Impact of obesity on life expectancy among different European countries: secondary analysis of population-level data over the 1975–2012 period

Objective This study assesses the impact of obesity on life expectancy for 26 European national populations and the USA over the 1975–2012 period. Design Secondary analysis of population-level obesity and mortality data. Setting European countries, namely Austria, Belarus, Belgium, the Czech Republic, Denmark, Estonia, Finland, France, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, the Netherlands, Norway, Poland, Portugal, the Russian Federation, Slovakia, Spain, Sweden, Switzerland, Ukraine and the UK; and the USA. Participants National populations aged 18–100 years, by sex. Measurements Using data by age and sex, we calculated obesity-attributable mortality by multiplying all-cause mortality (Human Mortality Database) with obesity-attributable mortality fractions (OAMFs). OAMFs were obtained by applying the weighted sum method to obesity prevalence data (non-communicable diseases (NCD) Risk Factor Collaboration) and European relative risks (Dynamic Modeling for Health Impact Assessment (DYNAMO- HIA)). We estimated potential gains in life expectancy (PGLE) at birth by eliminating obesity-attributable mortality from all-cause mortality using associated single-decrement life tables. Results In the 26 European countries in 2012, PGLE due to obesity ranged from 0.86 to 1.67 years among men, and from 0.66 to 1.54 years among women. In all countries, PGLE increased over time, with an average annual increase of 2.68% among men and 1.33% among women. Among women in Denmark, Switzerland, and Central and Eastern European countries, the increase in PGLE levelled off after 1995. Without obesity, the average increase in life expectancy between 1975 and 2012 would have been 0.78 years higher among men and 0.30 years higher among women. Conclusions Obesity was proven to have an impact on both life expectancy levels and trends in Europe. The differences found in this impact between countries and the sexes can be linked to contextual factors, as well as to differences in people’s ability and capacity to adopt healthier lifestyles.

impact of obesity on mortality diminish with increasing age. How the authors addressed those differences in their analysis? 2. The effect of obesity on mortality is mediated through chronic diseases, such as cardiovascular disease. In the recent decade, there is an improvement in the treatment of cardiovascular disease, and more individuals are living with cardiovascular disease and contributing to increased life expectancy. Therefore, the contribution of obesity to mortality could be smaller due to better treatment of chronic diseases. I think authors should discuss in the discussion to explain the smaller estimates compared to the previous study (Preston et al.). 3. What is the contribution of the differences in socioeconomic status across 26 European countries, especially comparing East and West Europe to the results?

GENERAL COMMENTS
I have read the paper and although I thought that I could provide my expertise in revising it, I have realised that the methods used are not familiar to me, and although I tried to read about them I felt that the statistical analysis was not sufficiently described for me to assess the appropriateness. I feel that a demographer might be better placed in reviewing this article. I am really sorry for not being able to help as much as I would have liked on this occasion.

GENERAL COMMENTS
This study assessed the population level burden of disease associated with obesity by estimating potential gains in life expectancy by eliminating obesity-attributable morality from allcause mortality for 27 countries from 1975 to 2012.
This study used data from difference sources for the calculations: (1) prevalence of obesity (BMI≥30kg/m2), by age, sex, county, and year; (2) relative risk (RR) of dying from obesity, by age categories and sex; and (3) all-cause mortality, by age, sex, and year.
The method for the calculation of potential gains in life expectancy is a valid method. The validity and reliability of estimates depend on the data used in this study. More detailed description of data sources is needed.
1. All-cause mortality estimation: All-cause mortality data by age and sex were used for the estimation of life expectancy (life table method). The authors said that this is "single year of age, sex, and year". The authors did not mention whether this is by counties or combined data for all counties. If this is for each counties, what is reliability of the data? Particularly for some small counties. If this is combined data for all counties, further discussion of this weakness is needed. Since life expectancy varied greatly across European counties (approximately 10 years from the lowest to the highest), I wonder if you can use combined data.

2.
Obesity prevalence estimates: What data were used for the estimation? The estimates were model based. Different models for different counties, or a single model for all counties? What are predictors for these estimates?

3.
RR of dying from obesity: Is obese persons relative to not obese persons or obese persons relative normal weight persons? Many studies estimated relative risk (and sometimes, hazard ratio) of dying of obese persons relative to normal weight persons. If RR is obese persons relative to normal weight person, the estimates are not accurate. Please confirm that RR is relative to not obese persons.

4.
Another weakness is applying same RRs to all 27 countries from 1975 to 2012. Many factors (including race/ethnicity and geographic regions) were associated with the impact of obesity on morality. It might be better applying different RRs for different counties (at least for select counties where such data are available).
Reliability of estimates: This study did not provide standard error or confidence limits of estimates. This information might be important and should be reported (see my comment on reliability life expectancy estimation for small counties above).

2.
Figures are different to see. ) in our analysis. First, we were particularly interested in assessing the impact of obesity, as the prevalence of obesity is alarmingly high, and obesity represents a significant health burden. Given that obesity is one of the biggest public health challenges facing countries today, this issue has attracted extensive attention in the literature. Second, the existing findings on the relationship between overweight and the risk of mortality are not yet conclusive, with some studies reporting that overweight people have a lower relative risk of mortality than the normal weight reference group (Flegal et al., 2013;Flegal et al., 2018) and other studies reporting that the relative risk of mortality is higher among overweight than normal weight people (Aune et al., 2016;Global BMI Mortality Collaboration et al., 2016). In order to clarify our sole interest in obesity, we added the following in the introduction, page 5, lines 90-92: "Our sole focus is on the impact of obesity, given the significant health burden caused by obesity, the large body of literature on its impact, and the well-documented association of obesity with mortality".

VERSION 1 -AUTHOR RESPONSE
Comment 2: To discuss the future implications of increasing obesity would strengthen the paper.

Reply
We extended our discussion of the likely future implications of a continued increase in obesity in our discussion of our results on page 18, lines 368-374: "Moreover, the impact of obesity on life expectancy levels and on life expectancy trends is likely to increase, as previous studies have also suggested (Alley et al., 2011). There are several indicators pointing in that direction, including evidence that obesity"s impact is already substantially greater in the USA (13% among men and 15% among women) than elsewhere; obesity prevalence is increasing rapidly in most European countries (see Supplementary Material Figure S3); obesity is increasing in severity; and the duration of obesity is rising in younger generations (Alley et al., 2011)".
Moreover, on page 18-19, lines 383-391, in our conclusions section, we stressed this point again: "It is likely that in the future obesity will have a larger impact on mortality and life expectancy in Europe, as obesity prevalence and obesity-attributable mortality continue to increase in the majority of countries. These trends will have important health, economic, and social implications. Specifically, the increasing prevalence of obesity among European populations, and especially at younger ages, will lead to an increased prevalence of obesity-related disorders, as well as to increases in the mortality burden associated with obesity and in obesity"s effects on life expectancy and quality of life.
Thus, obesity will constitute an additional burden for societies, economies, and public health".
Reviewer: 2 In this manuscript, the authors aimed to assess the impact of obesity in life expectancy across 26 European countries. The authors estimated the potential gain in life expectancy if obesity-attributable mortality were eliminated using the prevalence data from NCD risk factor collaboration and relative risk from a systematic review conducted in Western Europe and the US.
The analysis is novel, relevant, and the manuscript is well-written. I have a few suggestions: Comment 1: Epidemiological studies have shown distinct differences regarding the contribution of overweight and obesity in mortality among young adults, middle-aged, and elderly. For example, data shows that the impact of obesity on mortality diminish with increasing age. How the authors addressed those differences in their analysis?

Reply
Previous research has indeed shown that the association of BMI with mortality differs across ages; and, specifically, that the relative risk of obesity associated with mortality is higher at younger ages.
To account for this pattern, we deliberately selected age-(and sex-) specific relative risks of dying from obesity from a recent meta-analysis (see Table S1, Supplementary Material of the manuscript).
After applying linear regression to turn these wide age group RRs into single-year RRs (18-100), we effectively used RRs declining from 1.53 (age 18) to 1.43 (age 100) for women, and from 1.57 and 1.48 for men, respectively (see the Appendix of this document, Figure 1). Although more recent studies reported greater differences in RRs between age groups (Global BMI Mortality Collaboration et al., 2016), these studies did not, unfortunately, provide sex-specific RRs, which were important for our study as well.
We clarified this point in the text in page 6, lines 113-117: "These age-and sex-specific RRs were largely in line with the overall European RR of 1.64 recently estimated by the Global BMI Mortality Collaboration (Global BMI Mortality Collaboration et al., 2016). The differences across age groups found in that study were similar with those reported in our findings (i.e., higher RRs at younger than at older ages), though they were less distinct (Global BMI Mortality Collaboration, 2016)".
Comment 2: The effect of obesity on mortality is mediated through chronic diseases, such as cardiovascular disease. In the recent decade, there is an improvement in the treatment of cardiovascular disease, and more individuals are living with cardiovascular disease and contributing to increased life expectancy. Therefore, the contribution of obesity to mortality could be smaller due to better treatment of chronic diseases. I think authors should discuss in the discussion to explain the smaller estimates compared to the previous study (Preston & Stokes, 2011).

Reply
A decrease in cardiovascular mortality has indeed been observed in recent years, partly as a result of improvements in medical care (Mehta 2011). This development could indeed exert an influence on the association of obesity with mortality in terms of relative risks. Previous studies that assessed changes over time in the association of obesity with mortality did so only for the US, and, unfortunately, provided mixed evidence, with some of these studies reporting a decline (Flegal et al., 2005;Mehta & Chang, 2011;Yu, 2012), and others finding an increase (Yu, 2016 RRs, even though it is possible that changes in the association of obesity with mortalitywhich could, for example, occur because of improvements in the treatment of chronic diseaseshave affected the impact of obesity on life expectancy. Previous studies that assessed changes over time in the association of obesity with mortality did so only for the US, and, unfortunately, provided mixed evidence, with some of these studies reporting a decline (Flegal et al., 2005;Mehta & Chang, 2011;Yu, 2012), and others finding an increase (Yu, 2016). Therefore, before implementing time-variant European RRs, more information on their direction is required".
We do not believe that this issue can explain why the differences in our estimates were smaller than those in the estimates of Preston et al. (2011). While it is true that the two studies used different RR and prevalence data, if RRs had been the source of the differences between our estimates and Preston"s estimates, then the observed differences would have had the same direction for all countries -which was not the case (see the manuscript"s Supplementary Material, Table S4).
Instead, we believe that the observed differences are primarily attributable to the different prevalence data used. We emphasised this point in our comparison of the results, page 15, lines 300-302: "Given that the observed differences do not have the same direction for the different countries, we believe that these differences are mainly attributable to the prevalence data used". The method for the calculation of potential gains in life expectancy is a valid method. The validity and reliability of estimates depend on the data used in this study. More detailed description of data sources is needed.

Reply
We clarified our description of the data sources.

Reply
We used the data by country, and clarified this point in the text on page 6, lines 123-124: "All-cause mortality numbers and exposure population data by single year of age, sex, year, and country were obtained from the Human Mortality Database (16).
We also clarified the high quality of these data on page 6, lines 124-126: "These data are of high quality, and are widely used within the demographic community and beyond (Barbieri et al., 2015)".
Comment 2: Obesity prevalence estimates: What data were used for the estimation? The estimates were model based. Different models for different counties, or a single model for all counties? What are predictors for these estimates? Reply: The obesity prevalence estimates used in our study come from the NCD Risk Factor Collaboration study (NCD Risk Factor Collaboration, 2016). As inputs for the obesity estimates, measured height and weight data from representative data sources were used. In total, 1698 population-based measurement studies with 19.2 million participants were used. These data were entered into a Bayesian hierarchical model, which also included as explanatory variables national income, proportion of population living in urban areas, mean number of years of education, and summary measures of the availability of different food types for human consumption. This model was applied to all countries. We clarified this point in the data section, page 5-6, lines 102-109.
Although we already evaluated the use of these data in our discussion, we have now added table S3 in the Supplementary Material of the manuscript (see also Appendix Table 1 of the current reply), which gives the confidence intervals around the age-standardised prevalence estimates for each country by sex, in order to provide more information on the relative reliability of the data for the different countries in our analysis.
We also included a comment on these confidence intervals on page 13, lines 254-261: "For those countries with less available obesity dataespecially the CEE countriesa portion of the data we used were merely the result of modelling. Thus, the resulting estimates should be treated with some caution (NCD Risk Factor Collaboration, 2016). By contrast, for the non-CEE countries, most of the data we used pertained to measured data. Supplementary Material, Table S3 gives the confidence intervals around the age-standardised prevalence estimates for each country by sex in order to provide more information on the relative reliability of the data for the different countries in our analysis".
Comment 3: RR of dying from obesity: Is obese persons relative to not obese persons or obese persons relative normal weight persons? Many studies estimated relative risk (and sometimes, hazard ratio) of dying of obese persons relative to normal weight persons. If RR is obese persons relative to normal weight person, the estimates are not accurate. Please confirm that RR is relative to not obese persons.

Reply
In line with previous studies that estimated obesity-attributable mortality (Allison et al., 1999;Banegas et al., 2003;Flegal et al., 2005;Katzmarzyk & Ardern, 2004a), we used the RRs from a metaanalysis, which included studies that used the normal weight group (BMI 18.5-24.9 kg/m 2 ) or a narrower range of the normal weight group as a reference group (Lobstein T 2010). The estimation of obesity-attributable mortality with such a RR can be considered the theoretically maximally possible attributable mortality (GBD 2017Risk Factor Collaborators, 2018.
Instead of simply mentioning this point in the supplementary material, we have now clarified it in the main text as well, on page 6, lines 110-113.
Such an estimation of obesity-attributable mortality, using RRs with the normal weight category as a reference group, is in line with the hypothetical situation we employed when calculating the potential gains in life expectancy, in which we hypothesized the complete elimination of obesity-attributable mortality. Our PGLE estimation is in line with previous estimations (Preston et al. 2011).
We clarified this point on page 6, lines 115-122: "In addition, the use of RRs with the normal weight category as the reference category is in line with previous studies that estimated obesity-attributable mortality ( Allison et al., 1999;Banegas et al., 2003;Flegal et al., 2005;Katzmarzyk & Ardern, 2004b), while the estimation of obesity-attributable mortality with such a RR can be considered the theoretically maximally possible attributable mortality (GBD 2017Risk Factor Collaborators, 2018". Reply This is indeed a limitation of our study, as we briefly discussed in our evaluation of data and methods section (page 13), lines 262-268. We have extended our discussion of this limitation as follows: "Because ageand sex-specific RRs of mortality associated with obesity are not readily available by country and year, we have decided to apply to all of the countries studied age-and sex-specific RRs from Western European and US populations that are largely suitable for our setting, as has previously been done (Preston & Stokes, 2011

Reply
It is indeed the case that we did not provide standard error or confidence limits for our estimates. The main reason why we did not do so is that the relative risks we used were not accompanied by uncertainty estimates. As we had uncertainty estimates for obesity prevalence only, we were unable to perform a formal analysis that would have provided uncertainty estimates that captured the full level of uncertainty. We have now clarified this point this in the manuscript on page 14, lines 283-285: "The lack of information on the uncertainty of the RRs we used limited us in estimating confidence intervals for the OAMFs and PGLEs".
To give the reader some idea of the level of uncertainty, we now provide in Table S2  To provide some information on how the uncertainty of the prevalence estimates affected our PGLE estimates, we calculated the lower bound and the upper bound of the PGLE for 2012, for which only the levels of uncertainty of obesity prevalence estimations were taken into account (see Table 2 in the Appendix of this document). However, we decided not to include these estimates in our manuscript because they do not capture the full degree of uncertainty of the estimates, and could therefore be misinterpreted.