Obesity and early mortality in the United States

Authors

  • James A. Greenberg

    Corresponding author
    1. Department of Health and Nutrition Sciences, Brooklyn College of the City University of New York, New York Obesity Research Center, New York, New York, USA
    2. Department of Medicine, St. Luke's-Roosevelt Hospital Center, New York, New York, USA
    • Department of Health and Nutrition Sciences, Brooklyn College of the City University of New York, New York Obesity Research Center, New York, New York, USA
    Search for more papers by this author

  • Disclosure: The author declared no conflict of interest.

Abstract

Objective:

Although obesity is a serious public health problem, there are few reliable measures of its health hazards in the United States. The objective of this study was to estimate how much earlier mortality is likely to occur for Americans who are obese (body mass index [BMI], ≥ 30).

Design and Methods:

Data from the National Health and Nutrition Examination Survey (NHANES) I (1971–1975), NHANES II (1976–1980), and NHANES III (1988–1994) for 37,632 participants who experienced 8,791 deaths during 15 years of follow-up were prospectively analyzed. The relative risk of death from all causes and its advancement period, adjusted for covariates, were calculated. Stratification was used to investigate the effects of pre-existing illness, smoking, and older age on the advancement period.

Results:

Compared to the participants of reference weight (BMI, 23 to <25 kg/m2), mortality was likely to occur 9.44 years (95% confidence interval [CI]: 0.72, 18.16) earlier for those who were obese (BMI, ≥ 30). For overweight (25 to <30 kg/m2), grade 1 obesity (BMI, 30 to <35) and grades 2–3 obesity (BMI, ≥ 35.0), the mortality was likely to occur earlier by 4.40 (−3.90, 12.70), 6.69 (−2.06, 15.43), and 14.16 (3.35, 24.97) years, respectively. These estimates apply to healthy nonsmoker young- and middle-aged (21–55 years) adults, who constituted an estimated 32.8% of Americans with age of >21 years between 1988 and 1994. Without stratifying simultaneously for preexisting illness, smoking, and age, values of the advancement period for obesity were markedly smaller than those observed for healthy nonsmoker young and middle-aged adults.

Conclusions:

For healthy nonsmokers young- and middle-aged adults who constitute about one-third of American adults, being obese is likely to hasten mortality by 9.44 years.

Introduction

There is evidence suggesting that obesity is a major public-health problem in the United States (1, 2) and around the world (3). However, there is uncertainty as to the magnitude and seriousness of the health consequences of obesity. For instance, the estimates of mortality attributable to obesity in the US population annually range from 111,909 by Flegal et al. (4) to 216,000 by Danaei et al. (5), 280,000–325,000 by Allison et al. (6) and 365,000 by Mokdad et al. (7, 8). The institute of Medicine of the National Academy of Sciences (9) and the CDC (10) have both sponsored workshops aimed at finding appropriate methods for estimating the health hazards of overweight and obesity.

The advancement period for the relative risk of mortality (advancement period) provides a readily understandable, intuitively compelling measure of the mortality hazards of a risk factor such as obesity (11). The advancement period for obesity is a measure of how much earlier a given risk of mortality is reached among obese than among reference-weight individuals. For example, an advancement period of 10 years for obesity means that an obese person is likely to die 10 years earlier than a person of reference weight. The advancement period is similar to the years of lost life (YLL), but YLL is calculated as the difference between an individual's actual and expected lifespan. The advancement period is based on fewer and more realistic assumptions about the effects of body weight on mortality risk than YLL (11, 12). For instance, calculating YLL owing to obesity requires devising assumptions to estimate the life expectancy of all participants under consideration and then assuming that those who are not obese survive till their estimated life expectancy. Estimates of the advancement period can be derived directly from relative risks calculated from regression-model survival analyses.

Stevens et al. (12) made estimates of the advancement period for the relative risk of mortality for obesity in the large prospective Cancer Prevention Study-I cohort. Estimates of the relative risk advancement period for obesity have not previously been made for the US population.

The objective of this study was to fill this gap by using prospective data from three mortality-linked National Health and Nutrition Examination Survey (NHANES) surveys.

Methods and Procedures

NHANES I, II, and III cohorts and participants

There were 14,407 adults of 21–75 years in the 1971–1975 NHANES-I survey, 9,252 adults of 31–75 years in the 1976–1980 NHANES II survey, and 16,219 adults of 21–90 years in the 1988–1994 NHANES III survey.

Survival analyses

Cox proportional-hazards regression (13) analyses were conducted to estimate multivariate-adjusted relative risks for all-cause mortality in different categories of body mass index (BMI in kg/m2). The proportionality assumption was checked by means of scaled Schoenfeld residuals (14), and stratified baseline-function analyses were used to assess the effects for covariates that appeared not to have satisfied the proportionality assumption (15). Model fit was assessed by means of the log likelihood test. Time to event was days between the physical exam and the death or censorship. To avoid potential bias owing to different lengths of follow-up (16–18) in the three NHANES cohorts, the longest follow-up period common to all three cohorts, 15 years, was adopted in primary survival analyses. A sensitivity analysis was conducted with a follow-up period of 10 years.

Stratified analysis to assess interaction effects of illness, smoking, and age

Prior investigators have reported evidence, suggesting that the association between mortality risk and BMI, and therefore the advancement period, could be different in different strata of health status, smoking status, and age. Manson et al. (19), for instance, has shown that the relative risk of mortality for obesity is greater for nonsmokers than smokers; and for healthy participants than those with a history of serious illness. These effects have been attributed to the ability of smoking and illness to simultaneously decrease body weight and increase mortality risk (19). The relative risk of mortality for obesity has been found to decrease with increasing age among older participants in NHANES (17) and other (20) cohorts. Also, Stevens et al. (12) found that the advancement period for obesity decreased with increasing age. Therefore, tests were conducted to assess the existence of interaction effects for illness, smoking, and age. The forward-inclusion and backward-elimination methods described by Kleinbaum et al. (21) were used to assess interaction effects. There were significant third-order interactions for BMI by illness status by age, and BMI by smoking status by age, BMI by illness status by age, and BMI by smoking status by age in the NHANES I, II, and III data. A stratified analysis was conducted to investigate the effect modification shown by the significant interaction effects involving illness, smoking, and age. Results for healthy and nonsmoker participants were given priority here because of potential confounding in the strata containing smokers or ill participants (19) and because of the greater public-health importance of nonsmokers and healthy persons owing to their greater numbers in the population than smokers and ill persons, respectively. No subgroups with fewer than 100 events were included in the primary analyses.

Ill participants were defined as those who reported having serious pre-existing illnesses that included heart attacks, stroke, and cancer. Nonsmokers were defined as never smokers. Survival analyses were conducted in three different age levels ≤55, 55–65, and >65 years. These levels were selected because compared to BMI for censored participants, BMI for deceased participants was higher for participants of <55 years, and was lower and decreased more rapidly with increasing age for >65 years (Figure 1). This pattern suggested a different relationship between relative risk and BMI, and therefore a different advancement period, in each of the three age levels (18).

Figure 1.

BMI by age group and survival status in NHANES I, II, and II cohorts.

Outcome variable

All-cause mortality was the outcome event. Deaths were identified through systematic computerized matching of data from the NHANES surveys and the National Death Index. The sensitivity of the National Death Index for epidemiologic purposes has been found to range from 87.0 to 97.9% (22).

Predictor and covariates

BMI, calculated from measured body weight and height, was the predictor. Covariates were as follows: age in years; gender (male/female); race (White/non-White); alcohol consumption (never; <weekly, weekly, and daily); history of serious illness (yes, no); smoking (never, former, and current), and NHANES cohort (I, II, and III). Covariates in the causal pathway between BMI and mortality, such as blood pressure, were not included. Sensitivity analyses with extra covariates were conducted, as described below.

After excluding the three NHANES II participants who were lost to follow-up, the number of missing values in the combined cohorts were as follows: BMI, 56 (0.14%); smoking status, 1,859 (4.66%); illness status, 230 (0.58%); and alcohol consumption, 123 (0.31%). A sensitivity analysis was conducted to assess the effects of the missing smoking data. After exclusions for missing values, there were 37,632 participants and 8,791 deaths during the 15-year follow-up available for analysis.

BMI reference categories

The following BMI categories (2, 23) were used: overweight, 25 to <30 kg/m2; obesity, ≥30 kg/m2; grade 1 obesity, 30 to <35 kg/m2; and grades 2–3 obesity ≥35 kg/m2. Separate analyses were conducted for four different reference-weight BMI referent categories, 23 to <25, 18.5 to <23, 18.5 to <25, and 21 to <25 kg/m2. The WHO defines 18.5 to <25, 18.5 to <23, and 23 to <25 kg/m2 as the normal, lower-normal, and upper-normal categories, respectively. Flegal et al. (4) found that the Cox-regression relative risks for obesity were largest for BMI referent category 23 to <25 kg/m2, intermediate for 21 to <25 kg/m2, and smallest for 18.5 to <25 kg/m2 in the NHANES I, II, and III data. Fontaine et al. (24) found that the 23 to <25 kg/m2 referent category was associated with the least YLL for obesity for NHANES I, II, and III participants. These findings reflect the fact that in these NHANES data the nadir of the relationship between relative risk and BMI is between 23 and 25 kg/m2 and that relative risk increases with decreasing BMI below 23 kg/m2. In this analysis, the relative risk for obesity was largest for BMI referent category 23 to <25 kg/m2 and decreased as the mean BMI in the reference-weight category decreased among healthy nonsmokers with age of ≦55 years, as shown in the first paragraph of the Secondary Analyses section This finding could be owing to the fact that even in the stratum of healthy nonsmokers with age of ≦55 years there was residual confounding that increased relative mortality risk below 23 kg/m2. To emphasize the results least likely to be affected by such possible residual confounding, the relative risk and advancement period data summarized in Tables 2 and 3 are for BMI referent category 23 to <25 kg/m2. Selected secondary analysis results are presented in the text for 21 to <25, 18.5 to <25, and 18.5 to <23 kg/m2.

Formula for the mortality risk advancement period

The formula of Brenner et al. (11) was used to calculate the relative risk advancement period for obesity and overweight, and the Delta method was used to calculate 95% CIs (Figure 2). The assumptions that mortality rate increases monotonically with age and that there be no competing risks were satisfied.

Figure 2.

Mortality risk advancement period formula.

Adjustments for NHANES complex sampling

All analyses were performed with SAS version 9.2 and SAS-Callable SUDAAN version 10.0.1. Sample weights provided by NHANES researchers were included to account for complex sampling and for nonresponses. The acquisition and processing of the NHANES data were approved by the Institutional Review Board of Brooklyn College. This manuscript follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) recommendations (25).

Results

Baseline characteristics

As BMI increased, the proportion of minorities, mean age, and the prevalence of heart attack and stroke increased, and smoking and alcohol consumption rates decreased (Table 1). The proportion of females was highest among obese participants.

Table 1. Baseline characteristics in NHANES I, II, and III cohorts, by BMI categorya
 Ideal weight (N = 6,311)bOverweight (N = 12,999)Obese (N = 7,875)Total (N = 37,632)
  • a

    BMI categories in kg/m2: ideal weight: 23 to 25; overweight: 25 to 30; Obese: ≥30.

  • b

    Unweighted sample size after exclusions for missing data.

  • c

    Numbers in parentheses are standard error of the mean. Age showed significant variations across BMI categories, by t-test.

  • Chi-square tests of differences in prevalence across BMI categories for categorical variables were all significant.

BMI, mean (kg/m2)24.00 (0.01)c27.15 (0.02)34.38 (0.08)26.03 (0.06)
Age, mean (years)46.27 (0.37)48.53 (0.23)48.73 (0.28)46.68 (0.22)
Female (%)47.440.559.252.3
Non-White (%)11.611.917.012.7
History of heart attack (%)3.64.25.13.7
History of stroke (%)1.81.92.11.7
History of cancer (%)3.32.73.53.2
Smoking (%)    
Never40.641.447.941.9
Former24.927.225.523.5
Current34.531.426.634.6
Alcohol (%)    
Never29.032.844.833.5
Weekly27.126.826.527.9
Weekly30.527.620.626.6
Daily13.412.88.212.0

Effects of illness, smoking, and age on the relative risk advancement period

For all participants, the advancement period for the relative risk of mortality for obesity was 3.27 years (95% CI: 1.99, 4.56), prior to considering the significant interactions involving pre-existing illness, smoking, or age (Table 3). Stratifying by illness and smoking and limiting focus to healthy participants or nonsmokers did not substantially change the estimate. The advancement periods were 5.5, 8.0, and 12.8% higher for nonsmokers, healthy participants, and healthy nonsmokers, respectively. Stratifying for age and limiting focus to age of ≦55 years increased the estimate by 78% to 5.81 (2.81, 8.81) years. Stratifying for all three factors simultaneously and limiting focus to healthy nonsmokers age of ≦55 years increased the advancement period by 189% to 9.44 (0.72, 18.16) years.

The advancement period for healthy nonsmokers who were ≦55 years old increased with increasing stages of obesity: for grade 1 obesity (BMI, 30 to <35 kg/m2) and grades 2–3 obesity (BMI, ≥35 kg/m2), the equivalent results were 6.69 (−2.06, 15.43) and 14.16 (3.35, 24.97) years, respectively.

After stratifying for smoking and illness and limiting focus to healthy nonsmokers, the relative risk advancement period for obesity decreased with increasing age of above 55 years (Table 3). For ages of ≦55, 55 to <65, 65 to <75, and >75 years, the advancement periods were 9.40, 6.64, 3.10, and −1.11 years, respectively.

The relative risks for overweight (Table 2) and the advancement periods for overweight (Table 3) showed similar patterns to those for obesity. For overweight, the only significant results were for age of >75 years: a relative risk significantly lower than 1.00 (Table 2) and an advancement period significantly lower than 0 (Table 3).

Table 2. Relative risk of all-cause mortality in subgroups of NHANES I, II, and III participantsa
 Overweight (BMI: 25 to 30 kg/m2)Obesity (BMI: ≥30 kg/m2)
 NDeathsRelative risk (95% CI)NDeathsRelative risk (95% CI)
  1. a

    The relative risk of all-cause mortality was calculated from the multivariate-adjusted relative risk of mortality determined by means of Cox Regression analyses using data from the mortality-linked NHANES I, II, and III cohorts for 37,632 participants who experienced 8,791 deaths during 15 years of follow-up. The following covariates were in the model: age in years; gender (male/female); race (White/non-White); alcohol consumption (never; weekly, weekly, daily); history of serious illness (yes, no); smoking (never, former, current); and NHANES cohort (I, II, and III).

All participants12,9993,1150.98 (0.88, 1.08)7,8751,9161.36 (1.20, 1.53)
Nonsmokers (never smokers)5,9501,2910.94 (0.79, 1.10)4,2179391.40 (1.17, 1.67)
Healthy participants11,6462,3290.97 (0.86, 1.09)7,0301,4711.40 (1.23, 1.59)
Healthy nonsmokers5,4069930.92 (0.79, 1.07)3,8137431.44 (1.19, 1.76)
Adults age ≤55 years6,8414041.02 (0.74, 1.39)4,4313501.58 (1.26, 1.97)
Healthy age ≤55 years6,6263681.05 (0.74, 1.48)4,2433131.61 (1.27, 2.04)
Nonsmoker age ≤55 years2,9711201.29 (0.77, 2.16)2,2831241.78 (1.07, 2.96)
Healthy nonsmoker age ≤55 years2,8951121.33 (0.78, 2.26)2,2121151.84 (1.08, 3.15)
Healthy nonsmoker age 55 to ≤65 years8801300.84 (0.49, 1.46)6401401.91 (1.14, 3.21)
Healthy nonsmoker age 65 to ≤75 years1,3225070.92 (0.74, 1.16)8223741.52 (1.15, 2.00)
Healthy nonsmoker age >75 years3092440.63 (0.46, 0.85)1391140.89 (0.60, 1.32)

Secondary analyses

When calculated using different reference-weight BMI categories, the advancement period and relative risk for both obesity and overweight showed the same patterns as summarized in Table 2. Both measures of relative risk of obesity were largest for the participants who were 23 to <25 years of age and decreased as the mean BMI in the reference-weight category decreased. For instance, the advancement periods for obesity for healthy nonsmokers of ≦55 years were 9.44 (0.72, 18.16), 8.46 (1.32, 15.60), 7.80 (1.39, 14.20), and 6.54 (0.60, 12.47) years for the 23 to <25, 21 to <25, and 18.5 to <25 and 18.5 to <23 kg/m2 referent category, respectively.

The analyses for Tables 2 and 3 were repeated with a follow-up period of 10 years. The results showed the same patterns as summarized in Tables 2 and 3, with larger 95% CIs. For instance, the advancement periods for obesity for the 10-year and 15-year follow-up, respectively, were 3.35 (1.42, 5.28) and 3.53 (2.20, 4.86) years, respectively, for healthy participants, and 8.96 (−2.05, 19.97) and 9.44 (0.72, 18.16) years, respectively, for healthy nonsmoker who were ≦55 years old.

Table 3. Relative risk advancement period in subgroups of NHANES I, II, and III participantsa
 Overweight (BMI: 25 to 30 kg/m2)Obesity (BMI: ≥30 kg/m2)
 NDeathsRelative risk advancement period (95% CI)NDeathsRelative risk advancement period (95% CI)
  1. a

    The relative risk of all-cause mortality advancement period was calculated from the multivariate-adjusted relative risk of mortality determined by means of Cox Regression analyses using data from the mortality-linked NHANES I, II, and III cohorts for 37,632 participants who experienced 8,791 deaths during 15 years of follow-up. The following covariates were in the model: age in years; gender (male/female); race (White/non-White); alcohol consumption (never; weekly, weekly, daily); history of serious illness (yes, no); smoking (never, former, current) and NHANES cohort (I, II, and III). The formulae in Figure 2 were used to calculate the advancement period.

All participants12,9993,115−0.22 (−1.30, 0.86)7,8751,9163.27 (1.99, 4.56)
Nonsmokers (never smokers)5,9501,291−0.67 (−2.36, 1.02)4,2179393.45 (1.60, 5.30)
Healthy participants11,6462,329−0.31 (−1.54, 0.91)7,0301,4713.53 (2.20, 4.86)
Healthy nonsmokers5,406993−0.84 (−2.39, 0.71)3,8137433.69 (1.73, 5.66)
Adults age ≤55 years6,8414040.23 (−3.76, 4.22)4,4313505.81 (2.81, 8.81)
Healthy age ≤55 years6,6263680.61 (−3.81, 5.03)4,2433136.11 (2.93, 9.28)
Nonsmoker age ≤ 55 years2,9711203.90 (−4.01, 11.82)2,2831248.83 (0.63, 17.02)
Healthy nonsmoker age ≤55 years2,8951124.40 (−3.90, 12.70)2,2121159.44 (0.72, 18.16)
Healthy nonsmoker age 55 to ≤<65 years880130−1.73 (−7.35, 3.89)6401406.64 (0.60, 12.70)
Healthy nonsmoker age 65 to ≤75 years1,322507−0.60 (−2.29, 1.09)8223743.10 (0.95, 5.26)
Healthy nonsmoker age >75 years309244−4.53 (−7.61, −1.45)139114−1.11 (−4.92, 2.71)

The analyses for Tables 2 and 3 were repeated separately for males and females. The results were similar and showed the same patterns as those summarized in Tables 2 and 3. The relative risks and advancement periods had larger 95% CIs, and were higher for males in some subgroups and higher for females in others. For instance, the advancement periods for obesity were 3.62 (1.99, 5.27) and 3.36 (1.55, 5.16) years for healthy males and healthy females, respectively; and 6.02 (1.78, 10.25) and 6.49 (1.75, 11.24) years for males and females who were healthy and of ≦55 years, respectively.

The analyses for Tables 2 and 3 were repeated with the following extra covariates in the model: physical activity (% maximum); marital status (married, single, widowed, separated/divorced), and educational level (continuous, 0–17 completed years of school). The analyses for Tables 2 and 3 were also repeated after excluding all 5,689 NHANES I participants for whom smoking status was assessed retrospectively in 1982–1984 to determine the effects of this retrospective assessment. The results of both of these secondary analyses were essentially the same as those summarized in Tables 2 and 3.

It has been suggested that CIs for the advancement period based on Fieller's theorem (26) would be more accurate than those presented here, which are based on the Delta method. The analyses in Table 3 were repeated using the Fieller-theorem formula. The results were very similar to those summarized in Table 3. The Fieller-theorem 95% CIs were slightly larger and further from the null of zero. For instance, the Fieller-theorem 95% CIs for the advancement period for obesity were (1.99, 4.58), (1.17, 19.57), and (−5.10, 2.69) for all participants, healthy nonsmokers with age of ≦55 years and healthy nonsmokers with age of >75 years, respectively.

Although the focus of this analysis was obesity and overweight, results were also obtained for underweight. There is great uncertainty as to the causal factors underlying the magnitude of the relative mortality risk in the underweight category, so that such relative risks should be interpreted with caution. The following relative mortality risks for underweight, based on the methods described in Table 2, are presented here as follows: 1.18 (1.06, 1.32), 1.15 (0.96, 1.37), 1.18 (1.04, 1.34), 1.20 (0.90, 1.64), 1.21 (0.73, 2.01), and 0.79 (0.57, 1.09) for all participants, nonsmokers, healthy participants, participants with age of ≦55 years, healthy nonsmokers with age of ≦55 years, and healthy nonsmokers with age of >75 years, respectively.

Discussion

Compared to Americans of reference weight (BMI, 23 to <25 kg/m2), mortality was likely to occur 9.44 years (95% CI: 0.72, 18.16) earlier for those who were obese (BMI, ≥30 kg/m2). For overweight (25 to <30 kg/m2), grade 1 obesity (BMI, 30 to <35 kg/m2), and grades 2–3 obesity (BMI, ≥35.0 kg/m2), mortality was likely to occur earlier by 4.40 (−3.90, 12.70), 6.69 (−2.06, 15.43), and 14.16 (3.35, 24.97) years, respectively. These estimates were derived from a prospective 15-year survival analysis in NHANES data, and apply to American nonsmokers of ≦55 years without pre-existing serious illness. These findings are of considerable consequence. About one-third of Americans were obese and two-thirds were overweight during the period 2007–2008 (1). Also, Americans to whom the estimates apply, healthy nonsmokers of ≦55 years, constituted an estimated 32.8% of Americans older than 21 years of age at the time of the NHANES III survey (1988–1994); and they probably represent the majority of working Americans today. An advancement period of 9.44 years for Americans of age ≦55 years emphasizes the importance of maintaining a reference body weight in young- and middle-aged adulthood, as is suggested by the results of other studies of the association between mortality risk and body weight for young- and middle-aged adults (27–30). The present estimates were robust in that they showed similar patterns in separate analyses: a) using different reference-weight BMI categories; b) for males and females; c) with a wider range of covariates; d) with a shorter (10-year) follow-up period; and e) after excluding all NHANES-I participants who had their smoking status retrospectively assessed a decade after the NHANES I survey.

All except one of the estimates of the advancement period for overweight were not significantly different from zero. Larger NHANES data sets would probably have yielded more significant values. Analyses in larger cohorts such as the Nurses Health Study (27) and the NIH-AARP cohort (37) found relative risks of mortality for overweight among healthy young- and middle-aged nonsmokers, which were significantly >1.00.

The present estimates of the advancement period for obesity are in relatively good agreement with the previously published estimates. For instance, Stevens et al. (12), who studied the Cancer Prevention Study I cohort, found that for nonsmoker healthy participants without unintentional weight loss, mortality was likely to occur earlier for obese than reference-weight participants by between 5.9 and 11.7 years among young- and middle-aged adults. This range includes the present estimate of 9.44 years for healthy nonsmokers of ≦55 years. The present estimate for nonsmokers of ≦55 years, 8.83 years, is somewhat larger than the estimates by Peeters et al. (31) that YLL for nonsmokers of 40–49 years were 5.8 and 7.1 years for males and females, respectively. The Prospective Studies Collaborative (32), based on a survival analysis in their data set with 894,576 Western Europe and North-American participants, estimated that for participants with a BMI of 30–35 kg/m2 in middle age, survival would be curtailed by 2–4 years compared to those with a BMI of 22.5 to <25 kg/m2 in middle age. This estimated range overlaps the present estimate of 2.75 years for NHANES I, II, and III participants with grade 1 obesity. The relative risks for all NHANES I, II, and III participants for obesity, overweight, and underweight in this analysis are in general agreement with the results presented by Flegal et al. (4). It is not possible to make specific comparisons because of the different age categories used in the two analyses.

Failure to stratify simultaneously for smoking, pre-existing illness and age and limit focus to healthy nonsmokers with age of ≦55 years in this analysis yielded estimates of the advancement period for obesity that were sizeable underestimates for healthy young- and middle-aged nonsmokers. This was not owing to the fact that smoking or pre-existing illness caused large increases in the estimates. Stratifying for pre-existing illness alone, smoking alone or pre-existing illness and smoking simultaneously and limiting focus to healthy nonsmokers increased the advancement period by no more than 13%. Stratifying for age alone caused a larger increase. The advancement period was 78% larger for participants of ≦55 years than for all participants. Simultaneous stratification for preexisting illness, smoking, and age yielded an advancement period for healthy nonsmokers of ≦55 years that was 189% larger than for all participants. Previous investigators have arrived at results consonant with these findings. Adams et al. (33) who studied 527,265 US adults in the NIH-AARP cohort stratified for smoking and found that the relative risk for obesity was substantially larger for nonsmokers than for all adults; after simultaneously stratifying for smoking and age, they found that the relative risk for healthy nonsmokers was even larger. Manson et al. (27) found that the relative mortality risk in the Nurses Health Study when all nurses were of age of ≦55 years was larger for healthy nonsmoker than for healthy nurses. Similarly, Koster et al. (34) found that stratifying for smoking and adiposity had an additive effect on relative mortality risk in the NIH-AARP cohort. Flegal et al. (4, 35) found that the separate effects of stratifying for smoking and illness were small in their analysis in the NHANES I, II, and III data, as was found in this analysis. Finally, the BMI in Diverse Populations Collaborative Group (36) found that stratifying for smoking had little effect on the relationship between relative risk and BMI in an analysis without stratification for illness or age. Taken together, this evidence of the additive nature of interaction effects suggests that not stratifying for all significant interaction effects can yield estimates of relative risk and advancement period that are substantial underestimates for subgroups such as the healthy nonsmokers of age ≦55 years in this analysis.

The estimates of the advancement period and relative risk for obesity for healthy nonsmokers decreased as age advanced above 55 years. This decreasing trend is similar to the decrease in the advancement period for obesity with age among older participants observed by Stevens et al. (12). The present decreasing trend is most likely owing to the fact that compared to BMI for censored participants, BMI for decedents was higher for age of ≦55 years, and lower for age of >65 years, and increasingly lower as aging advanced (Figure 1). This pattern is consonant with the recent observation by Adams et al. (37) that among elderly NIH-AARP participants “weight loss after the age of 50 years was more strongly associated with the risk of death than was weight gain.” Most older participants in this analysis appeared not to have suffered from a pre-existing serious fatal illness that induced the weight loss. For instance, only 36.2% of participants over the age of 75 years reported a history of cardiovascular disease, cancer, or emphysema. It seems likely, therefore, that older participants tend to lose weight toward the end of the lifespan owing to nonpathological aging-related factors such as sarcopenia, cachexia, organ atrophy, and bone-mineral-density loss that are associated with a decrease in functionality (38). This pattern would tend to associate higher mortality risk with lower BM in a survival analysis, in a manner similar to that postulated for smoking and pre-existing illness (19).

There were several limitations in this analysis. First, data were not available on weight loss prior to the start of follow-up. This prevented an assessment of whether the advancement period would have been higher than the estimates reported here for older participants without prior weight loss. Figure 1 supports this possibility because it suggests that older participants tended to lose weight with increasing age, and that lower BMI was associated with higher risk of mortality during follow-up. A second limitation is that, as in most prospective epidemiological analyses of the association between adiposity and mortality, the present smoking, pre-existing illness, and alcohol consumption data were self-reported. Reporting errors undoubtedly induced some bias in the present estimates of the advancement period. There is some evidence showing that the smoking data in the NHANES III cohort contain low levels of error (39).

One of the strengths of this analysis is that the predictor, BMI, is derived from measured values of body weight and height. A second strength is that the advancement period is based on fewer and more realistic assumptions that other similar measures, such as YLL.

Conclusions

In conclusion, this 15-year prospective analysis in the NHANES I, II, and III data showed that after stratifying for pre-existing illness, smoking, and age, mortality was likely to occur 9.44 years earlier (0.72, 18.16) for participants who were obese than for those who were of reference weight. The estimates increased with increasing levels of obesity, and apply to healthy young- and middle-aged (age, ≦55 years) nonsmokers who constitute a sizeable portion of American adults.

Acknowledgements

The National Center for Health Statistics (NCHS) is the original source of the NHANES data. The author is responsible for all analyses, interpretations, and conclusions, and not NCHS, which is responsible only for the data. The research in this article was conducted while the author was a Special Sworn Status researcher of the U.S. Census Bureau at the Center for Economic Studies. Research results and conclusions expressed are those of the author and do not necessarily reflect the views of the Census Bureau. Approval to access and process the NHANES data was obtained from the Brooklyn College Institutional Review Board. There were no external sources of support for the project.

Ancillary