Association between Class III Obesity (BMI of 40–59 kg/m2) and Mortality: A Pooled Analysis of 20 Prospective Studies

In a pooled analysis of 20 prospective studies, Cari Kitahara and colleagues find that class III obesity (BMI of 40–59) is associated with excess rates of total mortality, particularly due to heart disease, cancer, and diabetes. Please see later in the article for the Editors' Summary

Effective interventions and accurate projections of future healthcare costs requires a better understanding of the health risks associated with class III obesity, but the necessary data are scarce. Because class III obesity was relatively uncommon in the recent past, the few studies that have specifically evaluated mortality rates associated with class III obesity generally had limited sample sizes (e.g., fewer than 400 deaths [2,11]) or did not separately evaluate risks of death for BMI of 50 kg/m 2 or higher [2,3]. As such, there is little quantitative information about the burden of disease, including total and cause-specific mortality rates, for individuals with BMI values of 40 kg/m 2 and above.
We combined original data from 20 prospective studies from the United States, Sweden, and Australia to evaluate-in what is to our knowledge the largest study on this topic to date-the excess rates of death overall and due to a wide range of specific causes, as well as the expected number of years of life lost attributable to class III obesity.

Study Population
The study sample included participants from the 20 cohort studies in the National Cancer Institute Cohort Consortium that met the eligibility criteria (.5 y of follow-up, .1,000 deaths among non-Hispanic white participants, baseline year 1970 or later) and had the ability to submit data for a previous pooled analysis of BMI and mortality [3], including the Adventist Health Study-I [12], Agricultural Health Study [13], Breast Cancer Detection Demonstration Project [14], California Teachers Study [15], Cancer Prevention Study-II [16], CLUE-I and -II [17], Cohort of Swedish Men [18], Health Professionals Follow-Up Study [19], Iowa Women's Health Study [20], Melbourne Collaborative Cohort Study [21], New York University Women's Health Study [22], National Institutes of Health-American Association of Retired Persons Diet and Health Study (NIH-AARP) [23], Nurses' Health Study-I [24], Physicians' Health Study-I and -II [25], Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial [26], Swedish Mammography Cohort [27], Swedish Women's Lifestyle and Health Study [28], United States Radiologic Technologists Study [29], VITamins and Lifestyle Study [30], and the Women's Health Study [31]. Baseline was defined as the date of completion of the first questionnaire that collected information on key variables (e.g., height, weight, smoking, and personal history of chronic diseases). Personal history of chronic diseases and other important covariates, including race/ethnicity, education, marital status, alcohol consumption, and physical activity level, were selfreported. Height and weight information was self-reported in all but one cohort [21]. Individual-level data were formatted uniformly across studies and combined.
Participants who were younger than 18 y or older than 85 y old at baseline, participants with less than 1 y of follow-up, and participants with BMI$60 kg/m 2 were excluded. We also excluded participants who reported having ever smoked cigarettes and/or a history of heart disease, cancer, stroke, or emphysema, as tobacco use and these preexisting illnesses have been demonstrated to confound the association between BMI and mortality [3]. The final sample consisted of 9,564 individuals with a BMI of 40.0 to 59.9 kg/m 2 (1,575 men and 7,989 women) and 304,011 with a BMI of 18.5 to 24.9 kg/m 2 (75,680 men and 228,331 women).

Follow-Up
Participants were followed from the date of completion of the baseline questionnaire to death, loss to follow-up, or administrative end date, whichever occurred first. Causes of death were ascertained from death records or registries and coded according to the Surveillance, Epidemiology, and End Results Program recodes [32] based on cause of death codes defined by the International Classification of Diseases, eighth, ninth, or tenth revisions [33][34][35], and categorized according to the 2002 National Vital Statistics Report of the US Centers for Disease Control and Prevention [36].

Statistical Methods
We calculated age-adjusted mortality rates (number of deaths per 100,000 persons per year) for the class III obesity (BMI 40.0-59.9 kg/m 2 ) and normal-weight (BMI 18.5-24.9 kg/m 2 ) groups using direct standardization for age [37], in which rates were weighted according to the age distribution of the total US population in 2000, restricted to ages 20-84 y, using 5-y categories of attained age. We used the US standard to most closely approximate the age distributions of the countries from which the study populations were selected. Specific causes of death were chosen from among the top leading causes of death in the US population in 2000 [36], including (but not limited to) heart disease, cerebrovascular disease, malignant neoplasms, chronic lower respiratory diseases, accidents, diabetes mellitus, influenza and pneumonia, kidney disease (nephritis, nephrotic syndrome, and nephrosis), septicemia, and chronic liver disease and cirrhosis. Mortality rates from suicide and homicide and from major causes of death resulting in fewer than five events in both men and women with BMI values of 40.0 kg/m 2 and above (e.g., Alzheimer disease) are not shown. Within the major categories of heart disease, malignant neoplasms, and cerebrovascular disease, where the sample size was sufficient, we additionally calculated mortality rates due to more specific causes (e.g., ischemic heart disease). Rate differences were calculated as the difference in age-adjusted mortality rates between the two BMI groups, and two-sided pvalues for these differences were calculated using a two-sample Z test with the variance calculated under the assumption that the number of deaths in each group followed an independent Poisson distribution.
We used proportional hazards regression models, with attained age as the underlying time metric, to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for total mortality and mortality due to major causes in relation to class III obesity (BMI 40.0-59.9 kg/m 2 ) compared with normal weight (BMI 18.5-24.9 kg/m 2 ). All models were adjusted for sex, race/ ethnicity, education, alcohol intake, level of physical activity, and study. In separate models, we additionally adjusted for selfreported history of diabetes mellitus and hypertension to evaluate the potential for mediation by these co-morbid conditions. In separate models restricted to participants with BMI values between 40.0 and 59.9 kg/m 2 , BMI was included as a continuous variable to evaluate HRs and 95% CIs per 5 kg/m 2 BMI and to calculate tests for trend. We observed no evidence of violation of the proportional hazards assumption based on graphical assessment of cumulative hazards plots by BMI category.
To calculate years of life lost, we calculated direct adjusted survival curves for each BMI category [38,39]. Proportional hazards regression models were used to estimate probabilities of survival at each age for each individual, which were then averaged to obtain an overall summary curve. Curves for each BMI category were estimated by counterfactual (i.e., applying the beta for each BMI category in turn to estimate individual and overall survival curves). Years of life lost were calculated as the difference between the adjusted median life expectancy for a given BMI category and the reference category of BMI 18.5-24.9 kg/m 2 . Models were restricted to participants who were 40 y and older at baseline.
As a secondary analysis and alternative approach to combining results across studies, we additionally calculated pooled HRs for class III obesity and total mortality using a random effects meta-analysis model. Heterogeneity between studies was calculated using the I 2 index [40]. Confidence intervals for the I 2 index were calculated according to the method described by Thorlund et al. [41]. We conducted subgroup analyses within the aggregate dataset to evaluate potential sources of betweenstudy heterogeneity. For instance, because the association between BMI and mortality has been shown to differ by age at BMI measurement/report and sex [3], and medical care for extremely obese individuals may have improved over calendar time, we calculated HRs and 95% CIs within strata of these factors. We also evaluated geographic location (US versus non-US cohorts) as another potential source of between-study heterogeneity. Tests for interaction by sex, geographic location (US versus non-US cohorts), attained age, and calendar period of follow-up (before and after age 65 y, and before and after the year 2000) were conducted by including a cross-product term in the model. Models evaluating interactions by attained age and calendar period were time-dependent, in which person-time for each participant was split before and after age 65 y or the year 2000.

Results
A description of participants with BMI in the class III obesity range from each of the 20 cohorts is shown in Table 1. In total, 1,036 of the 9,564 participants died during the follow-up period (median = 9 y, maximum = 27 y).
Compared with participants in the normal-weight range at baseline, those in the class III obesity range included a higher proportion of women, more individuals with baseline age 50 to 69 y (as opposed to younger [19-49 y] or older [70-83 y] individuals), fewer non-Hispanic white individuals, fewer individuals with a college education, more individuals with lower alcohol intake, and more individuals with lower levels of physical activity ( Table 2). Among participants within the range of class III obesity, we observed little variation in baseline age, sex, education, and alcohol intake. By contrast, the prevalence of non-Hispanic black race/ethnicity, low physical activity, and history of diabetes, as well as median waist circumference, was generally higher for the top BMI categories. Table 3 compares total and cause-specific mortality rates (deaths per 100,000 persons per year) for the class III obesity and normal-weight groups separately by sex. Total mortality rates for class III obesity participants were 856.0 in men and 663.0 in women. The differences in mortality rates for the two BMI groups were 509.3 and 382.5 in men and women, respectively. Heart disease was the most common underlying cause of death for the class III obesity group (mortality rate differences were 238.9 and 132.8 in men and women, respectively), followed by malignant neoplasms (mortality rate differences were 36.7 and 62.3 in men and women, respectively), and diabetes (mortality rate differences were 51.2 and 29.2 in men and women, respectively). Higher rates of death were also observed for nearly every other major cause of death that we examined, apart from cerebrovascular disease in men and malignant neoplasms of respiratory and intrathoracic organs and diseases of the arteries, arterioles, and capillaries in women, although not all of these differences were statistically significant because of the relatively small numbers of deaths in the class III obesity group for some of the major causes (e.g., malignant neoplasms, chronic lower respiratory disease, and nephritis/nephrotic syndrome/nephrosis deaths in men). The negative rate differences that were observed were also based on small numbers of deaths in the class III obesity group (five, eight, and five deaths due to cerebrovascular disease, malignant neoplasms of respiratory and intrathoracic organs, and diseases of the arteries, arterioles, and capillaries, respectively).
As an internal comparison, in the subset of participants in the pooled dataset without heart disease, cancer, stroke, or emphysema and having BMI values between 18.5 and 24.9 kg/m 2 , differences in mortality rates for current versus never smokers were 721.0 and 519.6 in men and women, respectively. Table 4 shows the minimally adjusted and multivariableadjusted HRs for death across categories of BMI and for continuous (per 5 kg/m 2 ) values of BMI restricted to the BMI 40.0-59.9 kg/m 2 range. In minimally adjusted models, we observed a sharp gradient in relative risk with higher values of BMI across the class III obesity levels that was consistent with a linear increase. The HRs from models of BMI categories were slightly, but not fully, attenuated with additional adjustment for race/ethnicity, education, alcohol intake, and physical activity level. Compared with adults with BMI in the 18.5-24.9 kg/m 2 range, multivariable-adjusted HRs (95% CIs) for those with BMI  Although the pooled HRs calculated from random effects metaanalysis models were similar to those based on the aggregate dataset (BMI 40.0-59.9 versus 18.5-24.9 kg/m 2 , HR = 2.57, 95% CI: 2.23-2.96; per 5 kg/m 2 BMI, HR = 1.42, 95% CI: 1.25-1.62), the heterogeneity estimates for differences in the associations by cohort were statistically significant (Figures 1 and 2). We evaluated potential sources of heterogeneity in Table 5, which shows the multivariable-adjusted HRs conducted overall and by population subgroup. Overall, we observed a 2.57-fold (95% CI: 2.41-2.74) increased risk of death in the BMI 40.0-59.9 kg/m 2 versus the 18.5-24.9 kg/m 2 group, and the association increased linearly for BMI values above 40 kg/m 2 (HR per 5 kg/m 2 = 1.40, 95% CI: 1.31-1.51). As observed in studies of BMI across the entire range of values in relation to mortality [3], these associations were significantly modified by sex (p-interaction = 0.03), by geographic location (US versus non-US, p-interaction = 0.03), and follow-up time before and after age 65 y (p-interaction ,0.001), with stronger associations observed for men (HR = 2.83, 95% CI: 2.45-3.26), for US cohorts (HR = 2.60, 95% CI: 2.44-2.78), and for follow-up before age 65 y (HR = 3.10, 95% CI: 2.77-3.47). The interaction by geographic location may have arisen due to chance,

20.9
Other disorders of the circulatory system  Table 6. HRs were significantly elevated for deaths due to heart disease, malignant neoplasms, cerebrovascular disease, diabetes mellitus, nephritis/nephritic syndrome/nephrosis, chronic lower respiratory disease, influenza/pneumonia, accidents, and septicemia, ranging from 1.69 for deaths due to malignant neoplasms to 9.57 for deaths due to nephritis/nephrotic syndrome/nephrosis. Additional adjustment for self-reported history of diabetes and hypertension attenuated the HRs for all specific causes of death apart from malignant neoplasms, though all remained statistically significant. Within the BMI 40.0-59.9 kg/m 2 range, greater BMI values were linearly associated with increasing HRs for all of these major causes of death apart from cerebrovascular disease, accidents, and septicemia; additional adjustment for history of diabetes and hypertension had little influence on these continuous HRs.

Discussion
To our knowledge, this is the largest study to date on the association between class III obesity and mortality, and the first to show that class III obesity is associated with excess rates of total mortality and mortality due to a wide range of causes, particularly heart disease, cancer, and diabetes, and that the risk of death overall and from these specific causes continues to rise with increasing values of BMI. The accumulating excess risk resulted in major reductions in life expectancy after the age of 40 y that were comparable to those of cigarette smoking.
The estimated 6.5 to 13.7 y of life lost for BMI values between 40 and 59 kg/m 2 versus 18.5-24.9 kg/m 2 were in line with those of a previous pooled analysis that found that individuals with BMI values of 35-50 kg/m 2 had a median 8-10 fewer years of life than those with BMI 22.5-24.9 kg/m 2 [2]. Our study further demonstrates that the expected number of years of life lost continued to increase for BMI values beyond 50 kg/m 2 , at which point the loss in life expectancy (9.8 y) exceeded that observed for current versus never smoking (8.9 y) in this study. These results have great relevance to the current era, during which class III obesity rates have increased dramatically at the same time that smoking rates have declined. That we observed a 7.2-y decrease in life expectancy from BMI 40.0-44.9 to 55.0-59.9 kg/m 2 suggests that otherwise healthy, non-smoking adults having BMI values within the class III obesity range may considerably expand their life expectancy by avoiding additional weight gain.
We observed significant differences in the relative risks for class III obesity and total mortality by cohort, which may reflect differences in the ages, sex distribution, and geographical location of the participants, as these factors were shown to be significant effect modifiers in our study. Similar to observations from previous studies based on lower values of BMI [3], these results suggest that the risks associated with class III obesity may be even greater for men than women, and for younger compared to older adults. These results could be considered in the development of public health interventions aimed at reducing health risks associated with extreme obesity, particularly those that are targeted toward susceptible populations. The clear finding of excess risk for nearly every major cause of death suggests that excess weight can have a wide range of physiological effects. In particular, the results of our study suggest that diabetes and hypertension are common mechanisms that might explain the majority of the excess mortality burden in the class III obesity group. Most of this excess was attributable to heart disease, diabetes, cancer, and kidney and liver diseases, and risks of death due to these causes were largely attenuated with additional adjustment for diabetes and hypertension. Obesity is strongly associated with metabolic abnormalities, including insulin resistance, type 2 diabetes, hypertension, and dyslipidemia, mediated in part by the chronic inflammatory state induced by the secretion of adipocytokines, such as angiotensinogen, transforming growth factor-beta, tumor necrosis factor-alpha, and interleukin-six [42][43][44][45]. The severity of these conditions has been shown to increase with increasing BMI [44] and improve after weight loss either due to lifestyle modification or bariatric surgery [46][47][48][49]. Diabetes, hypertension, and other obesityrelated aspects of the metabolic syndrome have been shown to play an important role in the development of heart disease, chronic kidney disease, and end-stage renal disease [42,45], among other major causes of death observed to be elevated in the class III obesity group. Thus, these metabolic conditions could be effect targets in efforts to reduce the burden of morbidity and mortality in individuals with extreme obesity. Individuals with extreme obesity may experience other physiological impairments that could explain the higher rates of death from other causes. Specifically, deaths due to chronic lower respiratory disease were shown in this and the much smaller National Health and Nutrition Examination Survey III study to be elevated at extreme levels of BMI due to reduced lung capacity and airway obstruction [50]. It is possible that excess risks associated with class III obesity may decline as treatments for these co-morbid conditions improve. Results for cohorts with fewer than five deaths in the BMI 40.0-59.9 kg/m 2 group are not shown. Cohort-specific models use attained age as the underlying time metric, and are adjusted for sex, race/ethnicity (white [all participants in the Swedish and Australian cohorts were coded as white], black, Asian/Pacific Islander, Hispanic, other/unknown), education (high school or less, post-high school, college, unknown), alcohol intake (grams of ethanol per day: 0, .0 to ,10, $10, unknown), and physical activity level (cohort-specific tertiles corresponding to low, medium, and high). Random effects models were used to calculate overall (pooled) HRs and 95% CIs. The cohort acronyms are identified in Table 1, and references are given in parentheses. doi:10.1371/journal.pmed.1001673.g001 An important strength of this pooling study is its large size. With sufficient numbers, we were able to quantify the risks of total and cause-specific deaths associated with class III obesity with greater precision than has been possible in previous studies. Furthermore, our results were based on a population of never smokers without prevalent disease, thereby reducing the potential for confounding due to major preexisting illness and smoking. Our results were also based on original, as opposed to published, data, which allowed for standardization of the variables and statistical methods across studies, and thereby reduced the potential for methodological heterogeneity between studies.
Some limitations of this study include the use of mostly selfreported, as opposed to measured, height and weight. Objectively measured values are ideal, but epidemiologic studies have reported high correlations between self-reported and measured height and weight [51,52]. Nonetheless, misclassification across BMI categories due to self-reported height and weight has been shown to yield stronger relative risks than those based on measured values [53]. This potential source of bias may have inflated our relative risk estimates, although this effect is likely to have been modest [3]. Another limitation is that height and weight were ascertained only once at the beginning of follow-up, which precluded us from examining risks associated with weight change. Although BMI is the most commonly used indicator of adiposity in epidemiologic studies, it does not directly measure the amount of adipose tissue and cannot distinguish between fat and lean body mass [54,55]. Nonetheless, BMI has been shown to be just as strongly correlated with obesity-related metabolic indicators as more accurate methods for measuring total fat mass and body fat percent, such as dual-energy X-ray absorptiometry [55]. BMI also does not provide information on central versus peripheral body fat distribution, each posing different risks in relation to the metabolic Figure 2. Cohort-specific and overall hazard ratios and 95% confidence intervals for continuous BMI (per 5 kg/m 2 ). Analyses were restricted to participants with BMI 40.0-59.9 kg/m 2 . Results for cohorts with fewer than five deaths in the BMI 40.0-59.9 kg/m 2 group are not shown. Cohort-specific models use attained age as the underlying time metric, and are adjusted for sex, race/ethnicity (white [all participants in the Swedish and Australian cohorts were coded as white], black, Asian/Pacific Islander, Hispanic, other/unknown), education (high school or less, post-high school, college, unknown), alcohol intake (grams of ethanol per day: 0, .0 to ,10, $10, unknown), and physical activity level (cohort-specific tertiles corresponding to low, medium, and high). Random effects models were used to calculate overall (pooled) HRs and 95% CIs. The cohort acronyms are identified in Table 1, and references are given in parentheses. doi:10.1371/journal.pmed.1001673.g002 syndrome and related chronic diseases [56]. Studies that combine BMI with measures of central adiposity may yield more accurate estimates of disease risks associated with high levels BMI [57,58]. For instance, in a pooled analysis based on data from the same 20 cohort studies, waist circumference was positively associated with all-cause mortality, even among individuals at the highest range of BMI (35.0-49.9 kg/m 2 ) [58]. There has been some concern about the accuracy of cause of death reports [59], which may be differential by BMI level. The excess deaths due to both diabetes mellitus and overweight and obesity that we observed in the class III obesity group may reflect greater difficulty in identifying the primary cause of death among individuals with multiple comorbid conditions. Our results may not be generalizable to all populations, as the pooled dataset comprised mostly non-Hispanic white individuals from affluent countries (the majority being from the US). Also, our estimates rely on prospective cohort studies, whose volunteer participants typically are healthier than those in the general population.
In this large pooled analysis of 20 prospective studies, participants with BMI in the class III obesity range (40.0-59.9 kg/m 2 ) experienced substantially higher rates of death compared with those in the normal BMI range (18.5-24.9 kg/m 2 ), Table 5. Hazard ratios and 95% confidence intervals for risk of death according to BMI category (40.0-59.9 versus 18.5-24.9 kg/ m 2 ) or continuous values of BMI between 40.0 and 59.9 kg/m 2 (per 5 kg/m 2 ): subgroup analyses. with most of the excess due to deaths from heart disease, cancer, and diabetes. These higher rates appear to be largely attributable to metabolic abnormalities associated with excess adiposity, including diabetes and hypertension. We found that the reduction in life expectancy associated with class III obesity was similar to (and, for BMI values above 50 kg/m 2 , even greater than) that observed for current smoking. If current global trends in obesity continue, we must expect to see substantially increased rates of mortality due to these major causes of death, as well as rising health-care costs. These results underscore the need to develop more effective interventions to combat this growing public health problem. Compared to people with a healthy weight (a BMI between 18.5 and 24.9 kg/m 2 ), overweight and obese individuals (who have a BMI between 25.0 and 29.9 kg/m 2 and a BMI of 30 kg/m 2 or more, respectively) have an increased risk of developing diabetes, heart disease, stroke, and some cancers, and tend to die younger. Because people become unhealthily fat by consuming food and drink that contains more energy (kilocalories) than they need for their daily activities, obesity can be prevented or treated by eating less food and by increasing physical activity.

Author Contributions
Why Was This Study Done? Class III obesity (extreme, or morbid, obesity), which is defined as a BMI of more than 40 kg/m 2 , is emerging as a major public health problem in several high-income countries. In the US, for example, 6% of adults are now morbidly obese. Because extreme obesity used to be relatively uncommon, little is known about the burden of disease, including total and cause-specific mortality (death) rates, among individuals with class III obesity. Before we can prevent and treat class III obesity effectively, we need a better understanding of the health risks associated with this condition. In this pooled analysis of prospective cohort studies, the researchers evaluate the risk of total and cause-specific death and the years of life lost associated with class III obesity. A pooled analysis analyzes the data from several studies as if the data came from one large study; prospective cohort studies record the characteristics of a group of participants at baseline and follow them to see which individuals develop a specific condition.
What Did the Researchers Do and Find? The researchers included 20 prospective (mainly US) cohort studies from the National Cancer Institute Cohort Consortium (a partnership that studies cancer by undertaking large-scale collaborations) in their pooled analysis. After excluding individuals who had ever smoked and people with a history of chronic disease, the analysis included 9,564 adults who were classified as class III obese based on self-reported height and weight at baseline and 304,011 normal-weight adults. Among the participants with class III obesity, mortality rates (deaths per 100,000 persons per year) during the 30-year study period were 856.0 and 663.0 in men and women, respectively, whereas the mortality rates among normalweight men and women were 346.7 and 280.5, respectively.
Heart disease was the major contributor to the excess death rate among individuals with class III obesity, followed by cancer and diabetes. Statistical analyses of the pooled data indicate that the risk of all-cause death and death due to heart disease, cancer, diabetes, and several other diseases increased with increasing BMI. Finally, compared with having a normal weight, having a BMI between 40 and 59 kg/m 2 resulted in an estimated loss of 6.5 to 13.7 years of life.
What Do These Findings Mean? These findings indicate that class III obesity is associated with a substantially increased rate of death. Notably, this death rate increase is similar to the increase associated with smoking among normal-weight people. The findings also suggest that heart disease, cancer, and diabetes are responsible for most of the excess deaths among people with class III obesity and that having class III obesity results in major reductions in life expectancy. Importantly, the number of years of life lost continues to increase for BMI values above 50 kg/m 2 , and beyond this point, the loss of life expectancy exceeds that associated with smoking among normal-weight people. The accuracy of these findings is limited by the use of selfreported height and weight measurements to calculate BMI and by the use of BMI as the sole measure of obesity. Moreover, these findings may not be generalizable to all populations. Nevertheless, these findings highlight the need to develop more effective interventions to combat the growing public health problem of class III obesity.