit smoking at the same time, initially the effect on the population's health would be less than estimated in our study, whereas as the time since smoking cessation evolves, the health status of the future non-smoking population might be more favourable than expected from our results. To explore whether our conclusion would be valid in the case of a future non-smoking population in which no one had ever smoked, we increased the excess risks associated with smoking by 75% to obtain a difference in life expectancy between smokers and non-smokers comparable to the difference currently observed between smokers and those who have never smoked.2 Repeating our analysis using these higher excess risks did not change our conclusion. In this situation, the reduction in life expectancy with disability attributable to smoking elimination was −1.4 years in men and −1.7 in women at age 30 and −0.5 and −0.4 years, respectively at age 70.

Compression of morbidity, in the presence of an increasing life expectancy, is generally considered a favourable development of population health. The combination of a compression of morbidity and an increase in life expectancy means that disability free life expectancy increases more than total life expectancy, which implies that years in poor health are being replaced by years in good health. However, a compression of morbidity without an increase in life expectancy should not necessarily be considered a favourable development. A compression of morbidity can be the result of dramatic increase in premature mortality. Nor should an absence of compression of morbidity be generally considered as undesirable. Reduction of premature mortality is socially desirable, even if this implies that people live longer with disability. Nevertheless, is clear that the consequences of smoking elimination as shown in this study are socially favourable. Smoking elimination would not only produce a gain in total life expectancy (1.6 years in men and 0.8 years in women) and in disability free life expectancy (2.5 years and 1.9 years, respectively), but in addition, the number of years with disability would decline (−0.9 and −1.1 years respectively).

A recent study found that non-smoking will increase health care costs,26 which would seem to contradict our conclusion that non-smoking reduces the length of time spent with disability. However, it is entirely possible that because of smoking elimination lifetime costs will increase, while the number of years with disability will decrease. As many diseases, even when they do not cause disability, still generate costs, the increase in costs with increasing age exceeds that of disability. As a result, non-smokers saved from dying incur more costs when they age than that they accumulate disability.27 The finding that smoking elimination will produce compression of morbidity supports a previous study of the effect of non-smoking on the number of disability years28despite differences in data and methodology.

Our results have important implications for public health policy makers and doctors. They show that interventions aimed at eliminating smoking will extend the length of life and will reduce the number of years spent with disability. This is an important finding in view of the trade off often found between longer life and more years with disability. For example, interventions aimed at eliminating fatal diseases, such as cancer and heart diseases, will extend the length of life, but at the same time will extend the period with disability. On the other hand, interventions aimed at eliminating non-fatal diseases will compress disability into a shorter period, but will not extend life.22 A successful smoking intervention will extend total life expectancy, extend disability free life expectancy and reduce life expectancy with disability. Therefore, interventions to discourage smoking should receive high priority. A next step would be to pinpoint other possible interventions that will compress of disability into fewer years. Further research on the disablement process29 might provide clues for factors preventing, slowing down or undoing disability, which can be targeted at in further public health interventions.