key: cord-0320688-5yth8lk8
authors: Tilstra, A. M.; Aburto, J. M.; Gutin, I.; Dowd, J. B.
title: Coding of Obesity-related Mortality Impacts Estimates of Obesity on U.S. Life Expectancy
date: 2022-05-16
journal: nan
DOI: 10.1101/2022.05.16.22275140
sha: f652954a2bd64dbd5c03bb421a5422bf2b562d97
doc_id: 320688
cord_uid: 5yth8lk8

Background High levels of obesity remain an important population health problem in the U.S. and a possible contributor to stalling life expectancy. However, reliable estimates of the contribution of obesity to mortality in the U.S. are lacking, because of inconsistent coding of obesity-related causes of death. Methods We compare five International Classification of Diseases version 10 (ICD-10) coding schemes for obesity-related mortality used in the literature and examine how the magnitude of obesity-related mortality burdens varies across different schemes. We use U.S. multiple cause of death data and population estimates for the Black, white, and Latino population in the years 2010, 2015, and 2020. In sex- and race/ethnic-stratified analyses, we estimate the potential years of life expectancy gained if obesity-related mortality had not occurred as measured by each coding scheme. Results We estimate that obesity-related mortality contributes to up to 78 months (6.5 years) of lost U.S. life expectancy, though estimates range from as low as 0 months, with a median contribution across ICD-10 coding schemes of about 20 months (1.7 years). Despite substantial variation across coding schemes, obesity-related mortality consistently contributes more to life expectancy deficits for Black Americans compared to white and Latino Americans. Across all ICD-10 coding schemes, the age pattern of obesity follows a J-shaped curve, suggesting exponential increases in obesity-related mortality after age 25. Conclusions The estimation of the burden of obesity-related mortality on life expectancy in the United States varies widely depending on the causes of death used in analyses. This inconsistency may obscure our understanding of the contribution of obesity-related mortality to trends in life expectancy. We propose a standardization of the coding of obesity-related mortality for future studies and outline which causes should be included.

coding schemes, the age pattern of obesity follows a J-shaped curve, suggesting exponential 24 increases in obesity-related mortality after age 25. 25 26 

The estimation of the burden of obesity-related mortality on life expectancy in the United States 28 varies widely depending on the causes of death used in analyses. This inconsistency may obscure 29 our understanding of the contribution of obesity-related mortality to trends in life expectancy. 30 We propose a standardization of the coding of obesity-related mortality for future studies and 31 outline which causes should be included. 32 Introduction 36 37 In 2018, an estimated 42.5% of the United States population had obesity (body mass index 38 (BMI) ≥ 30) , an increase from 30.5% in 2000. Over the same period, rates of severe obesity 39 (BMI≥ 40) nearly doubled to 9.2% (1) . There are pronounced race/ethnic differences in 40

American obesity rates, as evidenced by the 57% obesity rate among non-Hispanic Black women 41 compared to 40% among non-Hispanic white women (1) . While the obesity epidemic defies easy 42 explanation, one hypothesis is that an increasingly obesogenic environment over past decades 43 (e.g., increased accessibility of calorically-dense food and more sedentary lifestyles) has made 44 maintaining a healthy body weight more challenging, in ways that are socially patterned (2) . 45 Consequently, the obesity epidemic may be contributing to alarming U.S. mortality and life 46 expectancy trends (3) (4) (5) , as well as racial disparities in these trends (6, 7) . Research also indicates 47 that these trends may continue to worsen as younger cohorts are exposed to more time in the 48 U.S. obesogenic environment (3, 8) . 49 50 Despite the increased prevalence of obesity in the United States and widespread recognition of 51 its consequences for population health, we lack consensus on how to track and report its impact 52

on key population health metrics such as mortality rates and life expectancy estimates. As argued 53 in this paper, a key obstacle in obtaining such estimates is the absence of a standardized 54 approach for measuring obesity-related mortality based on reporting standards in existing vital 55 statistics data. 56 57 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2022. ;  https://doi.org/10.1101/2022.05. 16.22275140 doi: medRxiv preprint Currently, statistical approaches for estimating obesity-related mortality generally fall into one of 58 two categories based on the data source used for analysis. One approach uses obesity prevalence 59 and all-cause mortality to estimate obesity-attributable mortality fractions (OAMF) (9) . To 60 estimate this, researchers use nationally-representative survey data (e.g., National Health 61

Interview Survey, National Health and Nutrition Examination Survey) to calculate the 62 prevalence of obesity and the relative risk of premature mortality associated with obesity as 63 compared to a lower weight status. The two are then combined to provide a counterfactual 64 proportion of deaths attributable to obesity that could have otherwise been avoided if all adults 65 were assigned to a lower weight status (10) . There are different techniques to estimate OAMF, 66 including the partially adjusted method and the weighted sum method (11, 12) . Yet, the OAMF 67 approach is often limited in estimating an association between obesity and/or BMI at a given 68 point in time and mortality many years later, as there are many possible sources of confounding 69 that introduce substantial uncertainty in the meaning of a single measure of obesity (13) (14) (15) . 70 71 A second, more recent approach -which is the subject of this analysis -uses underlying and 72 multiple cause of death mortality data to estimate obesity-related mortality rates (8, (16) (17) (18) . 73

Because of the very strong and likely causal association between obesity and a broad variety of 74 cardiometabolic conditions, many different causes of death are often categorized as obesity-75 related (9, (19) (20) (21) (22) (23) . Though categorizations from multiple cause of death data cannot definitively 76 capture obesity-related mortality, they may be a better proxy for such deaths and provide more 77 "direct individual-level evidence" given that they come from death certificates rather than 78

OAMFs from survey data (16) . These data are also valuable in highlighting the many co-morbid 79 conditions that contribute to adult mortality, as the information provided by a single, underlying 80 cause may be insufficient (24) (25) (26) . While the demographic analysis techniques used with 81 population-level cause-specific mortality data are less varied across studies (e.g., estimation of 82 age-standardized mortality rates), there is a clear lack of consistency in the classification of 83 obesity-or metabolic-related causes of death. The relatively standardized statistical approaches 84 with population-level mortality data warrant a standardized coding scheme for consistent and 85 comparable estimates of obesity-related mortality across studies, time, and subpopulations of 86 interest. 87

The International Classification of Disease (ICD) lists obesity as an independent cause of death 89 (ICD10: E65-E67), but researchers acknowledge that this coding is used too infrequently and 90 underestimates the true mortality burden of obesity, as there are many other causes of death 91 which are likely representative of obesity-related mortality, including hypertension and diabetes 92 (27) (28) (29) . Thus, studies using multiple cause of death data often include additional ICD codes in 93 their obesity coding schemes. However, we contend that the inconsistent use of ICD codes across 94 studies is a key limitation of this work as it leads to vastly different estimates of the contribution 95 of obesity to U.S. mortality. 96

In this paper, we test how different obesity-related ICD coding schemes equate to variation in the 98 estimated of years of life expectancy in the U.S. lost due to obesity. We also show how these 99 patterns differ by sex and race/ethnicity, given large disparities in obesity rates within the U.S. 100

population. Finally, we identify critical flaws in commonly used obesity-related ICD coding 101 schemes and propose best practices for standardizing. In turn, our systematic comparison of 102 these different coding practices -both over time and by sex and race/ethnicity -helps 103 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275140 doi: medRxiv preprint demonstrate the considerable variation in obesity's impact over past decades and across groups. 104

Ultimately, we argue that greater standardization is vital for improving our knowledge of 105 population health trends in obesity-related mortality and its impact on life expectancy. Appendix A includes the ICD-10 codes for each coding scheme. Several differences across 126 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275140 doi: medRxiv preprint coding schemes are worth noting. First, GBD and Acosta are the only two to include neoplasms. 127 Acosta restricts to "obesity-related" neoplasms, but GBD codes are much more expansive (e.g., 128 leukemia). Second, Acosta and UCOD do not include any hypertension or heart disease causes, 129 while the other three coding schemes do. Lifetables were calculated by sex and racial/ethnic groups following standard demographic 145 techniques (31) . From these, life expectancy at birth was used as the main indicator of our 146 analysis. Life expectancy is the average number of years a cohort of newborns is expected to live 147 if they were to experience the mortality rates present in a given year throughout their lives. It is 148 not a forecast or projection of any individual's lifespan, but it accurately reflects mortality levels 149 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. While under the GBD coding scheme it comprises a large proportion of deaths, from 28.5% for 168

Latino males to 37.0% among Black females. Across all schemes, there is a higher proportion of 169 obesity-related deaths for Black males and females than for white and Latino males and females, 170 consistent with racial and ethnic disparities in obesity prevalence. 171 172 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The age pattern of obesity-related mortality, shown in Figure 3 , is relatively consistent across 182 coding schemes. It is also consistent over time (as shown in Appendix C). The J-shaped curve 183 suggests that obesity-related mortality increases exponentially with age after around age 25, 184 showing that even at young ages (0-20), people are dying from causes deemed to be obesity-185 related, albeit at much smaller rates. The exception is, again, the UCOD scheme which shows 186 little to no obesity-related deaths at young ages. This scheme also shows a stable distribution 187 across ages, suggesting that the likelihood of a death being coded with obesity as the underlying 188 cause does not fluctuate much across ages. At older ages, the Acosta scheme deviates from other 189 trajectories, which may be indicative that it more accurately captures frailty or selection. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275140 doi: medRxiv preprint research (10, 12, 16, 18) . Thus, the measurement of obesity-related mortality is both substantively 238 and theoretically important, but approaches have not been standardized up to now. Our results 239 suggest several key takeaways for moving forward. This study has important implications for the measurement of obesity-related mortality. First, we 252 recommend that UCOD, or obesity as the single underlying cause of death, should not be used 253 for estimating the contribution of obesity to mortality in a population. It is an overly narrow 254 definition of obesity-related deaths, which necessitates that obesity is listed as not only a cause of 255 death on the death certificate, but as the underlying cause. Unfortunately, this strict criterion is 256 potentially subject to the biases/assumptions of the person coding the cause of death (27, 29) , 257 leading to the identification of less than a fifth of all deaths where obesity is implicated (24) . For 258 instance, a cross-national comparison of obesity-related mortality suggests that physicians' 259 propensity for reporting obesity as an underlying cause on death certificates might vary due to 260 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275140 doi: medRxiv preprint their "sensitiv[ity] to abnormally high BMIs," in reference to national differences in the 261 prevalence of obesity and thus expectations about the average BMI and body size (24) . Within 262 the context of the United States, reporting practices may be similarly influenced by sex and 263 racial/ethnic differences surrounding body size norms and health (38) (39) (40) , as well as sex and 264 racial/ethnic biases in how medical practitioners view their patients and health (41) (42) (43) . These 265 factors may in turn influence perceptions about obesity's role for individuals' health (44, 45) , and 266 thus its contribution to their death. Thus, it is evident that existing coding practices for obesity as 267 other schemes, are provided in Appendix A. It is inevitable that there is some margin of error in 281 how well each of these causes of deaths can be directly attributable to obesity, but past studies 282 clearly demonstrate where and when this association is stronger or weaker (9, 17, 21, 46) . Future 283 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275140 doi: medRxiv preprint research might consider developing a relative risk threshold for the association between obesity 284 and specific causes of death, to use direct epidemiologic evidence to determine which causes to 285 include. For now, the codes used by Adair and Lopez strike the best balance of including 286 conditions most strongly associated with obesity (16) . 287 288 Second, the age pattern of mortality across coding schemes emphasizes the importance of 289 considering when in the life course obesity can plausibly be considered a contributing cause of 290 death. Obesity has relatively immediate health consequences, mostly reflected in its association 291 with elevated biomarkers for poor cardiovascular and metabolic health (47) (48) (49) , as well as its 292 broad pro-inflammatory effect (50, 51) , which is implicated in numerous chronic conditions. 293

However, the health consequences that might increase the risk of mortality are typically 294 cumulative in nature -especially with respect to chronic cardiometabolic disease -and thus 295 unlikely to be more directly implicated until later in life, when adults have lived with one or 296 more chronic diseases for an extended period of time (52) (53) (54) . Moreover, some proportion of 297 obesity-related deaths are likely misclassified when using multiple cause of death data. This 298 misclassification bias may be more salient at younger ages where even causes of death that are 299 very strongly linked to obesity (e.g., diabetes, heart disease) have a more complex etiology. 300

Thus, it is unlikely that young individuals are dying directly of obesity-related mortality. To 301 account for this, we suggest that researchers consider restricting to ages 25 and older when 302 estimating obesity-related mortality research. This is particularly important for research 303 estimating the consequences of obesity-related mortality on life expectancy which, unless 304 specified otherwise, considers all ages in the estimation. 305 306 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2022. ;  https://doi.org/10.1101/2022.05.16.22275140 doi: medRxiv preprint

Limitations 308

As noted earlier, the only currently-available ICD coding options that specifically implicates 309 obesity as the underlying cause of death are the E66-67 codes, which likely heavily undercount 310 obesity-related deaths -as demonstrated in this analysis and suggested in earlier work (16,27-311 29) . Thus, it is inevitable that at least some proportion of the deaths in each scheme are not due 312 to obesity, but instead a function other lifestyle and health factors, such as smoking (17) . It is 313 difficult to say whether the estimates presented by different coding schemes are liberal or 314 conservative, as any "overestimation" noted above may be balanced out by "underestimation" in 315 obesity's association with numerous other causes of death not included in the schemes (e.g., 316

infectious disease [(24)]). 317 318 It is worth reiterating that the consequences of obesity for mortality can be estimated with both 319 population-level data and demographic techniques, as is shown here, or with obesity-attributable 320 mortality functions (OAMF). OAMFs, while useful because they estimate the direct relationship 321 between obesity and mortality, are flawed because they estimate population-level obesity-related 322 mortality using only a sample of the population, whose obesity was estimated at a time point 323 preceding their mortality. Additionally, there is some evidence that the BMI cutoffs for obesity 324 are perhaps not the same across all sex and race/ethnic groups, with a weaker association 325 between high BMI and mortality for non-Hispanic Black and Hispanic men (55) (56) (57) (58) . In turn, 326 there may be nonnegligible variation in how accurately multiple cause of death coding schemes 327 capture obesity-related mortality (i.e., overestimation or underestimation) across different 328

subpopulations. However, studies are largely equivocal in reaching consensus on the 329 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2022 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. We thank the Health Inequality Working Group at the Leverhulme Centre for Demographic 385

Science for their insightful comments on the manuscript. 386 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2022. ; https://doi.org/10.1101/2022.05.16.22275140 doi: medRxiv preprint

Prevalence of Obesity and Severe Obesity Among Adults: United States

Reducing Obesity: Motivating Action While Not Blaming the Victim. Milbank Q

A Potential Decline in Life Expectancy in the United States in the 21st Century

Projecting the Effect of Changes in Smoking and Obesity on Future Life Expectancy in the United States

The role of obesity in exceptionally slow US mortality improvement

The Contribution of Weight Status to Black-White Differences in Mortality

Would the elimination of obesity and smoking reduce U.S. racial/ethnic/nativity disparities in total and healthy life expectancy? SSM -Popul Health

Explaining recent mortality trends among younger and middle-aged White Americans

Cause-Specific Excess Deaths Associated With Underweight, Overweight, and Obesity

Contribution of Obesity to International Differences in Life Expectancy

Impact of Different Estimation Methods on Obesity-Attributable Mortality Levels and Trends: The Case of The Netherlands

Impact of obesity on life expectancy among different European countries: secondary analysis of population-level data over the 1975-2012 period

How Dangerous Is Obesity? Issues in Measurement and Interpretation: How Dangerous Is Obesity? Issues in Measurement and Interpretation

Revealing the burden of obesity using weight histories

Association of Weight at Different Ages and All-Cause Mortality Among Older Adults in the US

The role of overweight and obesity in adverse cardiovascular disease mortality trends: an analysis of multiple cause of death data from Australia and the USA

Body-mass index and all-cause mortality: individual-participant-data meta-analysis of 239 prospective studies in four continents. The Lancet

Cardiovascular mortality gap between the US and other high life expectancy countries in 2000-2016

Executive summary: Guidelines (2013) for the management of overweight and obesity in adults: A Report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines and The Obesity Society Published by The O

Obesity and overweight in relation to organ-specific cancer mortality in London (UK): findings from the original Whitehall study

Body-mass index and risk of 22 specific cancers: a population-based cohort study of 5·24 million UK adults. The Lancet

Causal relationships between obesity and the leading causes of death in women and men

Clinical Guidelines on the Identification, Evaluation, and Treatment of Overweight and Obesity in Adults--The Evidence Report. National Institutes of Health

Obesity-related mortality in France, Italy, and the United States: a comparison using multiple cause-ofdeath analysis

Analysing Multiple Causes of Death: Which Methods For Which Data? An Application to the Cancer-Related Mortality in France and Italy

Cause-specific mortality analysis: Is the underlying cause of death sufficient?

Certification of obesity as a cause of death in England 1979-2006

Postmortem examination in the morbidly obese

Correcting the Count: Improving Vital Statistics Data Regarding Deaths Related to Obesity

National Cancer Institute, DCCPS, Surveillance Research Program

Demography: measuring and modeling population processes

International Handbook of Adult Mortality

International Handbooks of Population

Provisional Life Expectancy Estimates for 2020

Social context explains race disparities in obesity among women

Racial and Ethnic Disparities in Early Childhood Obesity

Mind the Gap: Race/Ethnic and Socioeconomic Disparities in Obesity

Race/ethnicity, nativity and trends in BMI among U.S. adults: Race/Ethnicity, Nativity, and Trends in BMI

Ethnic and racial differences in body size perception and satisfaction

Beliefs About Personal Weight Among African American Women

Body Image and Modifiable Weight Control Behaviors Among Black Females: A Review of the Literature

Evidence of Nonconscious Stereotyping of Hispanic Patients by Nursing and Medical Students

The Associations of Clinicians' Implicit Attitudes About Race With Medical Visit Communication and Patient Ratings of Interpersonal Care

Implicit Stereotyping and Medical Decisions: Unconscious Stereotype Activation in Practitioners' Thoughts About African Americans

Impact of Patient-Doctor Race Concordance on Rates of Weight-Related Counseling in Visits by Black and White Obese Individuals

Racial and ethnic minority patients report different weight-related care experiences than non-Hispanic Whites

Body-mass index and cause-specific mortality in 900 000 adults: collaborative analyses of 57 prospective studies. The Lancet

Cardiovascular risk and obesity

A clinical perspective of obesity, metabolic syndrome and cardiovascular disease

Obesity and Cardiovascular Disease: A Scientific Statement From the American Heart Association

Obesity-induced inflammation: a metabolic dialogue in the language of inflammation

Obesity and Inflammation: Epidemiology, Risk Factors, and Markers of Inflammation

The Consequences of Childhood Overweight and Obesity

Cumulative Disadvantage and Health: Long-Term Consequences of Obesity?

CALCULATING DEATHS ATTRIBUTABLE TO OBESITY

Obesity and All-Cause Mortality Among Black Adults and White Adults

Body mass index and mortality rate among Hispanic adults: a pooled analysis of multiple epidemiologic data sets

Body-mass index and mortality risk in US Blacks compared to Whites: Body Mass Index and Mortality among Blacks and Whites

Body Mass Index and All-Cause Mortality in a Large Prospective Cohort of White and

Smoking-attributable mortality by cause of death in the United States: An indirect approach. SSM -Popul Health