Educational Differences in Mortality Associated with Central Obesity: Decomposing the Contribution of Risk and Prevalence

Abstract

Thousands of preventable deaths are attributed to obesity in the United States. However, the harmfulness of obesity varies across the population; individuals’ education determines access to healthful resources and exposure to competing risks, dampening/amplifying obesity-associated mortality risk. Using restricted U.S. National Health and Nutrition Examination Survey data (N=40,058; 1988–2015), this study estimates educational differences in mortality attributable to central obesity (waist-to-height ratio ≥0.5) – a dangerous form of abdominal adiposity. Over 30% of excess deaths are attributable to central obesity among college-educated adults, compared to 1–10% among their less-educated counterparts. This difference is larger for cardiometabolic-related mortality, as central obesity may explain 60–70% of excess deaths among college-educated adults. Decomposition analyses show differences are driven by greater obesity-associated risk among college-educated adults, rather than prevalence. Policies targeting health disparities should recognize central obesity as a key risk among highly-educated adults, but only one of many encountered by those with less education.

1. INTRODUCTION

Demographers and population health researchers have long anticipated a decline in the health of U.S. adults attributable to obesity. With over one-third of U.S. adults considered obese (Flegal et al. 2019; Hales et al. 2017) – and nearly 60% being “centrally-obese” (as defined by a high waist circumference [CDC 2018]) – government agencies and medical organizations continually stress the threat that obesity poses to the long-term health of the population. The warnings are substantiated by countless studies reaffirming the association between excess body fat and many health outcomes, including diabetes, cardiovascular disease, neurodegenerative diseases, osteoarthritis, and disability (Ahima and Lazar 2013), as well as at least 13 different types of cancer – representing 40% of all cancer diagnoses (Steele et al. 2017).

Unsurprisingly, obesity is consistently associated with increased mortality risk across numerous studies and meta-analyses, representing millions of adults across different sociodemographic groups (Aune et al. 2016; Di Angelantonio et al. 2016; Flegal et al. 2013), as well as many different causes of death (Flegal et al. 2007). This has led scholars to project declines in U.S. life expectancy due to increased obesity (Kontis et al. 2017; Preston et al. 2014), especially among recent cohorts who may “live less healthy and… shorter lives than their parents” (Olshansky et al. 2005:1163). Recent work suggests that the expanding obesogenic environment in the United States – defined by increased “availability of high energy dense, palatable, inexpensive food” and “mechanized labor-saving and entertainment devices” that promote sedentary lifestyles (Ard 2007: 1058) – has resulted in a slowing of mortality declines for middle-age men and women (Masters et al. 2017). Moreover, obesity is poised to overtake smoking as the leading cause of preventable death in the U.S. (Stokes and Preston 2016a); indeed, rising obesity rates may have already reduced U.S. life expectancy at age 40 by an estimated 0.9 years as of 2011 (Preston et al. 2018).

As research continues to document diverging life expectancy trajectories based on individuals’ educational attainment (Hayward et al. 2015; Montez and Zajacova 2013; Sasson 2016), examining educational heterogeneity in the contribution of obesity as a risk factor for early death can illuminate mechanisms underlying these disparities. Lower education is consistently associated with higher mortality (Hummer and Hernandez 2013; Montez et al. 2012; Montez et al. 2016) and increased obesity, especially among white adults in the U.S. and higher-income nations as a whole (Cohen et al. 2013a; Cohen et al. 2013b); however, it is unclear if the combined and negative effects of social disadvantage and unhealthy behaviors/lifestyles on health are additive or multiplicative (Mehta and Preston 2016; Pampel and Rogers 2004; Schaefer and Ferraro 2011). Consequently, educational variation in mortality risk associated with obesity, and in the proportion of excess deaths attributable to obesity, is less well understood than should be the case.

On the one hand, we would anticipate that adults with higher socioeconomic status [SES] – as suggested by their level of education – are better positioned to access key health resources (Link and Phelan 1995), potentially limiting the health and mortality consequences of obesity. Conversely, unhealthy lifestyles may be less consequential for low-SES adults as their comparatively worse baseline health and exposure to multiple sources of early death limits the ‘additional’ risk imposed by a single health hazard like obesity (Blaxter 1990). Obesity may then prove especially harmful to highly-educated adults, who are otherwise ‘healthy’ and free of such competing risks to premature mortality. Greater knowledge of these relative differences is crucial for avoiding population health interventions that advance the health of one group over another, potentially widening education disparities in mortality.

Given these competing perspectives for understanding the interplay between education, obesity, and mortality, the present study builds upon existing obesity-mortality literature by examining educational variation in mortality associated with central obesity among U.S. adults. Using nationally-representative survey data, we examine whether the proportion of excess deaths attributable to central obesity varies by education, and the extent to which any variation is explained by educational differences in central obesity-associated mortality risk as compared to differences in the prevalence of central obesity. As detailed in the following section, we specifically examine two plausible hypotheses – “amplified” and “saturated” risk – which anticipate higher obesity-associated mortality risk among lower-educated adults as would be expected given their greater social disadvantage and increased vulnerability, versus lower (or comparable) obesity-associated mortality risk among lower-educated adults as may be expected given the greater likelihood of competing sources of risk that lead to premature mortality. Ultimately, this study contributes to our understanding of how distal determinants of health like educational attainment – which gives rise to numerous advantages and resources – interact with more proximate determinants like obesity to influence population health.

2. BACKGROUND

2.1. “Amplified” Risk

Extant research on the strong, positive association between individuals’ SES and health suggests that obesity would be less harmful among more educated members of society. While not the central argument underlying fundamental cause theory (FCT), the framework’s perspective on the vital importance of SES (and education in particular) for health-relevant “flexible” resources – “including money, knowledge, prestige, power and beneficial social conditions that can be used to one’s health advantage (Phelan et al. 2004:267)” – proves particularly salient in the context of obesity-related mortality. Specifically, the fact that less-educated adults generally lack these resources may contribute to a process of “ amplified risk accumulation” (Schafer and Ferraro 2011); their limited access to resources for avoiding the harmful consequences of obesity might lead to a higher likelihood of early death, as observed in past research on smoking (Pampel and Rogers 2004), physical inactivity (Krueger and Chang 2008), and alcohol consumption (Probst et al. 2014).

Indeed, past studies have identified many mechanisms by which education is not only inversely associated with individuals’ propensity for becoming obese, but that, once obese, higher-educated adults are better able to mitigate the onset and/or severity of obesity-related morbidities. These include, but are not limited to: greater access to healthcare; increased material resources for alleviating the burden of poor health and/or modifying health behaviors (such as weight loss); reduced stress and inflammation; better neighborhoods providing opportunities for physical activity and quality nutrition; and the tendency for highly-educated adults to benefit first and most from medical innovations (Hummer and Hernandez 2013; Lleras-Muney 2005; Pampel et al. 2010; Rogers et al. 2013; Ross et al. 2012).

Specifically, Miech et al. (2011) conclude that socially-advantaged individuals experience the most immediate and substantial health benefits from medical innovations and interventions – findings mirrored across multiple medical technologies targeting cardiometabolic health (Glied and Lleras-Muney 2008). Relatedly, Chang and Lauderdale (2009) document a downward “social” diffusion of statin use in the 1990s, with higher-SES adults being the first to experience reduced cholesterol levels and improved health. Even taking action in modifying health behavior – such as adhering to dietary/exercise advice from a physician – is contingent upon educational attainment (Mirowsky and Ross 1998). Critically, while any of the above mechanisms disadvantages lower-SES adults, they are likely to co-occur and perpetuate further health decline, helping to explain greater obesity-related disability among low-educated older adults (Schaefer and Ferraro 2011).

Finally, and of particular relevance to the present study, Masters et al. (2015) document large education gradients in “preventable” mortality (i.e., causes of death modifiable through behavior/lifestyle and medical intervention [Phelan et al. 2004]) such as heart disease, a condition closely tied to obesity. Although heart disease mortality risk dropped over past decades, higher-educated adults experienced the steepest declines – likely due to greater access to medical interventions and the ability/resources to pursue more healthful lifestyles.

In sum, the studies noted above suggest that education serves as an effective proxy for the very broad and diverse set of social advantages and flexible resources that higher-educated (and thus, higher-SES) adults may draw upon to improve their health. Even if higher-educated adults already experience some degree of impaired health – such as having central obesity – extant research and theory suggests that their educational attainment would continue to serve this protective function, as those same advantages and resources would help mitigate the severity of their poor health and delay premature death. In turn, this expectation leads to our first testable hypothesis (H1): Central obesity is associated with increased mortality risk among those with lower education compared to higher-educated adults. As lower-educated individuals are more likely to lack many of the advantages and “flexible resources” necessary to address health risks posed by obesity, we expect to observe a stronger relationship between obesity and early death. However, risks are only meaningful in the context of the prevalence of a given risk factor (Northridge 1995), as the higher risk of mortality for one group maybe offset by lower prevalence, resulting in a lower proportion of excess deaths. Thus, using education-specific estimates of obesity-associated mortality risk in conjunction with obesity prevalence allows us to estimate the proportion of excess deaths attributable to central obesity by education group. Per H1, we also expect that obesity accounts for a greater proportion of excess deaths among lower-educated adults.

2.2. Risk “Saturation”

Despite the extensive body of work on education, health mechanisms, and mortality, relatively few studies examine educational variation in the relationship between obesity and mortality. Among this limited body of work, Schnohr et al. (2004) find that obesity-associated mortality risk was lowest among Danish men with no secondary schooling, as compared to their higher-educated counterparts. Similarly, in their U.S.-based analysis, Zheng and Yang (2012) document higher risk among higher-educated adults; though severe obesity was associated with elevated mortality risk for all, the observed hazard ratio of 2.49 for college-educated adults was nearly double that of less-educated respondents (HR=1.34). Zheng and Yang suggest that excess weight may become a health resource for lower-educated adults (e.g., a nutritional reserve during illness) and that confounding from competing causes of death, possibly unrelated to obesity, might explain their lower risk.

Expanding on this issue of ‘competing’ risks, Mehta and Preston’s (2016) recent work provides a theoretical and empirical framework for understanding how sociodemographic and behavioral risk factors interact in their association with mortality. Taking issue with uncritical assumptions about risk relationships in extant research, they offer an overview of mortality risks associated with pairwise combinations of race, gender, education, smoking, and obesity. Critically, their analysis suggests that obesity and low education (defined as a high school degree or less) have a purely “additive” risk relationship; i.e., the obesity-mortality relationship is not contingent upon educational attainment, such that lower-educated adults do not have increased obesity-associated mortality risk.

This finding is less consistent with the “super-additive” (or positive, multiplicative) relationship suggested by the “amplified risk” framework, and more in-line with expectations described by the Blaxter Hypothesis. Blaxter (1990) posits that specific health risks exert less influence on increasing mortality among low-SES groups than their high-SES counterparts, as “unhealthy behaviour does not reinforce disadvantage to the same extent as healthy behaviour increases advantage” (p.233). Specifically, given the many structural factors negatively impacting the health of low-SES individuals (e.g., unsafe housing, work, and neighborhoods [Krueger and Chang 2008]), the ‘additional’ harmfulness of unhealthy behaviors may be lower compared to high-SES adults who are free of these detrimental, structural sources of risk. Schafer and Ferraro (2011) astutely frame this as a matter of “diminishing returns”, such that lower-SES adults have “‘less to lose’ once other health disadvantages are taken into account” and their “risk factors approach saturation” (p.1341). Indeed, Mehta and Preston (2016) would describe these as “sub-additive” (or negative, multiplicative) relations, as additional behavioral exposures (e.g., obesity) are a “redundant” hazard among low-SES adults (p.95).

To this point, research on smoking, alcohol consumption, and adult mortality finds that more disadvantaged groups – as defined by race rather than education – experience a weaker relationship between unhealthy behaviors and mortality (Krueger et al. 2011). Likewise, Schafer and Ferraro (2011) find that educational gradients in inflammation are less pronounced among obese adults relative to their overweight and normal weight counterparts. While this finding is more indicative of “additive” rather than “sub-additive” relations, the lack of SES differences nonetheless reflects the unfortunate reality that high- and low-SES adults may only be on a ‘level playing field’ once they are already obese and at an elevated risk of poor health.

Critically, both additive and sub-additive relationships associated with educational attainment and obesity run counter to the expectations for the interaction of these two sources of mortality risk under the amplified risk framework. Rather, this saturated risk framework suggests that lower-educated adults are likely to encounter many more and varied sources of risk throughout their lives – and thus face competing causes for premature mortality that are unrelated to obesity. The fact that obesity-associated mortality is preventable given adequate resources and advantages is less critical considering these many other plausible explanations. With respect our research hypotheses, we examine whether (H2) central obesity is associated with lower mortality risk among low-educated adults relative to those with higher education. As lower-educated individuals cope with many competing causes of poor health and mortality, obesity may not lead to as great an increase in premature mortality risk compared to their higher-educated counterparts for whom obesity may be a leading explanation for early death. Likewise, central obesity is associated with a lower proportion of excess deaths among low-educated adults. To help better contrast these two sets of hypotheses, the amplified and saturated risk perspectives are shown in Figure 1, illustrating the hypothesized magnitude of the interactions between individuals’ SES (as measured by education in this study) and the extent to which central obesity leads to elevated mortality risk.

Figure 1:

Comparison of Amplified and Saturated Risk Perspectives on Interaction of SES and Central Obesity-Associated Mortality Risk

2.3. Past Limitations and Central Obesity

While the above studies shed much-needed light on the role of education in moderating obesity-associated mortality, they are limited in a few critical areas; this study seeks to fill these gaps. Likely owing to limitations in sample size and/or data availability, prior studies use all-cause mortality as the primary outcome. However, there is substantial variation in the association between obesity and different causes of death (Flegal et. al 2007; Prospective Studies Collaboration 2009) – especially extrinsic causes like homicide or motor vehicle accidents, which are overrepresented among lower-SES adults. Thus, in addition to all-cause mortality, this paper examines deaths from cardiometabolic causes having a more direct, biophysiological association with obesity.

Furthermore, past work has exclusively relied on obesity as defined by body mass index (BMI: weight[kg]/height[m]²). However, researchers have become increasingly concerned with the measurement of obesity in surveys, and the extent to which BMI is able to capture excess fat, rather than just weight, as the underlying mechanism by which obesity negatively impacts health. While BMI is convenient and easy to measure, its lack of generalizability across different demographic groups; varying estimates of health risk based on choice of reference group and weight status cut-points; and, most notably, its inability to describe body shape and body fat accumulation, are important limitations (Burkhauser and Cawley 2008). Differences in the meaning of BMI – with respect to key differences in underlying body composition and cardiometabolic health – may be especially important when examining educational groups, as research shows individuals with higher socioeconomic position may have a more ‘favorable’ distribution of lean to fat body mass (Bann et al. 2014), which may not be reflected in a BMI score.

This is not to suggest that BMI is impossibly flawed; rather, recent research emphasizes the public health significance of measuring individuals’ waists rather than weight in identifying the dangerous concentration of body fat around the internal organs (i.e., visceral fat) (Ashwell and Hsieh 2005; Browning et al. 2010). As compared to BMI, a high waist-to-height ratio (WtHR) has been found to be a more reliable and robust predictor of numerous obesity-related health outcomes – such as diabetes, cardiovascular disease, and other cardiometabolic outcomes (Ashwell and Gibson 2016; Savva et al. 2013) – across different demographics and without the need for group-specific cutoffs (Browning et al. 2010). Increased mortality risk is more strongly associated with waist-based measures (Czernichow et al. 2011), with less effect-modification due to smoking status (Bigaard et al. 2003; Koster et al. 2008), as is a major concern in BMI-based research (Stokes and Preston 2016b). Ashwell et al. (2014) find that the years of life lost attributable to obesity defined by WtHR≥0.5 is greater than based on a comparable level of obesity defined by BMI. As many researchers use a WtHR≥0.5 cutoff for increased risk, empirical evidence supports the “simple message” to “[k]eep your waist circumference… less than half your height” (Ashwell et al. 2014).

3. DATA AND METHODS

3.1. Data

Data come from the National Health and Nutrition Examination Survey (NHANES), a nationally representative survey of the United States valued for its extensive sociodemographic and health questionnaires combined with physiological and anthropometric examinations (National Center for Health Statistics [NCHS] 2017). For these analyses, data are pooled from NHANES III (1988–1994) and continuous NHANES (collected biennially: 1999–2014). Due to concerns about respondents’ privacy, NCHS limits cause of death data in publicly available dat. In order to obtain the most contemporaneous and detailed cause-of-death data available, this study received approval to conduct analyses at a Federal Statistical Research Data Center, where NHANES data are linked to non-perturbed death records through December 31^st, 2015 (NHANES-Linked Mortality File [NHANES-LMF]).

The analytic sample is limited to adults ages 30–74, in order to maximize the likelihood of respondents having completed their schooling, as well as to exclude younger adults who have not yet reached a peak body weight – a key variable in sensitivity analyses. Excluding pregnant women, adults who were not examined, and those not eligible for mortality follow-up, the remaining sample size contains 40,058 adults, of which 6,258 died during the follow-up (444,206 person-years; average follow-up of 11.1 years).

3.2. Measures

The main outcome is mortality, from all causes as well as limited to various causes of death commonly associated with obesity, including: those with hypertension or diabetes as underlying/contributing causes; cardiovascular disease (CVD); and a larger category of “cardiometabolic” deaths based on past literature (CVD; diabetes; colon, rectum, anus/liver and intraheptic bile ducts/pancreas, and prostate cancer; kidney-related conditions; and other obesity-related diseases: Masters et al. 2017). Survival time is based on attained age, defined as the time elapsed from respondent’s age at examination (calculated as day of birth subtracted from their day of examination, or survey if not available) until either their day of death, their 85^th birthday (due to NHANES top-coding individuals’ ages at 85), or the end of the follow-up period (December 31^st, 2015). In sensitivity analyses, respondents were censored at their 75^th or 65^th birthday, with no substantive changes in results (as seen in the Appendix Table B.2).

The primary predictor is respondents’ waist-to-height ratio, calculated from measured waist circumference and height. Based on extant research, we create a dummy variable for a “high” WtHR (≥0.5) indicative of central obesity and elevated health risk (Ashwell et al. 2014). We also tested continuous indicators of WtHR and WtHR², with little change in results but less easily interpretable estimates. Education is categorized as “Less than High School” (<HS), “High School or Some College” (HS/SC), “College or Greater” (BA+), reflecting major levels of educational attainment among U.S. adults. Differentiating between “High School” and “Some College” did not result in a change to the overall findings (i.e., those two categories had similar estimated mortality risks; as seen in the Appendix Table B.3), consistent with research documenting similar health profiles among adults with a sub-baccalaureate education (Zajacova and Johnson-Lawrence 2016). For parsimony, this three-level measure reflects the credential-based focus of the U.S. educational system, and the importance of having a high school and college degree.

Other measures include individuals’ cohort (10-year groups from 1915 to 1984), gender, race/ethnicity (non-Hispanic White, non-Hispanic Black, Mexican-American, Other), nativity, Census region (Northeast, Midwest, South, West), and smoking status (Never, Former, Current), in an effort to choose relevant controls whose effect is relatively constant through the follow-up and to help address issues of confounding relevant to research on obesity, education, and mortality (Stokes and Preston 2016a). In sensitivity analyses we include income-to-needs ratio (based on federal poverty guidelines), weight status based on BMI at survey, and maximum BMI based on highest-ever reported weight. The latter measures of weight status seek to account for confounding due to individuals’ weight histories and their association with mortality (Stokes and Preston 2016a; Stokes and Preston 2016c).

3.3. Methods

We initially estimate the mortality risk associated with central obesity across different causes of death; in testing our hypotheses, we then include an interaction term between WtHR and education to examine its moderating effect in the central obesity-mortality relationship. Due to the complex, multi-stage sampling design of the NHANES data, survey weights are used to accurately estimate variance and account for the selection of respondents. Multiple imputation with chained equations (MICE) – using an iterative series of regressions based on both available and imputed independent variables – retains the full possible analytic sample; 93% of data are complete on the set of sociodemographic covariates used in the main models and 10 imputed datasets are used to ensure sufficient statistical power based on established guidelines for MICE-based estimation (White et al. 2011). Models including income-to-needs, body mass index, and maximum body mass index were estimated using multiple imputation as well. Comparable results were obtained using listwise deletion.

Cox proportional hazard models estimate adjusted mortality risks associated with central obesity, using attained age as the underlying time metric (Thiebaut and Benichou 2004); this allows for a flexible estimation of the baseline hazard rate and the effect of age, well-suited for multiple imputation (Harrell 2015). An assessment of the proportional hazards assumption based on Schoenfeld residuals suggests that the mortality risk associated with central obesity is roughly proportional by age. Further, discrete-time based analyses using a Poisson rate model, as well as competing risks models for deaths limited to CVD or cardiometabolic causes, produced nearly identical estimates. Throughout these analyses, the concept of mortality “risk” (i.e., greater or reduced relative risk for one group or another) refers to differences in the instantaneous hazard rate – or the time to death – rather than absolute differences in the probability of experiencing a mortality event.

Hazard rates obtained from these models are used to calculate population attributable fractions (PAFs), which provide a counterfactual estimate of excess mortality associated with a high WtHR. These counterfactual estimates do not imply a direct causal change in excess mortality; rather, PAFs provide a meaningful estimate of differences in the population-level impact of central obesity that may be missed when comparing relative risks alone, as the interaction of high/low risk and high/low prevalence leads to a range of possible PAFs across education groups. Specifically, PAFs are constructed based on the appropriate, adjusted (i.e, using hazard ratios adjusted for other covariates [Rockhill 1998]) population attributable fraction equation,

$$PA{F}_j=p{d}_j\left(\frac{H{R}_j-1}{H{R}_j}\right)$$

with j representing central obesity and pd and HR as the corresponding proportion of deaths and hazard ratio for central obesity. Group-specific PAFs for central obesity are then calculated for each level of educational attainment.

Finally, the complex interaction of risk and prevalence across educational groups merits further examination; we use Das Gupta decomposition methods (Chevan & Sutherland 2009; Das Gupta 1978) to estimate the extent to which differences in the distribution of and relative risks associated with central obesity explain PAF differences across educational groups. In brief, a Das Gupta decomposition standardizes both mortality risk associated with and prevalence of central obesity across different groups, allowing for the estimation of counterfactual PAF scenarios based on the various combinations of hazard rates and prevalence of central obesity across groups; this enables us to calculate what proportion of the difference in these counterfactual PAFs is driven by cross-group differences in mortality risk compared to overall prevalence of central obesity. More information on the Das Gupta decomposition and statistical procedures is provided in Appendix A.

4. RESULTS

As seen in Table 1, the NHANES-LMF data ensure that results are nationally representative and generalizable to the U.S. adult population. Respondents’ mean age is 49.1; more than half are female (51.4%); 73.5% are non-Hispanic White; just over 15% are foreign-born; and a plurality live in the South (35.0%) – all consistent with Census estimates (Census 2020). The educational composition of the sample is nationally representative as well, with HS/SC-educated adults as the modal group (53.1%). While contemporary estimates find one-in-three adults have a college degree ([Census 2017] vs. 26.3% in this sample), the discrepancy is likely attributable to the longer time period covered in the analyses, with more recent adults having greater education on average. Finally, over half of respondents (52.9%) have ever smoked and 79% have a WtHR≥0.5 – close to the 78% of adults who have “ever” been overweight/obese (based on maximum BMI), but higher than the 66% overweight/obese at survey.

Table 1:

Baseline Characteristics for Sample and Survivors vs. Decedents, 1988–2015 NHANES-LMF (Ages 30–74)

	Overall	Survivors	Decedents
Mean Age at Exam	49.1	47.6	57.6
Cohort
1915–1924	4.5%	2.7%	14.6%
1925–1934	9.3%	5.3%	31.3%
1935–1944	15.2%	13.5%	25.0%
1945–1954	24.9%	26.2%	17.5%
1955–1964	27.6%	30.8%	9.5%
1965–1974	13.4%	15.5%	1.9%
1975–1984	5.1%	5.9%	0.3%
Gender
Female	51.4%	52.5%	45.1%
Male	48.6%	47.5%	54.9%
Race/Ethnicity
Non-Hispanic White	73.5%	73.0%	75.9%
Non-Hispanic Black	10.9%	10.5%	13.0%
Mexican-American	6.0%	6.2%	4.9%
Other	9.6%	10.2%	6.2%
Foreign born	15.3%	16.2%	10.2%
Region
Northeast	19.2%	19.2%	19.0%
Midwest	23.6%	23.6%	24.1%
South	35.0%	34.5%	37.7%
West	22.1%	22.7%	19.2%
Education
Less than HS	20.6%	17.8%	36.1%
HS or Some College	53.1%	53.5%	51.0%
College or Greater	26.3%	28.7%	12.8%
Income-to-Needs Ratio
0–1.00	11.9%	11.0%	17.0%
1.01–1.99	18.7%	17.5%	25.5%
2.00–3.99	33.8%	33.9%	33.4%
4.00+	35.5%	37.5%	24.1%
High Waist-to-Height Ratio (≥0.5)	79.0%	78.0%	84.8%
Maximum BMI Group
Underweight (<18.5)	0.2%	0.2%	0.2%
Normal (18.5–24.9)	21.8%	22.5%	17.6%
Overweight (25.0–29.9)	35.2%	35.7%	32.1%
Obese, Class I (30.0–34.9)	24.2%	23.8%	26.8%
Obese, Class II/III (35.0+)	18.6%	17.8%	23.3%
BMI at Exam Group
Underweight (<18.5)	1.7%	1.4%	2.9%
Normal (18.5–24.9)	32.6%	32.9%	30.6%
Overweight (25.0–29.9)	34.3%	34.7%	31.6%
Obese, Class I (30.0–34.9)	18.6%	18.4%	19.9%
Obese, Class II/III (35.0+)	12.9%	12.5%	14.9%
Smoking Status
Never	47.1%	50.4%	28.6%
Former	27.7%	26.5%	34.3%
Current	25.2%	23.1%	37.1%
Sample size	40,058	33,800	6,258

Notes:

Estimates based on weighted and imputed values.

Follow-up age censored at 85.

Bold indicates significant difference between decedents and survivors at p<0.05, based on two-tailed t-test.

There are notable differences in the sociodemographic and health profiles of those surviving the follow-up period as compared to decedents. On average, decedents are 10 years older (57.6 vs. 47.6) and significantly more likely to be male, non-Hispanic Black, and U.S.-born. Adults with less than a high school education are overrepresented among decedents as well, accounting for 36% of those who died as compared to 18% of survivors. Conversely, college-educated adults – representing 29% of survivors – only account for 13% of decedents. Adults dying during the follow-up were also generally ‘unhealthier’: 85% had a high WtHR, 82% were ever overweight/obese, and 71% were former or current smokers. By contrast, 78% of survivors had a high WtHR, 77% were ever overweight/obese, and 50% ever smoked.

Prior to conducting survival analyses, we examined baseline differences in sociodemographic and health measures across the educational groups (included in the Appendix Table B.1). Immediately apparent are the significant educational differences in the proportion of adults dying during the follow-up: While only 7.4% of college degree-holding adults died, this proportion was nearly double among HS/SC-educated adults (14.5%), and more than triple among <HS-educated adults (26.4%); however, it is important to note that college-educated adults were four years older, on average, than their <HS-educated counterparts. Higher-educated adults also had more healthful measures of adiposity: approximately 70% of college-educated adults had a high WtHR or were overweight/obese at time of exam, compared to ~82% of less-educated adults. Smoking prevalence was significantly lower among college-educated adults as well.

Examining the interaction between central obesity and educational attainment (Table 2) reveals that a high WtHR is associated with a significantly greater risk of mortality among college-educated adults compared to their less-educated counterparts – a result more in line with H2, or the “saturated/sub-additive” risk hypothesis. While central obesity is associated with an increased likelihood of premature death for all education groups, the relative risk is greatest among the college-educated, and comparable among those with HS/SC or <HS education. The degree of moderation in this relationship is smallest for all-cause mortality: Those with HS/SC education have approximately 30% lower central obesity-associated all-cause mortality risk than their college-educated counterparts (p<0.05), while <HS-educated adults have approximately 40% lower risk (p<0.01). While there is no evidence of moderation for deaths where hypertension/diabetes are underlying conditions, we observe the same pattern of lower relative mortality risk among lower-educated adults when using a higher WtHR threshold of 0.6 (as discussed in the sensitivity analyses), which suggests that certain cause-specific mortality risks may be more sensitive to the ‘severity’ of central obesity.

Table 2:

High Waist-to-Height Ratio (≥0.5) Mortality Risk, 1988–2015 NHANES-LMF (Ages 30–74)

	All-Cause				Underlying Hyp./Diab.				CVD				Cardiometabolic
	HR		95% CI		HR		95% CI		HR		95% CI		HR		95% CI
High Waist-to-Height Ratio (≥0.5)	1.62	**	1.24	2.12	2.47		0.90	6.81	4.40	***	2.02	9.57	3.00	***	1.81	4.97
High WtHR X Education (ref. BA+)
Less than HS	0.62	**	0.44	0.88	0.92		0.31	2.78	0.29	*	0.12	0.74	0.47	*	0.26	0.85
HS or Some College	0.70	*	0.51	0.96	0.64		0.22	1.92	0.26	**	0.12	0.60	0.41	**	0.23	0.74
Education (ref. BA+)
Less than HS	3.02	***	2.17	4.19	2.97	*	1.08	8.22	6.68	***	2.91	15.34	4.27	***	2.49	7.33
HS or Some College	2.14	***	1.60	2.86	2.82	*	1.01	7.86	5.32	***	2.54	11.16	3.51	***	2.09	5.90
Cohort (ref. 1945–1954)
1915–1924	1.26		0.99	1.60	1.00		0.68	1.46	2.49	***	1.67	3.71	1.65	**	1.19	2.31
1925–1934	1.21		0.97	1.52	1.10		0.77	1.57	1.82	**	1.23	2.69	1.31		0.95	1.80
1935–1944	1.04		0.85	1.27	0.97		0.70	1.34	1.35		0.94	1.93	1.08		0.84	1.40
1955–1964	0.91		0.73	1.13	1.24		0.77	2.01	0.73	*	0.54	0.98	0.86		0.68	1.10
1965–1974	1.22		0.85	1.73	1.40		0.57	3.44	0.80		0.52	1.24	1.25		0.74	2.11
1975–1984	1.36		0.60	3.07	2.35		0.29	19.34	2.06		0.55	7.65	1.79		0.67	4.79
Female	0.74	***	0.68	0.80	0.87		0.72	1.05	0.65	***	0.56	0.75	0.65	***	0.57	0.74
Race/Ethnicity (ref. NH White)
NH Black	1.31	***	1.20	1.42	1.67	***	1.39	2.00	1.51	***	1.32	1.73	1.62	***	1.44	1.84
MX-American	1.23	**	1.09	1.38	1.38	**	1.13	1.69	1.15		0.91	1.44	1.26	*	1.05	1.52
Other	0.98		0.82	1.17	1.10		0.78	1.54	0.99		0.72	1.36	1.01		0.78	1.30
Foreign born	0.80	**	0.70	0.91	0.68	**	0.53	0.88	0.79		0.63	1.00	0.80		0.64	1.00
Region (ref. Northeast)
Midwest	1.05		0.87	1.26	1.26		0.98	1.63	0.97		0.76	1.25	0.99		0.80	1.22
South	1.15	*	1.01	1.32	1.49	**	1.14	1.94	0.98		0.78	1.22	1.05		0.87	1.25
West	1.05		0.91	1.20	1.55	**	1.14	2.11	0.98		0.74	1.30	0.98		0.78	1.22
Smoking Status (ref. Never)
Former	1.38	***	1.25	1.52	1.29	*	1.02	1.63	1.26	**	1.08	1.48	1.33	***	1.15	1.54
Current	2.55	***	2.34	2.78	2.37	***	1.95	2.88	2.20	***	1.90	2.53	2.06	***	1.83	2.33

Notes:

N=40,058.

Age at death restricted to 85.

Significance based on two-tailed t-test, indicated as:

^*p<0.05;

^**p<0.01;

^***p<0.001.

In further support of H2, educational variation is greater for CVD and cardiometabolic-related causes of death. While baseline mortality risk associated with central obesity is higher for both causes, adults with less than a college degree have only one-third the relative central obesity-related CVD mortality risk of their college-educated peers, and less than half the relative risk of cardiometabolic mortality. However, the mortality risk associated with lower education (relative to BA+) is significantly higher for both causes: Among adults with WtHR<0.5, those with less than a college degree have five to six times greater relative CVD morality risk, and three to four times greater cardiometabolic mortality risk.

Finally, we examine the population-level impact of central obesity-associated mortality risk by estimating educational differences in the proportion of excess deaths attributable to a high WtHR. Higher mortality risk associated with a health hazard is only meaningful if said hazard also affects a large proportion of the population; this is especially important when both risk and prevalence may vary across groups. Consequently, Table 3 presents estimates of population-attributable fractions (PAFs) for excess mortality by educational attainment, as well as the results of a Das Gupta decomposition which quantifies the proportion of differences in PAFs due to variation in risk as compared to prevalence. Comparing two educational groups at a time, the “Total PAF Difference” is a function of differences in both prevalence and risk. The “absolute contribution” indicates the counterfactual PAF difference assuming either risk or prevalence is held constant across groups, while the “proportional contribution” shows how much of the difference is due to either component PAF input.

Table 3:

Decomposing Contribution of High Waist-to-Height Ratio (≥0.5) Prevalence and Risk to Educational Differences in Population Attributable Fractions for Mortality, 1988–2015 NHANES-LMF (Ages 30–74)

	Difference between College or Greater and Less than High School
	All-Cause			Underlying Hypertension/Diabetes			Cardiovascular Disease			Cardiometabolic Diseases
	% of Dead w/C.O.a	HR	PAF	% of Dead w/C.O.	HR	PAF	% of Dead w/C.O.	HR	PAF	% of Dead w/C.O.	HR	PAF
Education
BA+	83.9%	1.62	32.1%	88.8%	2.47	52.8%	93.7%	4.40	72.4%	91.0%	3.00	60.6%
<HS	86.4%	1.01	0.7%	94.2%	2.28	52.9%	90.0%	1.29	20.4%	90.5%	1.40	25.8%
Total PAF Difference			31.4%			0.0%			52.0%			34.8%
Das Gupta Decomp.	Abs. Cont.b	Prop. Cont.c		Abs. Cont.	Prop. Cont.		Abs. Cont.	Prop. Cont.		Abs. Cont.	Prop. Cont.
Prevalence	-0.5%	-1.5%		-3.1%	-		1.9%	3.6%		0.2%	0.6%
Risk	31.9%	101.6%		3.1%	-		50.1%	96.4%		34.6%	99.4%
	Difference between College or Greater and Less than High School
	All-Cause			Underlying Hypertension/Diabetes			Cardiovascular Disease			Cardiometabolic Diseases
	% of Dead w/C.O.	HR	PAF	% of Dead w/C.O.	HR	PAF	% of Dead w/C.O.	HR	PAF	% of Dead w/C.O.	HR	PAF
Education
BA+	83.9%	1.62	32.1%	88.8%	2.47	52.8%	93.7%	4.40	72.4%	91.0%	3.00	60.6%
HS or SC	83.5%	1.13	9.7%	87.4%	1.59	32.5%	84.7%	1.16	11.6%	85.5%	1.24	16.5%
Total PAF Difference			22.4%			20.4%			60.8%			44.1%
Das Gupta Decomp.	Abs. Cont.	Prop. Cont.		Abs. Cont.	Prop. Cont.		Abs. Cont.	Prop. Cont.		Abs. Cont.	Prop. Cont.
Prevalence	0.1%	0.4%		0.7%	3.3%		4.1%	6.7%		2.4%	5.4%
Risk	22.3%	99.6%		19.7%	96.7%		56.7%	93.3%		41.7%	94.6%
	Difference between College or Greater and Less than High School
	All-Cause			Underlying Hypertension/Diabetes			Cardiovascular Disease			Cardiometabolic Diseases
	% of Dead w/C.O.	HR	PAF	% of Dead w/C.O.	HR	PAF	% of Dead w/C.O.	HR	PAF	% of Dead w/C.O.	HR	PAF
Education
HS or SC	83.5%	1.13	9.7%	87.4%	1.59	32.5%	84.7%	1.16	11.6%	85.5%	1.24	16.5%
<HS	86.4%	1.01	0.7%	94.2%	2.28	52.9%	90.0%	1.29	20.4%	90.5%	1.40	25.8%
Total PAF Difference			9.1%			20.4%			−8.8%			−9.3%
Das Gupta Decomp.	Abs. Cont.	Prop. Cont.		Abs. Cont.	Prop. Cont.		Abs. Cont.	Prop. Cont.		Abs. Cont.	Prop. Cont.
Prevalence	-0.2%	-2.0%		-3.2%	15.6%		-0.2%	-0.2%		-1.2%	13.1%
Risk	9.3%	102.0%		-17.2%	84.4%		9.3%	9.3%		-8.1%	86.9%

Notes:

N=40,058.

Age at death restricted to 85.

^aCentral Obesity is abbreviated as “C.O.”, indicative of a waist-to-height ratio ≥0.5.

^bAbsolute Contribution to differences in PAFs.

^cProportionate Contribution to differences in PAFs.

Although the prevalence of central obesity is similar among college-educated adults as compared to less-educated groups, their significantly elevated risk of mortality results in greater PAFs – also more consistent with the PAF prediction under the saturated/sub-additive risk hypothesis (H2). PAF differences are smallest when examining all-cause mortality and deaths with underlying hypertension/diabetes. For instance, the all-cause mortality PAF among those with a college education (32.1%) is 22.4 points higher compared to those with HS/SC education (9.7%), and 31.4 points higher than those with <HS education (0.7%). The PAF for college-educated adults is even higher for deaths with underlying hypertension/diabetes (52.8%), but group differences are smaller as the equivalent PAF for those with HS/SC education is 20 points lower (32.5%) and the PAF for those with <HS education is nearly identical (52.9%). Conversely, deaths with underlying hypertension/diabetes have the largest PAF difference (20.4%) between those with HS/SC or <HS education.

The PAF differences grow considerably when examining causes of death more closely associated with central obesity and its related morbidities, primarily driven by higher PAFs for college-educated adults. For CVD mortality, the PAF among college-educated adults is 72.4%, which is 60 and 52 points higher than those with HS/SC or <HS education, respectively. This pattern extends to cardiometabolic disease-related causes of death – albeit differences are smaller – as the 60.6% PAF among college-educated adults is approximately 40 points higher than their less-educated counterparts.

The Das Gupta decomposition underscores how much of the higher PAF among college-educated adults is attributable to elevated central obesity-associated mortality risk as compared to its prevalence among decedents. For every cause of death under consideration, the majority – if not entirety – of the PAF difference between college-educated and others is accounted for by higher risk. The consistently small absolute and proportional contributions corresponding with prevalence suggest that college-educated adults would instead have a similar PAF compared to their lower-educated counterparts if risks were held equal.

Sensitivity Analyses

In addition to the previously mentioned sensitivity tests using earlier ages of death and a four-category measure of education (Appendix Tables B.2 and B.3), we assess the robustness of results to alternate specifications of central obesity, the inclusion of additional measures of SES and weight status, and other causes of death, as seen in Appendix Table B.4. The first analysis uses a more conservative cutoff for a risky WtHR≥0.6 (Ashwell et al. 2014); this limits the proportion of the population ‘at-risk’ (~35%) – and resulting PAFs – but the effect moderation of education is similar. While the interaction is attenuated for all-cause, CVD, and cardiometabolic mortality risk, we observe a significant interaction effect for mortality risk when hypertension/diabetes are underlying conditions. The second analysis introduces individuals’ past and current weight status – based on BMI – as a proxy measure of “weight history” (Stokes and Preston 2016c) to help account for bias due to illness-related weight loss, which may not be captured by the single measure of WtHR available in the NHANES data. However, the inclusion of these measures has little effect on the original estimates. Similarly, including income-to-needs ratio in the third analysis – as a more direct measure of a type of flexible resource that may be particularly important for maintaining good health – has no effect either, suggesting that education does not merely reflect individuals’ economic status in these analyses.

The fourth analysis attempts to replicate the results using maximum BMI rather than WtHR. The unstable estimates reflect the small cell sizes for deaths among the various combinations of education and weight status, even in a survey as large as NHANES. Though these findings are somewhat inconclusive, some of the point estimates for the interactions between lower education and obesity are similar to those in the original results. Indeed, the obesity-associated CVD mortality risk among <HS-educated adults is approximately half that of their college-educated counterparts, though the interaction term is insignificant.

Finally, we ran models examining deaths due to accidents (e.g., falls, drownings, fires and explosions, accidental discharge of firearms, non-motor vehicle transport accidents, etc.) as a kind of “placebo test,” to demonstrate how educational differences in obesity-associated mortality risk may be less salient for causes of death having very little to do with one’s obesity status, and generally somewhat random in nature (Appendix Table B.5). While the number of deaths from such accidents is quite small, we clearly see that there are no educational differences in the relationship between central obesity and these causes of death. Moreover, looking at accidents independently of other causes helps further explain why we see smaller differences in PAFs for all-cause mortality, which encompasses accidents and other potentially ‘irrelevant’ causes of death in terms of their association with central obesity.

5. DISCUSSION

Population health in the U.S. is increasingly defined by growing socioeconomic inequality in mortality and rising rates of obesity and related comorbidities. However, owing to variation in individuals’ health resources and risks on the basis of their SES, there may be differences in the relative population-level impact of “preventable” sources of mortality such as obesity. Examining these relative differences is critical for understanding who stands to benefit most from public health efforts targeting a specific health behavior or condition. To fill this gap, the present study examined central obesity-associated mortality risk across different levels of educational attainment among U.S. adults. Findings revealed significant variation in mortality risk, contributing to large differences in the population attributable fractions for excess deaths associated with central obesity when comparing college-educated adults to those with lower educational attainment.

Adults at risk for central obesity on the basis of WtHR had anywhere from 15 to 87% higher risk of mortality across different causes of death – comparable to risks associated with being a former smoker – with significantly elevated risk for deaths closely linked to cardiometabolic health. College-educated adults had significantly higher relative central-obesity associated mortality risk compared to both their less-educated counterparts, similar to patterns observed for BMI-based measures in the work of Schnohr et al. (2004) and Zheng and Yang (2012), but differing from the results of Mehta and Preston (2016). For deaths due to cardiovascular or caridometabolic disease, college-educated adults had more than triple and double the relative risk, respectively; sensitivity analyses show that college-educated adults have higher relative risk for causes with underlying hypertension/diabetes as well, albeit at a higher WtHR threshold. Consequently, the PAF for central obesity was larger among highly-educated adults for all-cause mortality, where one-in-three excess deaths would be averted if adults’ WtHRs were below 0.5, compared to only one-in-ten deaths among HS/SC-educated and less than 1-in-100 deaths for adults with <HS education. These differences were even larger among cardiovascular disease (PAF: 72% for college or more vs. ≤20% for less-educated groups) and cardiometabolic deaths (PAF: 61% vs. ≤26% for less-educated groups). New to the literature, a Das Gupta decomposition demonstrated that educational differences in mortality risks associated with central obesity almost entirely explain the large PAF disparity.

Prior to discussing the implications of these results, we consider important limitations of the study and how they may be addressed in future research. Though WtHR is a validated measure of central obesity, there are other anthropometric markers of an unhealthy body shape and composition, such as body-fat percentages, skinfolds, waist-to-hip ratios, and appendage circumference (Burkhauser and Cawley 2008). WtHR data are most complete in NHANES, though future research may consider how these results compare across other measures of excess adiposity measured over multiple points in time. More critically, we are not able to fully account for confounding in the obesity-mortality relationship, such as the role of smoking – which leads to weight loss, yet higher mortality (Flegal 2007) – as well as reverse causation due to formerly obese adults losing weight attributable to weight-related illness, both of which may artificially inflate mortality risks among the referent “non-obese” category (Flegal et al. 2011; Stokes and Preston 2016a; Stokes and Preston 2016b).

With respect to smoking, we examined the interaction of education and central obesity-associated mortality risk among non-smokers, but results were inconclusive as the cell sizes for these interaction categories become quite small. Moreover, it is unclear if selection plays a role as well given that non-smokers are not representative of the U.S. population and do not yield generalizable results (Stokes and Preston 2016a), especially among lower-educated adults who generally have a much lower propensity to be non-smokers (Gilman et al. 2008). Recent work has cleverly employed maximum BMI to address the issue of reverse causality (Stokes and Preston 2016a; Stokes and Preston 2016c) and we incorporate this measure in sensitivity analyses. While the primary results were unchanged, re-analyzing the data using maximum BMI as a measure of obesity in lieu of WtHR was inconclusive. Whether this is attributable to cell-size issues, or substantive differences in the meaning of measures across groups, requires further study – especially as illness-related weight loss is more common among less-educated adults (Vierboom 2017), and may downwardly bias central obesity-associated mortality among these groups, for whom the “healthy” referent weight category is potentially quite unhealthy.

Indeed, this speaks to a broader issue of whether the obese and non-obese populations across different educational groups are comparable in terms of health and selectivity; future work may consider whether high-SES individuals who are obese – as well as low-SES individuals who are non-obese – are fundamentally different groups, and whether obesity represents equal states of “poor health” (based on other measures) regardless of SES. Nevertheless, there is emerging evidence to suggest that confounding due to reverse-causality may be less troublesome for waist-based measures, due to a genetic predisposition to attain – and retain – a certain body shape (Zhang et al. 2018), or the fact that weight loss and reductions in waist size do not always move in tandem (Freedman et al. 2015; Gearon et al. 2018; Peeters et al. 2014). However, longitudinal studies – with measures of waist circumference and a sufficient count of deaths – would be ideal for addressing this issue, as well as accounting for any left-censoring that may occur from particularly ill individuals dying before being sampled in a cross-sectional survey like NHANES.

Future studies could also examine gender- and race/ethnicity-based differences in the associations observed in this study. Our study stratified results by gender and observed no major differences between women and men; however, cell sizes were considerably smaller and thus our estimates were less stable. With respect to race/ethnicity, we observe the same patterns among non-Hispanic White and Black adults, which is somewhat unexpected given the shallower education and health/mortality gradient among Black adults in the U.S., for whom more education does not necessarily translate to greater resources and lower health risks (Masters et al. 2012; Montez et al. 2010; Kimbro et al. 2008; Vable et al. 2018; Zajacova and Hummer 2009). Indeed, despite the smaller sample sizes, the findings are very robust for both groups across different causes of death. Patterns among other race/ethnic groups and causes of death are less clear, but this preliminary evidence strongly suggests that future analyses should consider race/ethnic differences in the relationship between education, obesity, and mortality, as well as other variations across different surveys, populations, and contexts.

Despite these limitations, the present study highlights the population health significance of educational variation in the obesity-mortality relationship. Greater WtHR-associated mortality risk and PAFs for highly-educated adults favors the “saturated risk” hypothesis – that mortality risks associated with obesity are lower for lower-educated adults due to other, competing sources of risk for early death. As highly-educated individuals represent a more advantaged population facing fewer hazards in their lives and environments, the risk associated with central obesity is magnified given a lack of these extraneous threats to health and causes of death, accounting for a greater proportion of excess deaths. Moreover, the fact that educational differences in relative mortality risk and PAFs are larger for causes most commonly associated with central obesity suggests that competing risks not only vary by education and across specific causes of death.

Recent trends in U.S. health and life expectancy underscore this issue of competing risks; multiple studies have documented the stark contrast in leading causes of death based on educational attainment – most notably accidental poisonings and suicide that have little to no association with obesity and disproportionately affect lower-educated individuals at younger ages (Case and Deaton 2015; Sasson 2016; Sasson and Hayward 2019). This educational divergence in mortality trends reduces the likelihood that the cumulative toll of central obesity is a primary risk factor for early death among lower-educated adults. Central obesity may be the primary risk factor for mortality among highly-educated adults, as compared to only one of many risk factors among those with less education; consequently, the additional risk presented by obesity for this group is small relative to existing risks. This important point is highlighted in Figure 2, illustrating the cumulative mortality risks associated with educational attainment, central obesity, and their interaction. Despite elevated relative mortality risk associated with central obesity among college-educated respondents, it is consistently lower than non-central obese mortality risk among less-educated respondents.

Figure 2:

Relative Mortality Risk by Education and Central Obesity, NHANES 1988–2015 (Ages 30–74)

Lower-educated individuals may experience mortality from causes associated with but not resulting from central obesity, and that are instead attributable to adverse social circumstances. While obesity is associated with increased risk for poor cardiovascular health – inclusive of hypertension, diabetes, CVD, and other cardiometabolic disease – research finds numerous other socioenvironmental factors that contribute to educational differences in the prevalence of these conditions. Unregulated blood pressure, due to deleterious psychosocial exposures and increased stress; the co-incidence of hypertensive diseases, such as diabetes, due to diet and lifestyle; and reduced socio-emotional support and networks all increase the risk of poor cardiovascular health for those with low education (Matthews et al. 1989; Ross et al. 2012; Shim 2014). In fact, qualitative differences in these sources of risk may contribute to SES differences in risk of subsequent mortality as well (Wu et al. 2013). Moreover, the prevalence of metabolic syndrome is greater among lower-SES adults (Loucks et al. 2007a; Loucks et al. 2007b), suggesting that some degree of underlying metabolic dysregulation – which is associated with a larger waist circumference along with hypertension, hyperglycemia, and dyslipidemia – constitutes a greater risk among more disadvantaged adults than central obesity in isolation. Though this study is not able to fully disentangle the various competing and co-occurring sources of health risk across educational groups, our results are consistent with the notion that central obesity is but one piece of the much larger constellation of risks encountered by low-educated adults.

Knowledge of relative differences is critical for anticipating the success or failure of a given health intervention, especially among the less-advantaged members of the population often targeted by these initiatives. Namely, support for the “saturated risk” hypothesis shows how distal and fundamental inequalities in SES inhibit the effectiveness of efforts to address more proximate and cascading determinants of health, as is the case with central obesity and mortality. For instance, past research on smoking and mortality has shown that solely focusing on curbing tobacco use among low-SES groups may be unlikely to result in significant health or longevity benefits as expected (or desired) given the competing sources of stress and danger in their lives (Bosma et al. 1999; Pampel and Rogers 2004). Population-wide initiatives aimed at reducing smoking behavior may instead exacerbate existing health inequalities by failing to address deeper-seeded causes (Lawlor et al. 2003). With smoking on the decline, and obesity on the rise, anti-obesity efforts should avoid similar pitfalls.

Indeed, these results suggest current efforts to directly target obesity may disproportionately benefit highly-educated adults. Anti-obesity policies and interventions can and should contribute to absolute declines in mortality across all SES groups – which has tremendous public health value, given the impact of obesity on morbidity and mortality. However, relative difference would likely grow, perpetuating overall educational inequalities in mortality; this widening of differences may be especially severe if the aforementioned policies and interventions are contingent upon “individual resources or initiative” to be successful (Freese and Lutfey 2011: 76), such as investing time and money to improve one’s diet or engage in greater physical activity.

Critically, there is a need for interventions that do not directly place the onus on individuals to take action in reducing their body size, as actions require resources that lower-educated individuals tend to lack. Many obesity interventions in the U.S. are well-intentioned in their desire to promote a healthier lifestyle; however, these interventions require a high level of agency, as individuals tasked with absorbing this information and using it to improve their health (Adams et al. 2016). Such interventions are far more actionable among higher-educated adults, who have the resources necessary to make these changes; however, successful interventions requiring “breaking the bonds” between SES and health, such that one’s level of education does not dictate the success of a given intervention (Freese and Lutfey 2011; Link and Phelan 1995). As others have noted (Novak and Brownell 2011; Novak and Brownell 2012), structural and community-level interventions – which reduce the importance of “agency” (Adams et al. 2016) – are far more likely to have a population-level impact by reducing the barriers to healthier behaviors among lower-SES adults. For example, increasing access and affordability to more nutritious foods, such as reducing costs through the use SNAP or WIC benefits (Leung et al. 2014; Leung et al. 2017), or facilitating easier access outdoor activities and physical activity in low-SES communities by investing in improvements to the infrastructure and built environment in low-SES communities (Durand et al. 2011; Sallis et al. 2012), are both strategies that effect behavioral change while minimizing the need for additional time, money, or effort. In effect, obesity-specific interventions may be most effective when they bridge – rather than exacerbate – the resource gap between high- and low-educated adults.

Yet this study emphasizes that an ‘unhealthy’ waist-to-height ratio is likely not the only risk factor contributing to mortality from heart disease, hypertension, diabetes, and many other conditions; likewise, mortality risk associated with central obesity is not uniformly distributed throughout the population. Breaking the ties between SES and obesity-related health is no panacea for resolving the issue of competing risks to mortality. Current efforts to target obesity only target obesity, and often ignore the larger social and structural factors that shape mortality both through and independent of obesity. Chronic stress and uncertainty lead to harmful behaviors associated with obesity and poor health (Scott et al. 2012); however, they give rise variety of other harmful sequelae that increase both overall and cause-specific mortality (Nielsen et al. 2008; Russ et al. 2012). Addressing these more fundamental issues of how low SES and education leads to poor health is far more complicated than eliminating obesity. Improving the quality and length of individuals’ lives requires addressing a multitude of different health risks and pathways; indeed, this is a central premise of fundamental cause theory and other frameworks for understanding social determinants of health, as low SES leads to poor health via multiple, replaceable mechanisms and pathways (Link and Pheland 1995). Ideally, we can target many of these mechanisms at once by addressing the underlying structural issues at work.

In the contemporary U.S. – and as emphasized in these analyses – education is one of the foremost structural factors shaping individuals’ lives. Being low-educated is in and of itself a health risk, to the extent that it exposes one to the kinds of harmful social environments and scenarios that engender harmful health behaviors and exposures that may lead to early death (Zajacova and Lawrence 2018). Yet lower educational attainment presents independent risks to health as well. “Education allows individuals to make sense of the world around them, and understanding the world empowers individuals to make changes effectively” (Cohen and Symes 2013: 999); higher-educated individuals are better able to grapple with instability and minimize the uncertainty of day-to-day life, which helps minimize the kinds of chronic and accumulated stress that is implicated in so many disease and mortality processes. Health researchers are well-attuned to this reality, and numerous solutions have been offered, ranging from targeting educating itself via policies that lead to greater educational attainment in the U.S. as a whole (Hahn and Truman 2015), to policies that minimize the importance of education for better health by decoupling education from key health resources and mechanisms (Hummer and Hernandez 2013). Regardless of the approach, reducing the extent to which education stratifies health in the U.S. should take precedent over our current focus on behavioral risk factors like obesity.

Indeed, policymakers and public health advocates should be more careful in generalizing about “leading threats” to population health, which may give the wrong impression that we can achieve universal improvements in health through reducing obesity-associated morbidity and mortality. Obesity is a key population health risk that affects the lives of many adults regardless of their social strata, but we must also acknowledge group differences in the kinds and variety of mortality risks they encounter, and the fundamental sources of these risks. Decades of research shows that lower-SES adults are subject to multiple and intersecting risks to their health and wellbeing; failing to recognize education as a key distal determinant of these risks helps to explain why decades of progress in targeting proximate causes of risk like obesity has not reduced educational disparities in mortality.

RESEARCH HIGHLIGHTS.

• Obesity is a leading preventable cause of mortality in the United States.

• Educational attainment may be a key moderator of the obesity-mortality relationship.

• Central obesity accounts for 30–70% of excess deaths among college-educated adults.

• Less-educated adults have lower obesity-associated mortality risk and proportion of excess deaths.

• Population-wide anti-obesity efforts may widen educational health disparities.

APPENDIX A

The advantages of the Das Gupta decomposition are twofold: (1) Das Gupta’s approach is reference invariant, using an average of two decompositions referenced to each comparison group, thus solving the “index problem” whereby results are contingent upon the choice of referent group; and (2) this method allows for more easily interpretable results, as the separate effect of the interaction between components (e.g. the “double exposure” of being centrally-obese and low education) is instead incorporated into the main additive effects of the two components (Chevan & Sutherland 2009; Das Gupta 1978; Li 2017). Standardization of both mortality risk associated with and prevalence of central obesity across education groups is critical to the Das Gupta decomposition, as it allows us to estimate counterfactual PAF scenarios based on the various combinations of hazard rates and prevalence of central obesity across groups, and then calculate what proportion of the difference in these counterfactual PAFs is driven by cross-group differences in mortality risk compared to overall prevalence of central obesity.

Using the recently available -rdecompose- procedure in Stata (Li 2017) – based on the Das Gupta decomposition technique – allows for percent estimates of the relative contributions of differences in obesity distributions and differences in relative risks associated with central obesity to overall PAF differences between high- and low-educated adults. The estimate of the percent contribution of differences in the prevalence of central obesity between high- and low-educated adults to the overall PAF difference is given as

$${\omega }_{Prev.}=\frac{Ris{k}_{Hi\hspace{0.17em}Ed}+Ris{k}_{Lo\hspace{0.17em}\hspace{0.17em}Ed}}{2}\left(Prev{.}_{Hi\hspace{0.17em}Ed}-Pre{v}_{Lo\hspace{0.17em}\hspace{0.17em}Ed}\right),$$

where the relative risk is standardized between the two groups. Differences in the relative risk associated with central obesity between high- and low-educated adults contributes to the remained of the PAF difference, and is given as

$${\omega }_{Risk.}=\frac{Pre{v}_{Hi\hspace{0.17em}Ed}+Pre{v}_{Lo\hspace{0.17em}\hspace{0.17em}Ed}}{2}\left(Risk{.}_{Hi\hspace{0.17em}Ed}-Ris{k}_{Lo\hspace{0.17em}\hspace{0.17em}Ed}\right),$$

where the central obesity prevalence rate is now standardized between the two groups, with the relative contributions both summing to 100%.

APPENDIX B

Appendix Table B.1:

Educational Differences in Baseline Characteristics, 1988–2015 NHANES-LMF (Ages 30–74)

	<HS N≈8,250	HS or SC N≈21,274	BA+ N≈10,534
Died During Follow-up	26.4%	14.5%	7.4%
Mean Age at Exam	51.8	48.6	47.9
Cohort
1915–1924	8.8%	3.7%	2.7%
1925–1934	15.4%	8.7%	5.4%
1935–1944	18.0%	15.1%	13.3%
1945–1954	19.4%	25.1%	28.9%
1955–1964	22.7%	29.4%	27.9%
1965–1974	11.7%	13.2%	15.3%
1975–1984	3.9%	4.8%	6.4%
Gender
Female	50.3%	53.8%	47.4%
Male	49.7%	46.2%	52.6%
Race/Ethnicity
NH White	53.4%	76.6%	82.8%
NH Black	15.7%	11.4%	6.2%
MX-American	16.5%	4.0%	1.7%
Other	14.3%	7.9%	9.3%
Foreign born	28.2%	10.6%	14.7%
Region
Northeast	17.4%	18.4%	22.3%
Midwest	20.0%	25.2%	23.4%
South	41.7%	34.4%	31.0%
West	20.9%	22.0%	23.4%
Income-to-Needs Ratio
0–1.00	28.7%	10.0%	2.5%
1.01–1.99	33.5%	19.0%	6.6%
2.00–3.99	28.2%	39.8%	26.2%
4.00+	9.6%	31.1%	64.7%
High Waist-to-Height Ratio (≥0.5)	85.3%	80.2%	71.6%
Maximum BMI Group
Underweight (<18.0)	0.2%	0.1%	0.4%
Normal (18.0–24.9)	16.8%	20.2%	28.8%
Overweight (25.0–29.9)	34.4%	33.8%	38.4%
Obese, Class I (30.0–34.9)	27.3%	25.2%	20.0%
Obese, Class II/III (35.0+)	21.3%	20.7%	12.4%
BMI at Exam Group
Underweight (<18.0)	2.3%	1.3%	1.9%
Normal (18.0–24.9)	28.2%	30.6%	40.0%
Overweight (25.0–29.9)	34.8%	34.0%	34.4%
Obese, Class I (30.0–34.9)	21.0%	19.4%	15.3%
Obese, Class II/III (35.0+)	13.7%	14.8%	8.4%
Smoking Status
Never	38.9%	43.8%	60.3%
Former	25.9%	28.1%	28.2%
Current	35.2%	28.1%	11.5%

Notes:

N=40,058.

Education group (N) vary slightly among imputations; count is approximate.

Follow-up age censored at 85.

Bold indicates significant difference between college-educated and other educational group at p<0.05, based on two-tailed t-test.

Appendix Table B.2:

High Waist-to-Height Ratio (≥0.5) Mortality Risk for Death Before Age 75 or Age 65, 1988–2015 NHANES-LMF

	Before Age 75				Before Age 65
	HR		95% CI		HR		95% CI
High Waist-to-Height Ratio (≥0.5)	1.69	**	1.22	2.34	1.88	*	1.10	3.20
High WtHR X Education (ref. BA+)
Less than HS	0.58	*	0.38	0.90	0.62		0.32	1.22
HS or Some College	0.68	*	0.47	0.99	0.70		0.40	1.24
Education (ref. BA+)
Less than HS	3.50	***	2.36	5.17	3.97	***	2.02	7.83
HS or Some College	2.27	***	1.62	3.19	2.42	**	1.35	4.34
Cohort (ref. 1945–1954)
1915–1924	1.22		0.90	1.65	-		-
1925–1934	1.19		0.96	1.48	0.98		0.70	1.37
1935–1944	1.04		0.84	1.27	1.18		0.91	1.52
1955–1964	0.91		0.73	1.13	0.93		0.75	1.17
1965–1974	1.21		0.85	1.71	1.19		0.84	1.70
1975–1984	1.35		0.60	3.02	1.33		0.59	2.96
Female	0.74	***	0.67	0.82	0.75	***	0.66	0.86
Race/Ethnicity (ref. NH White)
NH Black	1.43	***	1.29	1.58	1.63	***	1.40	1.90
MX-American	1.43	***	1.26	1.62	1.57	***	1.32	1.86
Other	1.00		0.80	1.24	0.98		0.72	1.33
Foreign born	0.75	**	0.64	0.90	0.84		0.66	1.07
Region (ref. Northeast)
Midwest	1.01		0.85	1.20	0.90		0.69	1.18
South	1.12		0.99	1.27	1.02		0.83	1.27
West	0.98		0.85	1.13	0.87		0.66	1.14
Smoking Status (ref. Never)
Former	1.34	***	1.19	1.50	1.29	**	1.09	1.53
Current	2.45	***	2.21	2.72	2.15	***	1.78	2.59

Notes:

N=40,058; N(dead <75)=4,014; N(dead <65)=2,011.

Significance based on two-tailed t-test, indicated as:

^*p<0.05;

^**p<0.01;

^***p<0.001.

Appendix Table B.3:

High Waist-to-Height Ratio (≥0.5) Mortality Risk with Four-category Educational Attainment, 1988–2015 NHANES-LMF (Ages 30–74)

	All-Cause				Underlying Hyp./Diab.				CVD				Cardiometabolic
	HR		95% CI		HR		95% CI		HR		95% CI		HR		95% CI
High Waist-to-Height Ratio (≥0.5)	1.62	**	1.22	2.13	1.98		0.89	4.39	3.35	***	1.80	6.23	2.41	***	1.50	3.88
High WtHR X Education (ref. BA+)
Less than HS	0.62	*	0.44	0.89	1.01		0.41	2.47	0.37	*	0.17	0.82	0.56		0.31	1.02
HS or Equivalent	0.74		0.54	1.01	0.71		0.29	1.74	0.37	**	0.18	0.75	0.50	*	0.28	0.88
Some College	0.64	*	0.43	0.96	1.18		0.44	3.19	0.29	**	0.14	0.61	0.48	*	0.26	0.90
Education (ref. BA+)
Less than HS	3.01	***	2.16	4.18	2.38	*	1.05	5.38	4.97	***	2.47	10.00	3.35	***	2.00	5.60
HS or Equivalent	2.07	***	1.57	2.74	2.83	*	1.23	6.49	4.02	***	2.11	7.68	3.02	***	1.82	4.99
Some College	2.25	***	1.54	3.29	1.31		0.50	3.47	4.48	***	2.33	8.60	2.83	***	1.64	4.87

Notes:

N=40,058.

Age at death restricted to 85.

All models control for age, cohort, gender, race/ethnicity, nativity, region, and smoking status.

Significance based on two-tailed t-test, indicated as:

^*p<0.05;

^**p<0.01;

^***p<0.001.

Appendix Table B.4:

Sensitivity Analyses of Central Obesity Mortality Risk, 1988–2015 NHANES-LMF (Ages 30–74)

	High Waist-to-Height Ratio Cutoff at 0.6
	All-Cause				Underlying Hyp./Diab.				CVD				Cardiometabolic
	HR		95% CI		HR		95% CI		HR		95% CI		HR		95% CI
High Waist-to-Height Ratio	1.81	**	1.45	2.25	5.42	***	2.95	9.94	3.04	***	1.90	4.86	2.37	***	1.63	3.45
High WtHR X Education (ref. BA+)
Less than HS	0.75	*	0.59	0.96	0.46	*	0.25	0.84	0.50	**	0.30	0.83	0.69		0.46	1.02
HS or Some College	0.83		0.65	1.06	0.45	*	0.24	0.85	0.51	*	0.29	0.90	0.70		0.45	1.10
	Controlling for Previous Weight Status (Based on Maximum BMI and Current BMI)
	All-Cause				Underlying Hyp./Diab.				CVD				Cardiometabolic
	HR		95% CI		HR		95% CI		HR		95% CI		HR		95% CI
High Waist-to-Height Ratio	1.66	***	1.26	2.17	1.70		0.63	4.56	4.06	***	1.89	8.73	2.77	***	1.66	4.63
High WtHR X Education (ref. BA+)
Less than HS	0.67	*	0.48	0.95	0.92		0.31	2.78	0.31	*	0.12	0.78	0.47	*	0.26	0.86
HS or Some College	0.68	*	0.50	0.95	0.60		0.20	1.79	0.25	**	0.11	0.58	0.40	**	0.22	0.71
	Controlling for Income-to-Needs Ratio
	All-Cause				Underlying Hyp./Diab.				CVD				Cardiometabolic
	HR		95% CI		HR		95% CI		HR		95% CI		HR		95% CI
High Waist-to-Height Ratio	1.61	**	1.22	2.13	2.45		0.89	6.76	4.36	***	1.99	9.59	2.98	***	1.78	5.00
High WtHR X Education (ref. BA+)
Less than HS	0.64	*	0.45	0.91	0.95		0.32	2.87	0.30	*	0.12	0.77	0.48	*	0.26	0.88
HS or Some College	0.69	*	0.50	0.95	0.63		0.21	1.88	0.26	**	0.11	0.60	0.41	**	0.23	0.74
	Central Obesity Defined by Maximum BMI
	All-Cause				Underlying Hyp./Diab.				CVD				Cardiometabolic
	HR		95% CI		HR		95% CI		HR		95% CI		HR		95% CI
Maximum BMI Group (ref. Normal [18.5–24.9])
Underweight (<18.5)	-		-	-	-		-	-	-		-	-	-		-	-
Overweight (25.0–29.9)	0.72		0.50	1.04	1.58		0.56	4.46	0.66		0.33	1.32	0.61		0.30	1.22
Obese, Class I (30.0–34.9)	1.10		0.72	1.70	3.72	*	1.25	11.11	2.25	*	1.08	4.71	1.48		0.79	2.78
Obese, Class II/III (35.0+)	1.63	*	1.06	2.49	7.30	***	2.50	21.36	3.07	**	1.64	5.73	2.06	*	1.14	3.72
Max. BMI (ref. Normal) X Education (ref. BA+)
Less than HS X Underweight	-		-	-	-		-	-	-		-	-	-		-	-
Less than HS X Overweight	1.08		0.72	1.61	0.72		0.25	2.09	1.28		0.57	2.84	1.53		0.67	3.50
Less than HS X Obese, Class I	0.80		0.51	1.27	0.48		0.15	1.52	0.43		0.18	1.01	0.74		0.37	1.50
Less than HS X Obese, Class II/III	0.75		0.48	1.18	0.41		0.13	1.31	0.48	*	0.24	0.96	0.86		0.46	1.61
HS or SC X Underweight	-		-	-	-		-	-	-		-	-	-		-	-
HS or SC X Overweight	1.36		0.91	2.03	0.49		0.15	1.67	1.52		0.68	3.38	1.62		0.74	3.57
HS or SC X Obese, Class I	1.00		0.64	1.57	0.38		0.12	1.20	0.59		0.25	1.36	0.88		0.44	1.78
HS or SC X Obese, Class II/III	1.10		0.71	1.73	0.44		0.14	1.32	0.72		0.31	1.70	1.15		0.55	2.40

Notes:

N=40,058.

Age at death restricted to 85.

All models control for age, cohort, gender, race/ethnicity, nativity, region, and smoking status.

Significance based on two-tailed t-test, indicated as:

^*p<0.05;

^**p<0.01;

^***p<0.001.

Appendix Table B.5:

High Waist-to-Height Ratio (≥0.5) Mortality Risk due to Accidental Causes by Educational Attainment, 1988–2015 NHANES-LMF (Ages 30–74)

	All-Cause
	HR	95% CI
High Waist-to-Height Ratio (≥0.5)	1.17		0.36	3.79
High WtHR X Education (ref. BA+)
Less than HS	0.81		0.24	2.72
HS or Some College	0.97		0.26	3.64
Education (ref. BA+)
Less than HS	2.69		0.86	8.41
HS or Some College	1.56		0.47	5.15
Cohort (ref. 1945–1954)
1915–1924	0.85		0.38	1.88
1925–1934	0.70		0.32	1.52
1935–1944	0.65		0.33	1.29
1955–1964	1.46		0.82	2.59
1965–1974	2.78	*	1.12	6.88
1975–1984	3.61		0.95	13.65
Female	0.54	***	0.40	0.72
Race/Ethnicity (ref. NH White)
NH Black	0.91		0.58	1.42
MX-American	1.45		0.93	2.27
Other	0.67		0.35	1.28
Foreign born	0.95		0.64	1.41
Region (ref. Northeast)
Midwest	0.79		0.42	1.47
South	0.93		0.54	1.59
West	0.88		0.48	1.62
Smoking Status (ref. Never)
Former	0.99		0.64	1.52
Current	1.70	*	1.12	2.58

Notes:

N=40,058; N(deaths)=330.

Age at death restricted to 85.

Significance based on two-tailed t-test, indicated as:

^*p<0.05;

^**p<0.01;

^***p<0.001.