RETIREMENT RESEARCH January 2014, Number 14-2 HOW WILL MORE OBESITY AND LESS SMOKING AFFECT LIFE EXPECTANCY? By Samuel H. Preston, Andrew Stokes, Neil K. Mehta, and Bochen Cao* Introduction Obesity and Mortality Personal behaviors can have a major influence on how The methodology for the obesity analysis consists long people live. Two especially damaging behaviors of three steps. The first step estimates the current are smoking and the poor nutrition and exercise hab- impact of obesity on mortality rates. The second its that result in obesity. Estimates from the Centers step forecasts changes in obesity levels to 2040. The for Disease Control and Prevention suggest that, in third step applies the results from step one – obesity’s 2000, 15 percent of U.S. deaths were caused by obesity impact on mortality – to the results of step two – the and 18 percent by smoking.1 But obesity is on the future prevalence of obesity – to estimate how the rise while smoking is on the decline. The question is projected changes in obesity affect future mortality. whether the benefits from less smoking will outweigh The main data source is the National Health and the harm from rising obesity. This brief, based on Nutrition Examination Survey (NHANES), a nation- a recent study, projects how changes in obesity and ally representative survey conducted by the National smoking will impact life expectancy in 2040.2 Center for Health Statistics. The NHANES was con- The discussion proceeds as follows. The first two ducted periodically beginning in 1971 and has been sections describe the methodologies for estimating conducted annually since 1999. The survey includes the impact of obesity and smoking on mortality rates extensive medical data on individuals, collected by and for projecting how the prevalence of these behav- trained nurses. These data include current height iors will change over time. The third section presents and weight, which are used to calculate body mass in- the results, expressed as changes in future life expec- dex (BMI), the standard measure used in determining tancy. The final section concludes that, overall, the obesity. The survey also asks respondents to recall benefits of reduced smoking will trump the damage their weight at age 25. from increased obesity. However, the results differ by gender, with men showing a solid net gain, while women see only a small improvement. * Samuel H. Preston is a professor of sociology at the University of Pennsylvania and a research associate of the Univer- sity’s Population Studies Center. Andrew Stokes is a graduate student in demography at the University of Pennsylvania. Neil K. Mehta is a professor of public health at Emory University. Bochen Cao is a graduate student in demography at the University of Pennsylvania. 2 Center for Retirement Research Estimating Obesity’s Effect on Mortality between the 1990s and the 2000s. Therefore, data from the relatively stable 1998-2008 period were used Regression analysis is used to assess how obesity to generate the transition probabilities needed for the affects mortality. The analysis uses two different mea- projections. sures of obesity: 1) BMI at the date of the interview The transition probabilities were then applied to (“baseline BMI”); and 2) BMI at age 25.3 These mea- the 2010 baseline population to project changes in sures rely on NHANES survey data from 1988-1994 the prevalence of obesity over time. So, for example, and 1999-2002. The baseline BMI measure has four consider individuals who were age 50 in 2010. Their categories: Normal, Overweight, Obese I, and Obese obesity prevalence is projected every five years from II-III; for “age-25 BMI,” the two obese categories are age 50 to age 80 (2010 to 2040).6 This same process merged. In addition, due to strong evidence that the is used for all of the age groups in the 2010 baseline relative risk of death for obese individuals declines population. The combined results for all age groups with age, an interaction term for age and the two show that, by 2040, nearly half of the adult population obese categories is included.4 The basic equation is: will be obese, up from about 38 percent in 2010 (see Figure 1). Death rate = ƒ (baseline BMI category, age-25 BMI category, age, age-obesity interaction, sex, other demo- graphic factors) Figure 1. Actual and Projected Trends in Obesity Prevalence among U.S. Adult Population, As expected, both measures of obesity are related 1976-2040 to an increased risk of death, and the age-obesity 60% interactions indicate a decreasing mortality risk of obesity by age. These results, which provide the risk 50% of death for a given individual, are then applied to the projections of obesity prevalence for the sample 40% population, described below. 30% Projecting Changes in Obesity to 2040 20% Men 10% Both obesity measures – baseline obesity and obesity Women at age 25 – are projected to 2040. The projections 0% start with a sample of individuals age 25-84 in 2010. 1980 1990 2000 2010 2020 2030 2040 Over time, this initial sample changes as some mem- bers die and as, after each decade, a new cohort age Source: Preston et al. (2013). 25-34 is added. The procedure for projecting growth in the first obesity measure is as follows. Historical BMI data For the age-25 BMI measure, fewer projections are used to calculate the “transition probability” of are necessary because everyone in the 2010 baseline moving from non-obese to obese (or vice versa) dur- sample is already age 25 or over, so BMI can be calcu- ing different 10-year periods.5 For example, among lated directly from the NHANES interview responses. individuals with normal BMI in 1980, 67 percent were Projections of age-25 obesity are thus only needed for still in the normal category in 1990 while 30 percent the younger cohorts that are added to the sample over had moved into the overweight category and 3 percent time – those who are not yet age 25 in 2010. ended up as obese. So those starting out with normal The final step in the obesity analysis is to apply BMI had a 3-percent chance of becoming obese dur- the coefficients for the impact of obesity on mortality, ing the period. described above, to the projections of the prevalence The results of this analysis showed that the prob- of obesity.7 The results of this exercise determine ability of moving up to a heavier weight category rose how the projected change in obesity from 2010-2040 between the 1980s and the 1990s but then stabilized affects future mortality. Issue in Brief 3 Smoking and Mortality Projecting Changes in Smoking The methodology for the smoking analysis is broadly Since smoking behavior is only measured prior to similar to that for obesity – estimate the relationship age 40, only limited projections were needed for the between smoking and mortality; determine the future period 2010-2040. The reason is that, for anyone 40 prevalence of smoking behavior; and apply the results or older in the 2010 baseline population, the mea- of the current relationship to the prevalence of future sure was calculated directly from the NHIS inter- behavior. view responses. For cohorts younger than age 40 in Data on smoking come from the National Health 2010, the measure was projected based on activity at Interview Survey (NHIS), another nationally represen- younger ages. tative survey conducted by the National Center for Figure 2 shows the average number of years as a Health Statistics. This survey has been conducted an- smoker before age 40 by birth year. All of the cohorts nually since 1957; the data used in the analysis cover included in the analysis – those alive in 2010 – ap- 1965-2009.8 These data allow for an assessment of pear in the figure.12 Overall, smoking used to be birth cohorts stretching back to the late 19th century. much more prevalent among men than women, but Such a lengthy period is necessary due to the long it peaked with men born between about 1910-1920 lag between smoking and its mortality impact. For and then declined rapidly. Female smoking behavior example, smokers often do not die from lung cancer peaked much later, with those born around 1935- until decades after they pick up the habit. 1940, followed by a more gradual decline. The final The NHIS includes several questions on smok- step is to apply the results from the smoking-mortali- ing, including whether an individual is (or has been) ty analysis to the smoking prevalence data, by cohort, a smoker, when he started, and when he quit. These shown in the figure to estimate the effects of changes data were used to calculate the average number of in smoking behavior on mortality from 2010-2040. years spent as a smoker before age 40. Figure 2. Mean Number of Years Spent as a Estimating Smoking’s Effect on Mortality Cigarette Smoker before Age 40, for Birth Cohorts from 1885-1995 While the risk of death from smoking depends on several smoking-related behaviors, the death rate 20 from lung cancer is one clear indicator of the cu- Men mulative effects of smoking.9 Nearly 90 percent of Women 15 U.S. lung cancer deaths are related to smoking.10 Therefore, to assess smoking’s effect on mortality, the analysis starts with the relationship of smoking to 10 lung cancer deaths and then considers the effect on other types of deaths. 5 A regression equation is used to relate lung cancer mortality to age and average smoking behavior for each birth cohort. The basic equation is: 0 1890 1915 1940 1965 1990 Lung cancer death rate = ƒ (age, years as a smoker before age 40) Source: Preston et al. (2013). Armed with these results, the next step is to assess smoking’s influence on other causes of death, such as The Results heart disease. Here, the analysis relies on the histori- cal relationship between lung cancer deaths and all A common way to present mortality results is to other deaths. This relationship is assumed to remain translate them into life expectancy. This final step constant so that, as lung cancer deaths are projected produces a simple summary measure: changes in life to decline along with smoking, all other deaths are expectancy at age 40. For the purposes of this analy- reduced accordingly.11 Together, these results are an sis, the risks of death from obesity and smoking are indicator of the full effects of smoking on mortality. assumed to be independent of each other, so they are simply added together to produce the net result.13 4 Center for Retirement Research Figure 3 shows the impact of the changes in Conclusion obesity and smoking on life expectancy at age 40 in 2040.14 The first bar in each cluster is the net impact The two major behaviors that affect life expectancy on life expectancy, followed by the separate contribu- have been headed in opposite directions. Smoking tions of obesity and smoking. For men, the benefits rates have been dropping for decades while obesity of reduced smoking clearly trump rising obesity, with has been climbing. Over the next 30 years, the net a net gain of 0.8 years in life expectancy. For women, impact of these behaviors on life expectancy is esti- smoking and obesity roughly cancel each other out, mated to be positive, though this result is driven by with just a small net gain. The main reason for this men. discrepancy is that, compared to men, women see Both the gains from reduced smoking and the less of a decline in smoking during the projection losses from increased obesity are large compared period because their smoking behavior prior to age 40 to overall gains in life expectancy projected by other peaked later and declined less. Thus, while the effect researchers. For example, in 2005, the U.S. Social Se- of rising obesity is nearly the same for both men and curity Administration projected gains in life expectan- women, declines in smoking add 1.5 years to male cy at age 40 of about 2.6 years for men and 2.2 years life expectancy and just under 1 year to female life for women between 2010 and 2040.15 The projected expectancy. gains in life expectancy from smoking alone are equal to about half of this total gain, while obesity imposes a penalty equal to roughly a third of the gain. Given Figure 3. Projected Effect on Life Expectancy their prominence, both smoking and obesity merit at Age 40 of Changes in Smoking and Obesity continued monitoring and analysis. Between 2010 and 2040, in Years 2 Net effect of smoking and obesity 1.5 Effect of smoking alone Effect of obesity alone 1 0.8 0.9 0.1 0 -0.7 -0.8 -1 Men Women Source: Preston et al. (2013). Issue in Brief 5 Endnotes 1 Mokdad et al. (2004, 2005). 11 For details on the procedure used to connect lung cancer mortality to smoking-related mortality from all 2 Preston et al. (2013). other causes, see Preston, Glei, and Wilmoth (2011). For a discussion of the uncertainty analyses used to 3 For age-25 BMI, measured height at the date of assess the estimated impacts of smoking and obesity interview is used for individuals of all ages because on mortality, see Preston et al. (2013). self-reported height at age 25 was not available in the NHANES data from 1988-1994. 12 As with the obesity analysis, the individuals in the smoking sample are grouped into five-year age 4 Prospective Studies Collaboration (2009). cohorts. 5 In addition to current BMI and BMI at age 25, 13 See Preston et al. (2013) for a discussion of pos- the survey asks respondents to recall their weight sible interactions between mortality risks for obesity from 10 years prior to the survey. These responses and smoking. are combined with current height to estimate “recall BMI” for each respondent, and are adjusted to correct 14 These results differ somewhat from Stewart, Cut- for common reporting errors as discussed in Flegal et ler, and Rosen (2009), who forecast that the negative al. (1995). The combination of “corrected recall BMI” effects of obesity would outweigh the positive effects and current BMI is used as a data input for estimat- of reduced smoking during the 2005-2020 period. ing the probability of transition between different The main reason for the difference is the different BMI categories over a 10-year period. period of analysis; the results for 2010-2020 look more similar to Stewart, Cutler, and Rosen (2009). Another 6 This example just uses a single age for simplicity. reason is that the results show a smaller role for obe- For the analysis, five-year age groups were used. sity than Stewart, Cutler, and Rosen (2009), probably due to a slower increase in obesity and a lower associ- 7 Specifically, the coefficients are multiplied by the ated mortality risk. percentage of the total sample population projected to be obese and the percentage obese at age 25. 15 Bell and Miller (2005). 8 Data on smoking by cohort are based on Burns et al. (1998); which used 15 NHIS surveys from the 1965-1991 period. David Burns also supplied unpub- lished estimates through 2001. This series was then further updated to 2009. 9 These behaviors include the number of cigarettes smoked per day, the degree of inhalation, and the tar content of the cigarette. 10 This figure comes from Oza et al. (2011). The analysis in this brief obtains historical data on lung cancer deaths from several sources: Vital Statistics of the United States, the World Health Organization/ International Agency for Research on Cancer, and the Centers for Disease Control and Prevention. 6 Center for Retirement Research References Bell, Felicitie C. and Michael L. Miller. 2005. “Life Preston, Samuel, Dana Glei and John Wilmoth. 2011. Tables for the United States Social Security Area “Contribution of Smoking to International Differ- 1900-2100.” Actuarial Study No. 120. Washington, ences in Life Expectancy.” In International Differ- DC: U.S. Social Security Administration. ences in Mortality at Older Ages: Dimensions and Sources, eds. Eileen Crimmins, Samuel Preston, Burns, David M., Lora Lee, Larry Z. Shen, Elizabeth and Barney Cohen. Washington, DC: National Gilpin, H. Dennis Tolley, Jerry Vaughn, and Academy Press. Thomas G. Shanks. 1998. “Cigarette Smoking Behavior in the United States.” In Changes in Prospective Studies Collaboration. 2009. “Body-Mass Cigarette-Related Disease Risks and Their Implication Index and Cause-Specific Mortality in 900,000 for Prevention and Control (Smoking and Tobacco Adults: Collaborative Analyses of 57 Prospective Control Monograph No. 8), eds. D. M. Burns, L. Studies.” Lancet 373: 1083-96. Garfinkel, and J. Samet. Bethesda, MD: Cancer Control and Population Sciences, National Cancer Stewart, Susan T., David M. Cutler, and Allison B. Institute, U.S. National Institutes of Health. Rosen. 2009. “Forecasting the Effects of Obesity and Smoking on U.S. Life Expectancy.” The New Flegal, K. M., Richard Trioano, Elsie Pamuk, Robert England Journal of Medicine 361(23): 2252-60. Kuczmarski, and Stephen Campbell. 1995. “The Influence of Smoking Cessation on the Prevalence of Overweight in the United States.” The New Eng- land Journal of Medicine 333(18): 1165-70. Mokdad, Ali. H., James S. Marks, Donna F. Stroup and Julia L. Gerberding. 2004. “Actual Causes of Death in the United States, 2000.” Journal of the American Medical Association 291: 1238-45. Mokdad, Ali. H., James S. Marks, Donna F. Stroup and Julia L. Gerberding, 2005. “Correction: Actual Causes of Death in the United States.” Journal of the American Medical Association 293(3): 293-4. Oza, Shefali, Michael J. Thun, S. Jane Henley, Alan D. Lopez, and Majid Ezzati. 2011. “How Many Deaths Are Attributable to Smoking in the United States? Comparison of Methods for Estimating Smoking- Attributable Mortality When Smoking Prevalence Changes.” Preventive Medicine 52(2011): 428-433. Preston, Samuel H., Andrew Stokes, Neil K. Mehta, and Bochen Cao. 2013. “Projecting the Effect of Changes in Smoking and Obesity on Future Life Expectancy in the United States.” Demography. Available at: http://bit.ly/1l7tuXo. RETIREMENT RESEARCH About the Center Affiliated Institutions The mission of the Center for Retirement Research The Brookings Institution at Boston College is to produce first-class research Massachusetts Institute of Technology and educational tools and forge a strong link between Syracuse University the academic community and decision-makers in the Urban Institute public and private sectors around an issue of criti- cal importance to the nation’s future. To achieve this mission, the Center sponsors a wide variety of Contact Information Center for Retirement Research research projects, transmits new findings to a broad Boston College audience, trains new scholars, and broadens access to Hovey House valuable data sources. Since its inception in 1998, the 140 Commonwealth Avenue Center has established a reputation as an authorita- Chestnut Hill, MA 02467-3808 tive source of information on all major aspects of the Phone: (617) 552-1762 retirement income debate. Fax: (617) 552-0191 E-mail: crr@bc.edu Website: http://crr.bc.edu © 2014, by Trustees of Boston College, Center for Retire- The research reported herein was performed pursuant to ment Research. All rights reserved. Short sections of text, a grant from the U.S. Social Security Administration (SSA) not to exceed two paragraphs, may be quoted without ex- funded as part of the Retirement Research Consortium. The plicit permission provided that the authors are identified and opinions and conclusions expressed are solely those of the full credit, including copyright notice, is given to Trustees of authors and do not represent the opinions or policy of SSA, Boston College, Center for Retirement Research. any agency of the federal government, or the Center for Retirement Research at Boston College.