Trends in Cause-Specific Mortality by Race and Hispanic Origin, 1999–2019

Introduction

Assessing outcomes for Social Security retirement and disability program participants involves accounting for beneficiary mortality patterns, and studies of mortality have examined death rate variations by socioeconomic status (SES), earnings, and educational attainment far more than by race and ethnicity (RE). For instance, it is well established that mortality declines as a function of higher SES and that gains in longevity over time disproportionately accrue to those with higher earnings and educational attainment (Bor, Cohen, and Galea 2017). As a result, with all else equal, an “across the board” Social Security policy change such as an increase in the normal retirement age represents a disproportionately large benefit reduction for retired beneficiaries with lower lifetime earnings, whose life expectancies tend to be shorter than those of workers with higher earnings histories. In a similar fashion, understanding mortality differentials over time by race, ethnicity, and sex can shed light on the distributional effects of current and future retirement and disability program policies. This article summarizes and compares recent trends (1999–2019) in cause-specific mortality in the United States by RE group, sex, and age.

For this study, population and death counts by cause of death were obtained from the Centers for Disease Control and Prevention Wide-Ranging Online Data for Epidemiological Research (CDC WONDER) database, which provides underlying cause-of-death data using codes defined in the World Health Organization's International Statistical Classification of Diseases and Related Health Problems, 10^th revision (ICD-10).¹ I selected the 1999–2019 time frame deliberately to avoid mortality data beyond 2019. Doing so sidesteps the COVID-19 pandemic's distortion of historic trends.² Another reason to avoid the most recent year available in the data is that Census Bureau population estimates can change from year to year (particularly for older individuals), so that the estimated mortality rates for 2020 and later may be revised in future updates. Likewise, extending the data set to years before 1999 would require reconciling different ICD classification schemes, which lies beyond my expertise and the scope of this article.³

A crude death rate is simply the ratio of the number of deaths to the population count for a given group at a specific point in time. Comparing crude death rates across groups or time periods can be deceiving because the age distributions of population groups can change in divergent ways over time. Age-adjusting the death rates by holding the age distribution constant overcomes the effects of those changes and allows a meaningful comparison of mortality rates between males and females, between people of different races or ethnicities, or across different time periods for members of a single group. All mortality rates discussed in this article are age-adjusted and are expressed as deaths per 100,000 population. For consistency with the CDC WONDER database's default option, I use the direct method of computing age-adjusted death rates, based on the CDC definition of the 2000 U.S. standard population.⁴

There are inherent limitations to the quality of cause-specific mortality data, particularly when disaggregated by race and Hispanic origin. The population data in the denominator of a crude mortality rate and the death count data in the numerator derive from different sources. Specifically: All population counts used in this article are bridged-race estimates of the July 1^st resident population from the Census Bureau, in which the RE data are self-reported.⁵ By contrast, the death counts come from death certificates filed in the states and compiled into a national database by the CDC's National Center for Health Statistics (NCHS). Any information on race or ethnicity in a death certificate is either based on direct observation by the funeral director or reported by the surviving next of kin or other informant. As a result, there can be discrepancies between the decedent's self-identified RE group and the judgment of an outside observer.

Comparing the NCHS database with the Census Bureau's National Longitudinal Mortality Study, Arias and others (2008) found that the quality of death certificate data is excellent for the White and Black populations, reasonably good for the Hispanic and Asian or Pacific Islander (API) populations, and very poor for the American Indian or Alaska Native (AIAN) population. For example, for the 1990–1998 period, the percentage of respondents in a self-identified RE group who were correctly identified on the death certificate was only 55.2 percent among the AIAN community, but it was 88.1 percent for the Hispanic population, 89.7 percent for the API population, 98.1 percent for Black individuals, and 99.6 percent among the White population. Adjustment for death certificate misclassification had minimal effect on the mortality estimates associated with the API and Hispanic communities, but led to a reversal of the mortality differentials between the White and AIAN populations. Moreover, the authors found that the geographic distribution of the populations affected the misclassification rate. For AIAN, API, and Hispanic individuals, residence in areas with a high coethnic concentration improved the quality of the RE classification on the death certificate. Nativity was also an important factor, with the foreign-born Hispanic population considerably more likely to be correctly classified than the U.S.-born group. The quality of the death certificate data was better for people of Mexican, Puerto Rican, and Cuban origin than for those of Central American, South American, or other Hispanic heritage. Thus, although death rate estimates for the entire Hispanic population are reliable, accurately estimating mortality rates for specific Hispanic subgroups could be problematic.

Also complicating the RE classifications in the underlying data is the fact that as the U.S. population has become more diverse, Americans' perceptions on race have evolved. In 2000 and 2010, census respondents identifying as “some other race”—an option that has been part of the questionnaire for more than 100 years in one form or another and was always meant to be a minor residual category—became instead the third largest race group (Ashok 2016). Those identifying as some other race were mostly Hispanic individuals, as well as members of Afro-Caribbean, Middle Eastern, and North African populations who did not identify with any of the Office of Management and Budget's official race categories. The increasing share of ethnoracially mixed families over time suggests further challenges to the collection of RE data that can accurately illustrate trends, as increasing numbers of Americans with multiple backgrounds will decide how to define their own identity. The public's evolving perception of RE identification calls for more detailed and disaggregated data. The Census Bureau has undertaken a number of research projects aimed at improving RE question design and data quality, such as the 2010 Census Alternative Questionnaire Experiment and the 2015 National Content Test (Census Bureau 2022), but the challenge is ongoing.

A notable limitation of the cause-specific mortality data is the fact that one or more comorbidities may accompany a reported cause of death, raising potential uncertainty about the primary underlying cause. Typically, a funeral home director records the demographic information in a death certificate; but physicians, medical examiners, or coroners report cause of death using the current ICD standards. For its statistical tabulations, NCHS applies an automated coding system that can overrule the actual cause of death reported by a certifier, in an effort to standardize and improve the quality of the data.⁶ Nevertheless, death certificate information can be incomplete or certifiers may be poorly trained, leading to ambiguities. For example, based on a sample from the University of Michigan's Health and Retirement Study of noninstitutionalized U.S. adults, Stokes and others (2020) found that the death certificates underestimated the burden of dementia on mortality by a factor of 2.7.

For this article, I analyze mortality among four RE categories: (1) White, non-Hispanic (WNH); (2) Hispanic (of any race); (3) Black, regardless of Hispanic origin; and (4) API, regardless of Hispanic origin.⁷ The analysis explicitly excludes a separate AIAN category because it represents a relatively small sample in the CDC WONDER database (4.8 million people in 2019) and Arias and others (2008) documented a high misclassification rate in their death certificates. However, AIAN individuals of Hispanic origin are included in the Hispanic population group. Notice that the Black, Hispanic, and API categories are not mutually exclusive, as this would require arbitrarily suppressing one of two overlapping identity groups. Thus, for instance, an individual who identifies as both Black and Hispanic is included in both categories.⁸ Nevertheless, the amount of overlap is small. For example, the full population count in 2019 was 328.3 million people, while the sum of the four RE categories was 329.9 million. The difference (1.6 million people) represents individuals overlapping the Hispanic, Black, and API population groups, minus AIAN individuals of non-Hispanic origin. It amounts to one-half of 1 percent of the full population count.

Finally, this analysis is descriptive, in that there is no attempt to fit the computed age-adjusted mortality series to any formal statistical model. By contrast, in Woolf and Schoomaker (2019), for example, the authors fit a series of joint-point regression models to ascertain statistically whether a mortality series rose or declined over a particular period (which itself involves a number of assumptions and reliance on a modification of the Bayesian information criterion). This article reports rising or declining mortality simply as the difference in magnitudes between the beginning and end of the period observed.

The article comprises this introduction and 15 additional sections, arranged as follows:

• A section that describes features of the four RE groups whose counts enter the denominator of crude death rates, highlighting distributional differences in their age structures and the effects of immigration. It also addresses variation in the geographic distribution of the various population groups.
• A section that provides an overview of cause-specific age-adjusted mortality, based on cause-of-death categories corresponding to 17 ICD-10 chapters.
• A section that describes variation in all-cause mortality by state, county, and RE group.
• Individual sections discussing mortality for 10 cause-of-death categories. Each section explores mortality patterns for the specific subcategories that constitute those categories:
  • Diseases of the circulatory system.
  • Neoplasms (cancers).
  • Diseases of the respiratory system.
  • External causes of mortality, such as transportation accidents, assaults, falls, poisonings (including drug and alcohol overdoses), and self-inflicted harm (suicide).
  • Diseases of the nervous system, including Alzheimer's, Parkinson's, and other degenerative diseases.
  • Endocrine, nutritional, and metabolic diseases, such as diabetes and obesity.
  • Mental and behavioral disorders, including dementia and psychoactive substance use.
  • Diseases of the digestive system.
  • Diseases of the genitourinary system, such as renal failure.
  • Infectious and parasitic diseases, including human immunodeficiency virus (HIV), hepatitis, and sepsis.
• A brief section presenting summary statistics for other less frequent cause-of-death categories.
• A concluding section, with a summary of findings.