Body Mass Index and All-Cause Mortality in a Large Prospective Cohort of White and Black U.S. Adults

Remaining controversies on the association between body mass index (BMI) and mortality include the effects of smoking and prevalent disease on the association, whether overweight is associated with higher mortality rates, differences in associations by race and the optimal age at which BMI predicts mortality. To assess the relative risk (RR) of mortality by BMI in Whites and Blacks among subgroups defined by smoking, prevalent disease, and age, 891,572 White and 38,119 Black men and women provided height, weight and other information when enrolled in the Cancer Prevention Study II in 1982. Over 28 years of follow-up, there were 434,400 deaths in Whites and 18,702 deaths in Blacks. Cox proportional-hazards regression was used to estimate multivariable-adjusted relative risks (RR) and 95% confidence intervals (CI). Smoking and prevalent disease status significantly modified the BMI-mortality relationship in Whites and Blacks; higher BMI was most strongly associated with higher risk of mortality among never smokers without prevalent disease. All levels of overweight and obesity were associated with a statistically significantly higher risk of mortality compared to the reference category (BMI 22.5–24.9 kg/m2), except among Black women where risk was elevated but not statistically significant in the lower end of overweight. Although absolute mortality rates were higher in Blacks than Whites within each BMI category, relative risks (RRs) were similar between race groups for both men and women (p-heterogeneity by race  = 0.20 for men and 0.23 for women). BMI was most strongly associated with mortality when reported before age 70 years. Results from this study demonstrate for the first time that the BMI-mortality relationship differs for men and women who smoke or have prevalent disease compared to healthy never-smokers. These findings further support recommendations for maintaining a BMI between 20–25 kg/m2 for optimal health and longevity.


Introduction
Excess body weight is known to increase risk of premature mortality and various chronic diseases including cardiovascular disease, diabetes, and various types of cancer [1,2]. Both the disease etiology implications and public health impact of high body mass index (BMI) are vitally important given the extremely high rates of overweight and obesity in the U.S. and worldwide. Numerous observational studies [2][3][4][5] have reported on the relationship between obesity and mortality with quantitatively and qualitatively varied results.
Despite the large number of studies to date, four major unresolved issues in the study of obesity and mortality still exist. The first two related issues include whether overweight (BMI 25-,30 kg/m 2 ) is beneficial or detrimental in relation to total mortality and to what extent smoking and prevalent disease, conditions that both cause weight loss and shortened survival, influence the relationship between overweight and total mortality. A large pooled analysis that included 1.46 million Caucasian men and women found that smoking and prevalent disease significantly modified the BMI-mortality relationship. Among healthy, never smokers there was a statistically significant higher risk of mortality in underweight (BMI ,18.5 kg/m 2 ) and in the lowest end of normal weight (18.5-19.9 kg/m 2 ) groups as well as overweight and obese individuals compared to the upper end of normal weight (22.5-24.9 kg/m 2 ), and optimal BMI was between 20.0 and 24.9 kg/m 2 [3]. Conversely, a recent meta-analysis [5] including 97 studies reported that overweight was associated with lower all-cause mortality compared to normal weight, and reported that exclusion of smokers and individuals with prevalent disease had little effect on the magnitude of risk estimates. Unlike the large pooled analysis, the meta-analysis by Flegal et al, examined associations relative to the full range of normal weight (reference category was 18.5-24.9 kg/m 2 ). Inclusion of the low end of the normal weight in the reference group would result in mortality rates that are higher than if the reference group were limited to the apparently optimal range of 20-24.9 kg/m 2 , and would result in a lower observed relative risk in the overweight group, especially when smokers and individuals with prevalent disease are included.
The third issue is whether the association between BMI and mortality differs by race. While recent, large pooled analyses have examined the relationship between BMI and all-cause death rates in Caucasians [3] and Asians [6], the BMI-mortality dose-response relationship has not been well-characterized in African Americans.
Among the relatively few studies on this issue in African Americans [4,[7][8][9][10][11][12][13][14][15][16][17], most [8,10,[12][13][14][15][16][17] suggest that the association between BMI and mortality may be weaker in Blacks than Whites. However, studies of African Americans had relatively small sample sizes and/or short follow-up time. Finally, the fourth issue is that the appropriate age window of the exposure (the age at which body mass maximally predicts subsequent mortality) remains unclear. Results of various studies suggest that BMI measured at younger ages is more predictive of subsequent mortality long term because it is more effective in measuring adiposity and accounting for pre-existing disease than BMI measured later in life [4,8,9,[18][19][20][21].
The American Cancer Society's Cancer Prevention Study II is well-suited to the study of BMI in relation to mortality, due to its prospective design, large size of both Black and White men and women, and long follow-up period. Our initial paper examining the association between BMI and total mortality was published in 1999 when the cohort had been followed for 14 years [4]. That analysis was based on 14-years of follow-up and had a limited number deaths among Black men and women, especially among those who were obese (n = 83 deaths among men, n = 319 deaths among women). Here, we present results on the four issues described above with follow-up extended to 28 years including over 450,000 deaths.

Study Population
Subjects for this analysis were selected from the 1,184,387 participants (508,227 men and 676,160 women) in the Cancer Prevention Study II, a prospective study of mortality among men and women in the United States begun by the American Cancer Society in 1982 [22]. Participants were identified and enrolled by more than 77,000 volunteers in all 50 states, the District of Columbia, and Puerto Rico. Families were enrolled if at least one household member was 45 years of age or older, and all enrolled members were at least 30 years old. The average age of participants at enrollment was 57 years old. In 1982, participants completed a confidential questionnaire, through which they provided information on demographic characteristics, personal and family history of cancer and other diseases, environmental and occupational exposures, and factors related to lifestyle, behavior and diet. All aspects of CPS-II have been reviewed and approved by the Emory University Institutional Review Board, and all data were de-identified prior to analysis.
Participants were asked to provide their current weight, weight one year prior to enrollment, and height (without shoes). Subjects with missing values for height or current weight were excluded from the analysis (N = 29,320), as were those with extreme values (99.9 th percentile) for height or weight or extreme underweight (BMI ,15) (N = 6,484). In order to adequately control for smoking, we excluded subjects with missing data for the smoking questions (N = 46,900), smokers with no information on number of cigarettes per day (N = 87,342), former smokers with no information on years since quitting (N = 2,080), and smokers whose smoking status at baseline (current or former) was unknown (N = 20,068). We also excluded men who had smoked only pipes or cigars but not cigarettes (N = 40,462) because their amount of smoking was not well defined; pipe and cigar smoking was not queried in women. Finally, we excluded all races other than White or Black (N = 21,860 35.0 and higher. This categorization scheme includes finer groupings of the World Health Organization (WHO) definitions of ''underweight'' as body mass index (BMI) less than 18.5 kg/m 2 , the ''normal'' range 18.5 to 24.9 kg/m 2 , ''overweight'' range 25.0 to 29.9 kg/m 2 , and ''obese'' as greater than or equal to 30.0 kg/m 2 [23]. Additionally, these finer groupings were selected a priori for consistency with several previously published studies including the large pooled study of 1.46 million White participants reporting the nadir of the doseresponse curve was at BMI 22.5 to 24.9 [3].

End Points
Deaths occurring between the months of enrollment and December 31, 2010, were ascertained through personal inquiries by volunteers in September 1984, September 1986, and September 1988, and thereafter through linkage with the National Death Index (NDI) [24]. As of December 31, 2010, 50.5 percent of the participants had died and 49.5 percent were still living; 0.3 percent were lost to follow-up on September 1, 1988 due to insufficient data for linkage with the NDI. Multiple cause-of-death codes have been obtained for 99.3 percent of all deaths.
The primary end point in this analysis was deaths from all causes. In secondary analyses, deaths from all cardiovascular disease (ICD-9 codes 390 through 459; ICD-10 codes I00 through I99), all cancer (ICD-9 codes 140 through 208; ICD-10 codes C00 through C97), and all other causes, were examined separately in relation to BMI.

Statistical Analysis
Age-adjusted mortality rates according to BMI in men and women, race-specific, were calculated by direct standardization to the age distribution of the CPS-II male and female populations using 5-year age categories. Multivariable-adjusted relative risks (RRs) and 95% confidence intervals (CIs) were estimated in Cox proportional-hazards regression models stratifying on single year of age at enrollment. Models simultaneously adjusted for education (less than high school, high school graduate, some college/vocational training, college graduate and higher), physical activity (none, slight, moderate, or heavy exercise), alcohol use (nondrinker, ,1 drink per day, 1 drink per day, .1 drink daily), marital status (married vs. single, widowed, or divorced), aspirin use (yes/no), fat consumption, vegetable consumption (sex-specific tertiles), and estrogen replacement therapy among women (yes/ no). Models that included ever smokers were also adjusted for smoking status, frequency and time since quitting (current smoker with cigarettes per day (cpd) categorized as #10, 11-20, 21-30, 31-40, or 40 or more; former smoker having quit within the last year and smoked #10, 11-20, 21-30, 31-40, or 40 or more cpd; former smoker having quit 1 to 9 years prior and smoked #10, 11-20, 21-30, 31-40, or 40 or more cpd; former smoker having quit 10-19 years prior and smoked #10, 11-20, 21-30, 31-40, 40 or more cpd, and former smoker having quit at least 20 years prior and smoked #10, 11-20, 21-30, 31-40, or 40 or more cpd) and those that included prevalent disease also adjusted for disease status (no/yes for cancer, heart disease, stroke, or emphysema; yes for chronic bronchitis or asthma).
We examined the association between BMI and all-cause mortality for each of four mutually exclusive race and genderspecific subgroups according to smoking status [never, ever] and prevalent disease status [yes/no] defined as a history of cancer (except non-melanoma skin cancer), heart disease, stroke, respiratory diseases (chronic bronchitis, emphysema, or asthma), current illness of any kind, or weight loss of 10 or more lbs (4.5 kg) in the previous year. Effect modification by smoking and prevalent disease combined (healthy never smokers, healthy smokers, never smokers with prevalent disease, smokers with prevalent disease) and by race were evaluated in multivariate models using the likelihood ratio test and a p-value ,0.05 was considered statistically significant. In Black and Whites combined, we examined the association between BMI and mortality by age at enrollment (younger than 50, 50-59, 60-69, 70 years or older) but this analysis was limited to the healthy never smokers because smoking and prevalent disease significantly modified the association between BMI and mortality. Finally, we examined the associations of BMI with cardiovascular disease and cancer mortality separately.

Results
The mean age at baseline was 56 years old for White men, 55 for Black men, 57 for White women and 55 for Black women. The mean BMI was 25.8, 26.8, 24.6, 27.1 kg/m 2 for White and Black men and women, respectively. Selected baseline characteristics for each sex-race group by BMI category are presented in Table 1. BMI was inversely related to age at baseline for all groups except Black women. A strong positive relationship was observed between BMI and the prevalence of never smokers in both men and women at the time of enrollment, regardless of race. In contrast, there was a curvilinear relationship between BMI and prevalent disease, such that individuals at both high and low extremes of BMI were more likely to report disease at baseline than persons within the WHO normal range of BMI (18.5-24.9 kg/m 2 ). For all groups, BMI was also inversely related to educational status at BMI levels of 22.5 kg/m 2 or greater. Tables 2 and 3 present the BMI-mortality relationship by race for men and women, respectively, stratified by smoking and prevalent disease status. Smoking and prevalent disease modified the association between BMI and mortality in all four groups (interaction p = 2.4610 239 for White men, p = 0.055 for Black men, p = 1.6610 232 for White women, and p = 0.02 for Black women). For White men and women, regardless of smoking status or prevalent disease, underweight was associated with higher risk of mortality compared to BMI 22.5-24.9 kg/m 2 . For Black men, underweight was associated with higher risk of mortality in those with prevalent disease, regardless of smoking status, but not in Black men without prevalent disease. In Black women, underweight was associated with higher mortality in all groups, albeit a non-statistically significant elevated risk for the group defined as never smokers with prevalent disease. Among all underweight BMI categories, age-standardized death rates were highest in men and women who smoked and had prevalent disease, and rates were lowest in healthy never smokers.
Relative risks were higher for overweight and obese compared to the upper end of normal weight (BMI 22.5-24.9 kg/m 2 ) among the healthy never smokers compared to those who smoked and/or had prevalent disease (Tables 2 and 3). In this group of never smokers without prevalent disease, among both men and women, the RR's for overweight Blacks appear to be similar in magnitude to those for Whites whereas at the highest levels of BMI, the RR's for Blacks appear to be marginally lower compared to those for Whites. For example, among healthy never smokers, the multivariable-adjusted RR for White women with BMI. = 40 kg/m 2 Table 3. Cont.  for Black women (95% CI 1.43-2.21). However, there was no evidence of statistical interaction by race (interaction p = 0.23 for women and p = 0.20 for men). Due to the lack of effect modification by race and the profound modifying effect of smoking and prevalent disease on the association between BMI and mortality, analyses stratified by age and grouped cause of death are presented among never smokers without prevalent disease in both races combined. The magnitude of the association between BMI and total mortality varied substantially by the age at which BMI was reported (Table 4). BMI at ages younger than 70 years was much more strongly associated with risk of death in both men and women than was BMI at older ages. For BMI at ages less than 60 years and 60-69 years, all categories of overweight and obesity were associated with higher risk of mortality. The association between BMI and mortality was greatly attenuated when BMI was reported at age 70 and above; however, in women, associations remained statistically significant for all categories of BMI.27.4 kg/m 2 . Overweight and obesity were associated with all grouped causes of death, but were generally stronger for death from cardiovascular disease and other causes compared to cancer for both men and women (Table 5).

Discussion
In this large prospective study of approximately one million Black and White men and women, we were able to address four major unresolved issues in the study of BMI and mortality. Results showed that men and women who were underweight were at higher risk of mortality as were men and women who were overweight and obese compared to normal weight men and women, and that smoking and prevalent disease significantly modified the association between BMI and mortality, such that the strongest associations were among never smokers without prevalent disease for men and women. In healthy never smokers, mortality rates were lowest within the upper end of the normal BMI category (i.e., 22.5-24.9 kg/m 2 ) for all race sex groups. Although the results of our study showed no statistically significant differences in associations between BMI and mortality by race, overweight and obesity were associated with subsequent mortality among smokers and/or those with prevalent disease in White men, White women and Black women but not in in Black men. In addition, weight in late middle age but not older (i.e., 70 years or older) was strongly associated with future mortality.
There is growing evidence that residual confounding by smoking and/or reverse causality by prevalent disease attenuated the association between BMI and risk of mortality. Indeed, studies that included smokers (who tend to have a lower BMI) and individuals with diseases that cause weight loss generally show weaker associations with high BMI levels, and stronger associations with low BMI levels than do studies that exclude these subjects [4,8,[25][26][27][28]. While controlling for smoking history can reduce its confounding effects, eliminating the residual effects of current smoking is essential for clarifying associations of BMI with mortality, and can only be achieved through exclusion of current smokers. On the other hand, prevalent disease exclusions may be more dependent on the characteristics and age of the study population. Consistent with the largest pooled analysis to date [3], in our study underweight and the lower end of the normal weight range were associated with higher risk of mortality among those who smoked or had prevalent disease.
Numerous studies have reported on the association between BMI and mortality in Caucasians [2][3][4]8,9,[18][19][20][21]. In contrast, few studies have reported on the BMI-mortality relationship among Black men and women [4,[7][8][9][10][11][12][13][14][15][16][17], and among Table 4. Cont.  those studies, results are inconsistent with most studies showing a stronger association in Whites than in Blacks [8,10,[12][13][14][15][16][17]. In fact, only two previous studies examined the BMI-mortality association in African American among healthy never-smokers. The first study included only women [11] and found that BMI-mortality associations were similar in magnitude as those reported in other studies among Whites; however, they were unable to directly compare results between Whites and Blacks as the study population included only Black women. The second was the earlier CPS-II analysis which included both men and women and was based on 14-years of follow-up [4]. Prior to the present analysis, that was the largest study to date among African Americans, and included a total of 485 deaths in Black men and 1,188 deaths in Black women. In that analysis, there was an excess risk of mortality with higher BMI in all race-sex groups, but the association between BMI and mortality appeared to be weaker in African Americans that in Whites. However, there were many fewer deaths identified among obese Black men and women (83 and 319 deaths, respectively) in the 14 year follow-up than in this study (247 and 1,002 deaths, respectively). In the current analysis, while Blacks had higher absolute mortality rates, the relative risks associated with BMI did not differ by race for men or women. The increased statistical power and precision is likely to have resulted in more stable estimates in the current compared to earlier analysis. In addition, the longer follow-up time would likely reduce confounding by undiagnosed disease-related weight loss.
The present study showed that the BMI-mortality association is stronger when BMI is assessed at younger ages (i.e. ,70 years old in this study) in both men and women. Disease-related weight loss is less common at younger ages and BMI is a better measure of excess adiposity in young and middle-aged adults [48] than in the elderly [49]. This finding is consistent with many other reports in Whites [4,8,9,[18][19][20][21], but the variation in the BMI-mortality relationship by age had not been examined in Blacks.
The major strengths of this study are its large sample size, wide BMI range and long follow-up time. In addition, the large number of Black men and women in CPS-II allowed for the largest detailed analysis of the BMI-mortality relationship in Blacks to date. The main limitation of the study is the reliance on a single self-report of height and weight. When compared with measured height and weight, self-reported data systematically overestimate height in men and underestimate weight in women [50], although the magnitude of this error is small. A longitudinal study using national data collected over 20 years  showed that Black men aged 48-60 years had a higher average weight gain per year compared to White men (0.14 and 0.02 kg/year, respectively) [51]. In contrast, for women, especially Black women, weight gain tended to occur earlier in adulthood; thus, the single measure may more accurately reflect long-term weight in women and White men compared to Black men. In addition, it is also possible that exclusions for prevalent disease (cancer, heart disease, stroke, emphysema, chronic bronchitis or asthma) were not adequate to control for all disease-related weight loss.
In summary, this large nationwide study helps to clarify four major unresolved issues in the study of BMI and mortality by demonstrating that smoking and prevalent disease significantly modify the relationship between BMI and mortality, and that among never smokers without prevalent disease, overweight and obesity are strongly associated with subsequent risk of mortality and the optimal BMI range is 20.0-24.9 kg/m 2 . Additionally, this study demonstrated that among never smokers without prevalent disease, the BMI-mortality relationship is similar in Black and White men and women where overweight (BMI 25.0-29.9 kg/m 2 ) is associated with a modest and obesity (BMI 30.0+ kg/m 2 ) a more Table 5. Cont.  substantial increased risk of premature death. In the U.S., ageadjusted prevalence estimates of obesity are highest for African Americans (35.7%), followed by Hispanics (28.7%), and non-Hispanic Whites (23.7%) [52]. Given the high prevalence of obesity among all racial-ethnic populations, although disproportionately higher among African Americans, these findings are of considerable clinical and public health relevance.