In this paper, estimates of HALE were presented for different cohorts defined conditional on risk factor classes. Estimates of HALE are highest for 'healthy living' people (54.8 for men and 55.4 for women at age 20). Differences in HALE compared to 'healthy living' men at age 20 are 7.8 and 4.6 for respectively smoking and obese men. Differences in HALE compared to 'healthy' women at age 20 are 6.0 and 4.5 for respectively smoking and obese women. At older ages differences in HALE are larger than differences in LE. For all cohorts unhealthy life expectancies are approximately 8 years for men and 10 years for women. As a result, a slight relative compression of morbidity occurs if prevention of smoking and/or obesity is successful. Estimates of LE and HALE and conclusions about compression of morbidity should be made with great caution because of uncertainty with respect to future changes in disease epidemiology. In sensitivity analyses we investigated the sensitivity of our results to future changes in disease epidemiology and to changes in the age gradient of the health state valuations of the different cohorts. Main results of the sensitivity analyses were that although estimates of LE and HALE are sensitive to changes in disease epidemiology, differences in LE and HALE between the different cohorts remain substantial. Furthermore, although a sharper decrease in the health state valuations at older ages results in an absolute expansion of morbidity this does not result in a relative expansion of morbidity. Overall, in most scenarios the proportion of life expectancy spent in good health is a fairly stable proportion for the different cohorts.
It is argued that an incidence based estimate of HALE using a state-transition model is a better indicator than a prevalence based indicator for public health policy since it is not biased by the stock of diseases built up in the past [60
]. The drawback of an incidence based methodology is that state-transition models are required for which the data requirements are very large. To minimize data requirements, we only modeled marginal disease prevalence rates, and did not model comorbidity (joint disease prevalence rates). However, as with any modeling study, results depend on the assumptions used in constructing the simulation model and the input data used. We will first discuss the different assumptions and then proceed to discuss the sensitivity of the results in relationship to the input data used.
First of all, we did not distinguish between light and heavy smokers, and BMI was not treated as a continuous variable. The relative risks used for the BMI classes are risk factor class averages calculated using data on the BMI distribution in the Netherlands. Using risk factor class averages in a cohort implies, over age, the most obese tend to die early reducing the mean level of obesity, with age. To what extent the risk factor class averages calculated using the current BMI distributions in the Netherlands can be used as an approximation to simulate average relative risks of a cohort depends on the stability of these distributions over time.
To estimate baseline incidence and mortality rates for the 'healthy living' cohort, we assumed independence between risk factor classes and multiplicative relative risks. Furthermore, relative risks on disease incidence rates are used as an approximation for disease prevalence rates to estimate relative risk for the different cohorts on other causes of death. Although these assumptions may be violated in practice, necessary data to fill this gap are missing.
Another crucial assumption was the one regarding excess mortality risks. Patients with a specific disease have a higher mortality risk than persons without the disease, all other variables equal. The difference in mortality is expressed here as excess mortality which is used in calculations of the disease prevalence rates. However, in general only part of this excess mortality can be attributed to the specific disease, which is used in our calculations of population numbers. The difference between the excess mortality and the part uniquely attributable to the disease can be interpreted as mortality due to co-morbid conditions. The mortality due to co-morbid conditions is especially important on higher ages, such as for COPD for which smoking is an important risk factor with many other related chronic diseases. So part of the COPD excess mortality must be attributed to other smoking related diseases (e.g. coronary heart disease and lung cancer) [61
]. Therefore, in the calculation of the prevalence rates excess mortality rates are used, while in the calculation of the number of survivors disease specific attributed mortality rates are used. So far, the problem of excess mortality has received relatively little attention in most population models. Therefore, developing methods to establish relations between excess mortality rates and attributable mortality rates should deserve more attention.
Disability weights for comorbidity were defined assuming a multiplicative adjustment method. We tested for this using alternative weighing methods [22
]. Although this affected absolute estimates of HALE it only had a minor influence on differences in HALE. The same also goes for the influence of diseases not causally related to BMI and/or smoking. Excluding them raised HALE estimates, but did not affect substantially the differences found between groups.
Lastly, we assumed that no transitions occur between risk factor classes over time. In reality, of course, transitions between classes do occur: some smokers quit (and some of them might start again later) and obese people of course might lose weight. Moreover, obesity has a more complex age trajectory than smoking in that body composition changes with age [62
Recently, it was argued that the excess mortality due to obesity had been overestimated and that the effects of obesity attenuate with age, and are not strongly related to mortality above age 70 to 75 [63
]. Of course, our estimates of LE, and thus also of HALE, would increase for the obese cohort if we imputed relative risks as reported by a study finding lower risks. Other studies, however, reported higher mortality risks associated with obesity [39
] which would lead to lower LE estimates for the obese cohort [65
]. Moreover, in our analysis, our cohort was defined as being obese but non-smoking. It has been shown that excess mortality due to obesity is highest for never smokers [39
Even though successful prevention would result in health gains this is not necessarily accompanied by a reduction in health care costs. A decline in costs due to risk factor related diseases may well be outweighed by an increase in costs due to risk factor unrelated diseases, especially in life years gained. Prevention, when successful in prolonging life, may therefore cause more costs than it avoids [66
]. However, this will of course largely depend on the risk factor under study. A next step, therefore, would be to compare the effects of smoking and obesity prevention on health care costs. We conclude that losses in HALE due to smoking and obesity are substantial and that prevention of smoking and obesity can considerably increase both life expectancy and health-adjusted life expectancy. This knowledge underpins the importance of continuing public health policies to prevent unhealthy behavior. However, our results do not indicate that substantial compression of morbidity is to be expected as a result of successful smoking or obesity prevention.