Obesity and Life Expectancy

Definition

Obesity — defined as body mass index (BMI) ≥ 30 — raises mortality risk across all age groups and interacts with other behavioral risk factors (especially smoking) to shape aggregate life expectancy (LE) trends. Rising obesity rates in the United States have become a countervailing force against LE gains from declining smoking, with the net effect being a drag on future LE growth. Stewart, Cutler, and Rosen (2009) quantified the joint effect: for a representative 18-year-old, continuing BMI (+0.5%/year) and smoking (−1.4%/year) trends through 2020 produce a net −0.71 LE years and −0.91 quality-adjusted life expectancy (QALE) years relative to a no-trend-change baseline, with the BMI drag (−1.02) roughly 3× larger than the smoking benefit (+0.31).

Key Ideas

• Epidemiological base: Cox proportional-hazards (PH) models on National Health and Nutrition Examination Survey (NHANES) I–III follow-up data (N=24,758) across 16 joint smoking×BMI categories generate category-specific relative mortality risks. These risks are held fixed while population shares are shifted via demographic projection.
• Simultaneous and offsetting: Declining smoking has a real positive LE effect (+0.31 years if isolated), but the obesity drag (−1.02 years if isolated) more than triples it. The two trends cannot be assessed independently because the population's joint distribution across 16 risk categories changes nonlinearly.
• Obesity projections: Extrapolating 15-year historical trends (NHANES 1971–2006), ~45% of U.S. adults will be obese by 2020, up from ~30% in 2005. This is a demographic forecast from trend extrapolation, not a causal model of the drivers of obesity.
• Quality of life: Using Medical Expenditure Panel Survey (MEPS) 2003 EQ-5D data (N=18,913), quality-of-life weights are estimated by BMI-smoking category via ordinary least squares (OLS) regression. The morbid obesity–smoking quality penalty at age 65+ is ~0.17 EQ-5D points (0.61 vs. 0.78 for normal-weight never-smokers). Because obesity degrades quality of life above and beyond mortality, the QALE drag (−0.91 years) exceeds the LE drag (−0.71 years).
• Threshold finding: The adverse BMI effect on LE persists as long as annual BMI growth exceeds 0.15%/year. Given the observed 0.5%/year trend, essentially any positive BMI growth produces a net LE drag — the scenario where obesity's growth rate is too slow to matter is empirically implausible without a near-complete halt in the trend.
• Full elimination counterfactual: If all adults became nonsmokers of normal weight by 2020, the LE gain for a representative 18-year-old would be +3.76 years (smoking: +1.73; BMI: +1.40) and QALE gain +5.16 years (smoking: +2.17; BMI: +2.44). The QALE gain exceeds the LE gain because eliminating obesity and smoking restores quality of life for years that are already being lived, not just years that would have been lost to early death.
• Historical validation (1990–2004): Actual LE gain was 2.44 years; counterfactual holding 1990 risk-factor distribution fixed would have yielded 2.98 years. Adverse obesity and smoking trends cost ~0.54 years over that period, but did not reverse LE growth. The 2005–2020 base case forecasts a net decline (−0.71 years) relative to no-change, meaning the trend has worsened enough to turn the marginal effect negative.
• Precursor context: Olshansky et al. (2005, NEJM) was the first major projection warning that obesity could reduce 21st-century LE in the United States below 20th-century levels. Stewart et al. (2009) extends this by explicitly modeling the smoking offset, establishing a joint decomposition, and providing historical validation.

How It Works

Life-table simulation method:

1. Project 2005→2020 population shares across 16 joint categories (4 BMI: normal/overweight/obese/morbidly obese × 4 smoking: never/former/current light/current heavy) by extrapolating NHANES (BMI, four waves 1971–2006) and National Health Interview Survey (NHIS) (smoking, four waves 1978–2006) 15-year historical trends.
2. Estimate age-sex-specific mortality hazard ratios for each of the 16 categories using Cox PH models on NHANES I–III mortality follow-up (N=24,758), with normal-weight never-smoker as reference category.
3. Apply category-specific hazard ratios to the 2004 U.S. period life tables (published by CDC) to generate category-specific life expectancy and disability-adjusted life tables.
4. Compute population-weighted average LE/QALE by applying the projected category shares at each future year, yielding a single "representative 18-year-old" forecast.
5. Quality weights from MEPS 2003 EQ-5D (0–1 scale) are estimated by BMI-smoking category via age-sex-adjusted OLS and applied age-specifically to the life table to generate QALE.

Counterfactuals: Three separate scenarios are modeled — (a) only smoking changes, (b) only BMI changes, (c) both change — to isolate the individual contributions before combining.

Why It Matters

• LE growth not guaranteed: Unlike prior decades when smoking declines were large enough to dominate LE improvement, the shift in risk-factor composition — obesity rising, smoking declining at a decelerating rate — means the net effect on LE may be negative for the first time in the post-World-War-II era.
• SSA mortality forecasting: The Social Security Administration's (SSA) historical actuarial projections (Office of the Chief Actuary, OACT) relied on 70 subjectively chosen "ultimate rates of mortality decline" that did not explicitly incorporate behavioral risk-factor trends. Stewart et al. (2009) demonstrates why this approach is vulnerable to systematic overoptimism: if obesity continues rising, any forecast that ignores it overstates LE gains. Soneji and King (2012) showed this directly, finding the Bayesian alternative that includes smoking and obesity covariates implies substantially lower SSA trust fund solvency. See SSA Mortality Forecasting.
• Policy prioritization: Because the obesity LE drag (~3×) dominates the smoking benefit, interventions that reduce obesity have substantially higher returns than further smoking reduction at this point in the epidemiologic transition.
• Connection to compression of morbidity: Obesity expands morbidity at working ages (Lakdawalla, Bhattacharya, and Goldman 2004 show obesity accounts for ~50% of disability growth in ages 18–29 from 1984–2000), directly opposing the compression of morbidity trend observed in the elderly. The net effect of rising obesity is both shorter LE and more years of morbidity per life lived. See Compression of Morbidity.
• Socioeconomic status (SES) gradient: The paper does not model SES heterogeneity, but obesity rates have risen disproportionately among lower-income and less-educated Americans. This means the aggregate LE drag masks a steeper drag for disadvantaged populations, reinforcing the Education-Mortality Gradient and Income-Mortality Gradient.

Open Questions

• Has the obesity trend decelerated since 2009, and do revised BMI projections change the LE forecast?
• Does the threshold of 0.15%/year BMI growth vary across age/sex/race subgroups in ways that matter for aggregate LE?
• What share of the QALE drag is attributable to obesity's direct mortality effect vs. its morbidity/quality-of-life effect?
• Does eliminating the two risk factors additively produce +3.76 LE years, or are there interaction effects between obesity and smoking that make joint elimination more or less than the sum of parts?

Related

• Compression of Morbidity
• Disability-Free Life Expectancy
• Disability-Adjusted Life Year
• Obesity and Non-Employment
• SSA Mortality Forecasting
• SES Health Behavior Gradient
• Period Mortality
• Education-Mortality Gradient
• Income-Mortality Gradient
• S. Jay Olshansky
• Susan Stewart
• David Cutler
• Allison B. Rosen

Sources

• Stewart, Cutler, and Rosen 2009 — Forecasting the Effects of Obesity and Smoking on U.S. Life Expectancy