Education, Race/Ethnicity, and Causes of Premature Mortality Among Middle-Aged Adults in 4 US Urban Communities: Results From CARDIA, 1985–2017

Abstract

Objectives. To assess causes of premature death and whether race/ethnicity or education is more strongly and independently associated with premature mortality in a diverse sample of middle-aged adults in the United States.

Methods. The Coronary Artery Risk Development in Young Adults study (CARDIA) is a longitudinal cohort study of 5114 participants recruited in 1985 to 1986 and followed for up to 29 years, with rigorous ascertainment of all deaths; recruitment was balanced regarding sex, Black and White race/ethnicity, education level (high school or less vs. greater than high school), and age group (18–24 and 25–30 years). This analysis included all 349 deaths that had been fully reviewed through month 348. Our primary outcome was years of potential life lost (YPLL).

Results. The age-adjusted mortality rate per 1000 persons was 45.17 among Black men, 25.20 among White men, 17.63 among Black women, and 10.10 among White women. Homicide and AIDS were associated with the most YPLL, but cancer and cardiovascular disease were the most common causes of death. In multivariable models, each level of education achieved was associated with 1.37 fewer YPLL (P = .007); race/ethnicity was not independently associated with YPLL.

Conclusions. Lower education level was an independent predictor of greater YPLL.

Depending on race/ethnicity, education, and place of residence, Americans experience up to 30-year differences in life expectancy at birth.^1–3 These disparities in life expectancy have become a focal point of population-based research and subsequent national initiatives and local interventions.^4,5 Despite these efforts, geographic disparities in life expectancy persist and are worsening, especially among middle-aged adults.^6–8 Premature mortality is still most prevalent among Black men,⁹ but it has been rising among middle-aged (aged 45–54 years) White men with low education in recent years.⁷ Overall, this has resulted in an alarming decline in national life expectancy in 2015, 2016, and 2017 for the first time in decades.^10,11

Most previous studies have reported contributors to premature mortality at all ages, but the greatest disparities in premature mortality exist among young and middle-aged adults.^12,13 These studies have reported several factors contributing to disparities in life expectancy, including race/ethnicity, education level, income, access to health care, and lifestyle factors.^1,2,6,14 However, these factors are highly interrelated and are influenced by, among others, neighborhood segregation, especially in urban areas. As Chetty et al. described, Black and Hispanic children growing up in neighborhoods within the same city have starkly different socioeconomic and health trajectories based on the poverty rates of their neighborhoods.¹⁵ Thus, untangling the contributions of race/ethnicity and socioeconomic status to premature mortality in urban environments is essential. Though a few population-based studies have isolated educational attainment as the strongest predictor of life expectancy after controlling for other socioeconomic variables,^1,2 this relationship was still strengthened by the addition of race, suggesting that race has some additional effect.² While these previous studies have advanced our understanding of the root causes of disparities in life expectancy, they are limited by their reliance on ecologic analyses using census or other cross-sectional geographically based population estimates, making it difficult to know whether race/ethnicity or education is the underlying driver of these observations.

The Coronary Artery Risk Development in Young Adults study (CARDIA) is a longitudinal cohort of equally represented Black and White men and women of higher and lower educational attainment who were recruited from 4 urban centers at ages 18 to 30 years.¹⁶ Data from this cohort provide a rare opportunity to identify predictors of and causes of premature mortality among Black and White men and women before the age of 60 years. We previously reported large differences in causes of premature mortality by race/ethnicity and sex through 16 years of follow-up. This study goes beyond the year-16 report,¹⁷ including up to 29 years of participant follow-up with excellent retention of study participants. Much has changed in the interim years, with advances in medical treatment changing the trajectory of diseases such as HIV/AIDS, and although our cohort has aged, their deaths are still premature. As such, we used a years of potential life lost (YPLL) approach in this analysis. We investigated whether race/ethnicity or educational attainment was more strongly and independently associated with YPLL among Black and White middle-aged men and women. We also assessed which causes of death contributed to greater YPLL. We chose to use YPLL because it is a measure that is easy to comprehend and effectively emphasizes deaths among younger individuals, as in the CARDIA cohort.¹⁸

METHODS

CARDIA is an ongoing longitudinal study of Black or White men and women designed to identify risk factors for cardiovascular disease and its progression. Details of the study design have been published elsewhere.¹⁶ Briefly, the study recruited 5115 participants between the ages of 18 and 30 years at baseline in 1985 to 1986. One participant withdrew consent after recruitment, so we excluded that individual’s data and used data from the remaining 5114 participants in these analyses. Participants were recruited from 4 field centers across the United States: Birmingham, Alabama; Chicago, Illinois; Minneapolis, Minnesota; and Oakland, California. The sampling protocol was designed to oversample Black participants to obtain at each field center balanced numbers of Black and White men and women, those with a high-school education or less versus more than a high school education, and age group (18 to 24 and 25 to 30 years), resulting in a recruited cohort that is 48% White and 52% Black, with 54% women and 46% men.

Following the baseline examination in 1985 to 1986, field center staff attempted to contact all participants for follow-up clinic visits at years 2, 5, 7, 10, 15, 20, and 25. Participants were followed between examinations via 2 types of interim contacts, with follow-up windows around the participant’s baseline examination date. In annual contacts, field center staff obtained data directly from the participant or a designated proxy if the participant could not physically provide data. Various health events and characteristics (e.g., smoking status, diagnosis of diabetes) as well as hospitalizations and specific outpatient treatments (e.g., outpatient revascularization) were collected. Semiannual contacts confirmed contact information and vital status. At this contact, the verification could be obtained from an informant other than the participant (e.g., spouse, parent, roommate, friend), although direct contact with the participant was preferred. If neither a participant nor any of that participant’s designated informants could be contacted, we performed a vital status check yearly using various types of resources including various locator search engines, obituary search engines, and the Social Security Death Index. Additional contacts consisted of mailing yearly birthday and winter holiday cards and periodic newsletters with address correction service requested.

Finally, National Death Index searches for lost participants were conducted at approximately 5-year intervals. This analysis includes data through the 348-month annual contact, with ascertainment of deaths completed on August 28, 2015, and the most recent National Death Index data available from deaths in 2011. At that time, 92% of the surviving cohort was successfully contacted within the previous 5 years, though some who were unable to be contacted were deceased, but their death records were yet to be adjudicated.

Once field center staff ascertained a death, specific records were requested based on the location (e.g., during an admission to an acute care hospital) and the potential cause of death (e.g., potential stroke or coronary heart disease event) according to a detailed protocol. Death certificates were requested for all deaths. Other potential records included hospital records, autopsy and coroner’s reports, or emergency department records, and detailed informant interviews for outpatient (including emergency department) suspected cardiovascular deaths. Two study physicians reviewed all information and assigned cause of death independently by using specific definitions and a detailed manual. The CARDIA Endpoints Surveillance and Adjudication Subcommittee resolved disagreements.

Independent variables included age at baseline, race/ethnicity, sex, educational attainment, and difficulty paying for basic needs. We assessed years of education completed at all follow-up visits, and used education level reported at the most recent examination before death as our primary predictor variable, categorized into ordered, nonoverlapping categories as defined by the original study design: less than or equal to 12 years (high school or less), 13 to 15 years (some college), and 16 years or more (college graduate).¹⁶ We used race/ethnicity as Black or White and sex as male or female as self-reported during the baseline examination. We used difficulty paying for basic needs reported during the most recent follow-up examination.

Our dependent variable was YPLL. We calculated YPLL for those who were deceased by subtracting age at death from the average sex-specific national life expectancy at birth during the year of the baseline examination: 71.1 years for men and 78.2 years for women in 1985. We obtained sex-specific life expectancy data from the National Center for Vital and Health Statistics.

We first compared sociodemographic characteristics by vital status by using the t test for continuous variables and the χ² test for categorical variables. We calculated age-adjusted mortality rates by direct standardization among Black and White participants, overall and stratified by sex, education level at baseline, difficulty paying for basic needs, and field center of recruitment.¹⁹ We determined whether vital status differed by race/ethnicity within education groups and whether vital status differed by education level within race/ethnicity groups. We used the χ² test to assess differences among these groups. Next, we assessed underlying causes of death, stratified by race–sex subgroups.

We subsequently estimated multivariable linear regression models to identify independent associations of predictor variables with YPLL. We first estimated a model that included sociodemographic characteristics: age, race/ethnicity, sex, education level (categorical) at the most recent examination before death, difficulty paying for basics at the most recent examination before death, and the recruiting field center at baseline. We then estimated a model including reported causes of death. We then estimated a combined model, including all variables from the sociodemographic and cause-of-death models. We also tested whether the interaction between race/ethnicity and education was significant in the combined model. For these regression analyses, we excluded data from 46 deceased participants whose cause of death had not yet been fully reviewed.

Because participants who died when younger than 30 years may not have had time to complete higher education, results may be biased to suggest that low education is associated with premature mortality. In our cohort, most participants had achieved their highest level of education by the year-7 examination, with only 1% to 2% reporting a change in education level at each follow-up examination thereafter. As such, we performed a sensitivity analysis with data only from participants alive at age 30 years, and we included their highest level of education completed as of age 30 years. In addition, we performed a supplementary analysis defining YPLL as the participant’s expected age at the time of analysis to mitigate potential bias from projected life expectancy estimates because we do not actually know how long each participant would have lived if the early death had not occurred.

All reported probability values corresponded to 2-tailed tests, which we considered significant at the α = 0.05 level. We performed all analyses with SAS version 9.4 (SAS Institute Inc, Cary, NC).

RESULTS

Through 348 months of follow-up, 417 deaths had been reported, an overall mortality rate of 81.5 per 1000 persons. Table 1 compares characteristics of living and deceased participants. Compared with living participants, deceased participants were more often men and Black, had lower educational attainment, and more often reported having a hard or very hard time paying for basic needs.

TABLE 1—

Comparison of Sociodemographic Characteristics According to Vital Status at the End of Follow-Up: CARDIA, United States, 1985–2017

	Total No. (n = 5114)	Deceased (n = 395), Mean ±SD or No. (%)	Living (n = 4719), Mean ±SD or No. (%)	Pa
Age at enrollment, y	24.85 ±3.7	25.5 ±3.7	24.8 ±3.7	< .001
Sex				< .001
Female	2787	152 (5.3)	2635 (94.5)
Male	2327	243 (10.4)	2084 (89.5)
Race/ethnicity				< .001
Black	2637	246 (9.3)	2391 (90.7)
White	2477	149 (6.0)	2328 (94.0)
Highest education, y	15.2 ±2.6	14.0 ±2.5	15.3 ±2.6	< .001
Maximum education level				< .001
High school or less	1072	144 (13.4)	928 (86.6)
Some college	1676	137 (8.2)	1539 (91.2)
College graduates	2366	114 (4.8)	2252 (95.2)
Pay for basics				< .001
Not very hard	3559	237 (6.7)	3322 (93.3)
Somewhat hard	1026	98 (9.6)	928 (90.4)
Hard	287	33 (11.5)	254 (88.5)
Very hard	242	27 (11.2)	215 (88.8)

Note. CARDIA = Coronary Artery Risk Development in Young Adults study.

^at test for continuous variables and χ² for categorical variables.

The age-adjusted mortality rate was 30 per 1000 persons among Black participants and 17.25 per 1000 persons among White participants (Table 2). Age-adjusted mortality rates were higher among men compared with women, those with high-school education or less compared with those with more than a high-school education, and among those recruited from the Birmingham Field Center compared with all other sites. Within categories of highest education level attained, there was no statistically significant difference in vital status by race/ethnicity (Table 3). However, within both Black and White groups, there were differences in vital status by education level. Among both race/ethnicity groups, the high-school education or less group had the highest percentage of deceased and the college graduates the lowest percentage of deceased.

TABLE 2—

Age-Adjusted Mortality Rate per 1000 Persons by Race/Ethnicity Within Categories of Sex, Education, Income, and Baseline Examination Center: CARDIA, United States, 1985–2017

Race/Ethnicity	All	Black	White	Pa
Overall	24.17	30.00	17.25	< .001
Sex
Male	35.86	45.17	25.20	< .001
Female	14.23	17.63	10.10
Education
High school or less	61.69	48.21	36.66	< .001
Some college	30.74	27.03	20.50
College graduates	12.71	15.29	11.16
Difficulty paying for basics
Not very hard	21.51	29.65	13.59	< .001
Somewhat hard	31.32	31.14	31.54
Hard	28.83	31.15	24.70
Very hard	26.12	28.62	20.00
Field center
Birmingham, AL	30.36	35.50	23.91	< .001
Chicago, IL	20.69	26.83	13.93
Minneapolis, MN	22.35	30.58	14.55
Oakland, CA	23.55	27.32	17.78

Note. CARDIA = Coronary Artery Risk Development in Young Adults study.

^aχ² analyses performed to compare age-adjusted mortality rates among Black and White participants across categorical variables describing sociodemographic characteristics. Age-adjusted mortality rates were calculated by using direct standardization.¹⁹

TABLE 3—

Vital Status by Race/Ethnicity Within Each Educational Category and by Educational Category Within Each Race: CARDIA, United States, 1985–2017

	Deceased (n = 395), No. (%)	Living (n = 4719), No. (%)	P
Race/ethnicity
High school or less
Black	101 (13.5)	645 (86.5)	.88
White	43 (13.2)	283 (86.8)
Some college
Black	97 (9.0)	980 (91.0)	.1
White	40 (6.7)	559 (93.3)
College graduate
Black	48 (5.9)	766 (94.1)	.08
White	66 (4.3)	1486 (95.7)
Educational category
Black
High school or less	101 (13.5)	645 (86.5)	< .001
Some college	97 (9.0)	980 (91.0)
College graduate	48 (5.9)	766 (94.1)
White
High school or less	43 (13.2)	283 (86.8)	< .001
Some college	40 (6.7)	559 (93.3)
College graduate	66 (4.3)	1486 (95.7)

Note. CARDIA = Coronary Artery Risk Development in Young Adults study.

A total of 349 deaths had been fully reviewed for underlying cause of death. Table A (available as a supplement to the online version of this article at http://www.ajph.org) lists cause-specific rates of death by race–sex subgroup. The most common causes of death overall were cancer (16.3%), cardiovascular disease (17.5%), and AIDS (13.5%). The most common cause of death differed by race–sex subgroup, with homicide the most common cause of death among Black men (19.3%), AIDS the most common cause of death among White men (27.8%), and cancer the most common cause of death among Black (23.5%) and White women (31.5%). AIDS contributed 35.4 YPLL and homicide contributed 37.4 YPLL. At our 16-year assessment of mortality, when the cohort was an average age of 41 years, AIDS (28%; n = 35) and homicide (16%; n = 20) were the 2 most common underlying causes of death.¹⁷ Since that report, there have been 12 additional deaths from AIDS and 9 additional deaths from homicide. Over the same time period, there have been an additional 48 deaths from cancer, 52 deaths from cardiovascular disease, 17 deaths from unintentional injury, and 9 deaths from suicide.

In our multivariable regression model, age at enrollment, female sex, and educational attainment were independently associated with YPLL (Table 4). Being female was associated with 6.43 greater YPLL (95% confidence interval [CI] = 4.75, 8.12). Attaining 1 additional level of education was associated with 1.37 fewer YPLL (95% CI = −2.37, −0.37). Race/ethnicity was not independently associated with YPLL, and in an otherwise identical model that included an interaction term between race/ethnicity and education, the interaction term was not significant (b = −1.45; 95% CI = −3.44, 0.55). Homicide and AIDS were the causes of death associated with the greatest YPLL. Cardiovascular disease, cancer, and liver disease contributed fewer YPLL than did AIDS (reference group in the model).

TABLE 4—

Estimated Mean Years of Potential Life Lost (YPLL) Compared With Reference Group, by Selected Participant Characteristics, Among Deceased Participants: CARDIA, United States, 1985–2017

Predictors of YPLL	Sociodemographic Model,a Mean YPLL (95% CI)	Cause-of-Death Model,b Mean YPLL (95% CI)	Combined Model,c Mean YPLL (95% CI)
Sociodemographic characteristics
Black	0.50 (−1.31, 2.30)	. . .	0.81 (−0.92, 2.54)
Age at enrollment, y	−1.11 (−1.34, −0.87)	. . .	−0.94 (−1.16, −0.73)
Female	4.05 (2.29, 5.81)	. . .	6.43 (4.75, 8.12)
Educational attainment, 1 level increased,e	−1.22 (−2.32, −0.12)	. . .	−1.37 (−2.37, −0.37)
Difficulty paying basics, 1 level increasee,f	−0.67 (−1.62, 0.29)	. . .	−0.67 (−1.53, 0.19)
Field center
Birmingham, AL (Ref)	1	. . .	1
Chicago, IL	1.03 (−1.55, 3.61)	. . .	0.48 (−1.83, 2.79)
Minneapolis, MN	−1.44 (−3.65, 0.78)	. . .	−1.19 (−3.16, 0.79)
Oakland, CA	1.42 (−0.88, 3.71)	. . .	1.35 (−0.72, 3.42)
Cause of death
AIDS (Ref)	. . .	1	1
Asthma	. . .	0.19 (−11.77, 12.14)	−2.61 (−12.63, 7.40)
Cancer	. . .	−6.74 (−10.09, −3.39)	−9.05 (−11.96, −6.14)
CVDg	. . .	−7.97 (−11.13, −4.80)	−9.85 (−12.53, −7.17)
Diabetes	. . .	−0.88 (−10.75, 8.99)	−4.74 (−12.90, 3.43)
Homicide	. . .	2.00 (−2.11, 6.10)	−0.15 (−3.64, 3.35)
Kidney	. . .	−6.08 (−12.14, −0.02)	−10.22 (−15.37, −5.07)
Liver diseases	. . .	−10.74 (−16.33, −5.16)	−12.43 (−17.06, −7.80)
Lung diseases other than asthma	. . .	−7.32 (−13.12, −1.52)	−9.24 (−14.04, −4.44)
Others	. . .	−5.51 (−8.98, −2.04)	−7.76 (−10.68, −4.84)
Sepsis	. . .	−14.28 (−22.09, −6.47)	−16.14 (−22.68, −9.61)
Suicide	. . .	−0.84 (−5.48, 3.80)	−4.47 (−8.42, −0.52)
Unintentional	. . .	−0.88 (−4.94, 3.18)	−4.49 (−7.85, −1.12)

Note. CARDIA = Coronary Artery Risk Development in Young Adults study; CI = confidence interval; CVD = cardiovascular disease. Sample size n = 349. YPLL was calculated by subtracting age at death from the average sex-specific national life expectancy at birth during the year of the baseline examination, 71.1 years for men and 78.2 years for women, in 1985.

^aSociodemographic model includes race/ethnicity, age at enrollment, sex, education, difficulty paying for basics, and recruitment field center with YPLL.

^bCause-of-death model reports the independent associations of included categories of causes of death with YPLL. Reference group for the cause-of-death model is the YPLL associated with AIDS. AIDS contributed 35.4 YPLL.

^cCombined model includes all variables in sociodemographic and cause-of-death models.

^dDifficulty paying for basics and education level were assessed at the most recent examination before death.

^eEducation levels: 1 = high school or less; 2 = some college; 3 = college graduates.

^fDifficulty paying for basics: 1 = not very hard; 2 = somewhat hard; 3 = hard; 4 = very hard.

^gDeaths attributed to CVD included deaths from acute myocardial infarction, coronary heart disease, congestive heart failure, primary or secondary arrhythmic death, ischemic stroke, hemorrhagic stroke, abdominal aortic aneurysm, peripheral aterial disease, nonischemic cardiomyopathy, or death caused by invasive cardiovascular intervention.

Results from our sensitivity analysis, including data only from the 320 participants who were alive at age 30 years, were not substantially different from the primary results (Table B, available as a supplement to the online version of this article at http://www.ajph.org). Results from our supplemental analysis defining YPLL as years of life lost compared with expected age at the time of analysis also did not differ substantially from our primary results (Table C, available as a supplement to the online version of this article at http://www.ajph.org).

DISCUSSION

In a prospective cohort of more than 5000 Black or White young adults recruited at an average age of 25 years and followed over the course of 29 years, we found evidence of persistent disparities in premature mortality by race/ethnicity and education. We found key differences in mortality rates and causes of mortality by race/ethnicity, and differences in causes of mortality by stage of life, and we found that education level explained observed differences in mortality by race/ethnicity. Overall, cancer, cardiovascular disease, and AIDS were the most common causes underlying premature deaths, but homicide was the most common cause of premature death among Black men and contributed the most YPLL. Age-adjusted mortality rates were significantly higher for Blacks than for Whites and decreased as educational attainment increased. However, when we examined race and education simultaneously, we found that education—and not race—was associated with premature mortality, with education inversely associated with YPLL. Our study extends previous findings by quantifying racial disparity in premature mortality in terms of YPLL within a longitudinal cohort of Black and White young adults in the United States and showing that, once education is accounted for, this difference is no longer statistically significant.

The education-related disparity in life expectancy described as persisting and worsening through the 1980 to 2010 decades is thought to be attributable to greater gains in life expectancy for those with high education.^1,12,20,21 Studies that used data from nationally representative samples have reported that life expectancy differed by race and education level in years 1980 to 2001, and disparities were greatest among adults younger than 65 years. Similar to our observations, these studies reported educational disparities within all race groups, and they reported larger differences among Blacks. Education-related disparities widened over the decade between 1990 and 2000,¹ and persisted or worsened through the 2000s, especially among White men and women.^2,7 Among Whites, the mortality rate increased between years 1999 and 2013, with the greatest increase among those with high-school education or below, resulting in a narrowing of the race-based disparities gap, but for unfortunate reasons (rather than low-life-expectancy groups improving, high-life-expectancy groups worsened).⁷ Our results provide evidence that these education-based disparities continued well into the 2010s, despite national initiatives to curb disparities in health outcomes and premature mortality.^4,22

Two previous studies reported that race has a residual effect on the association of education with life expectancy.^1,2 However, we found inconsistent and weak relationships between race/ethnicity and mortality among education subgroups, and in our multivariable models, race/ethnicity was not independently associated with YPLL, and no significant interaction was found between race/ethnicity and education. These discrepancies may be attributable to cohort effects. It is possible that race-based disparities are seen in older cohorts because there were even larger differences in educational quality because of racial segregation among Whites and Blacks born before the 1970s.²³ It is also possible that education is the mediator of observed racial disparities in causes of premature mortality among those younger than age 60 years, as in our cohort, and the residual independent effects of race/ethnicity on health and longevity emerge in later years. This is plausible, in that current evidence suggests the long-term impact of Black race on health outcomes is at least in part attributable to the persistent and cumulative negative effects of perceived discrimination.^24,25 Persistent discrimination is thought to lead to chronic stress and inflammation, which are risk factors for hypertension, obesity, diabetes, and heart disease, all of which tend to have the greatest impact on mortality later in life.^14,25–29

Among our cohort, it appears that these racial/ethnic disparities in cardiovascular disease incidence may be emerging, as 70% of those who have died prematurely from cardiovascular disease were Black. Among our full sample, the most common causes of death overall were cancer, cardiovascular disease, AIDS, homicide, and unintentional injury. Homicide was the most common cause of death among Black men, and deaths from suicide were most common among White men and women, consistent with observed national trends.⁷ However, we observed shifts in overall dominant causes of mortality since our 16-year assessment from AIDS and homicide toward chronic diseases, specifically cardiovascular disease and cancer. As the CARDIA cohort continues to age and more deaths accrue from chronic diseases like cardiovascular disease, it is possible that a racial/ethnic disparity in YPLL independent of educational attainment may emerge. However, it remains to be studied whether these potentially emerging racial/ethnic disparities in YPLL from chronic diseases are truly independent of educational attainment or if they are, in fact, mediated by education.

Educational attainment may be indicative of several other psychosocial factors associated with health outcomes. Though we have not examined specific health behaviors, health care access, health beliefs, or other neighborhood factors that might explain these educational disparities in this study, our results support the extensive literature reporting common underlying risk factors over the life course associated with low education that contribute to premature mortality. These factors include behavioral risk factors for heart disease and cancer, such as smoking and obesity, and risk factors for HIV/AIDS, such as condomless sex and needle sharing.³⁰ Furthermore, other social factors such as low neighborhood cohesion and low social support may contribute to youths engaging in violent networks, resulting in higher rates of homicide among young Black men.^31–33 Consistent with this hypothesis, both smoking and low social support were found to be independent predictors of mortality in the CARDIA cohort at year 16.¹⁷ These findings suggest that local programs should simultaneously address both social and health-related risks to have a meaningful impact in mitigating disparities in health and longevity.

Limitations

Our sample was limited to Black and White adults recruited from 4 urban centers in the United States, so results may not be applicable to other racial/ethnic groups, rural communities, or communities outside the United States. Because participation in our cohort and subsequent follow-up examinations was voluntary, it is possible that those who participated are more likely to engage in regular medical care. However, less than 20% of participants were lost to follow-up, and the original cohort was 50% of the population-based sample that was approached. In addition, 46 deaths had not been fully reviewed at the time of the analysis and were excluded from our multivariable regression analyses. Though it is a small number that were lost to follow-up or excluded, it is possible that this group had different rates or causes of death than the group that continues to remain engaged in the study. However, the rates and causes of death we observed were similar to those reported by national cross-sectional samples of US residents of the same age. In addition, though we assessed education and the ability to pay for basic needs as markers of socioeconomic status, there may be other unmeasured markers of socioeconomic status that may contribute to residual confounding.

Finally, it is possible our results are biased because our cohort has not yet achieved their expected life expectancy and those who are alive are censored from the YPLL analysis.³⁴ However, we did not limit our analyses to a specific cause of death (a major source of bias in excess YPLL comparisons³⁴), and our exposures of interest were race/ethnicity and education level, which were balanced by the recruitment design of the CARDIA cohort and may mitigate the risk of this potential bias. In addition, we performed a sensitivity analysis using data among participants alive at age 30 years and their education at that time that yielded similar results, suggesting bias attributable to differential exposure did not significantly influence our results. We also performed a supplemental analysis using expected age at the time of analysis to calculate YPLL that did not yield different results, suggesting bias attributable to the fact that the cohort has not yet achieved their life expectancy did not influence our results; however, YPLL will continue to accrue as there are more deaths among this cohort until they reach their projected life expectancy.

Public Health Implications

In a large, longitudinal cohort of Black and White middle-aged adults recruited from 4 urban centers, we found evidence of persistent race/ethnicity–related and education-related disparities in premature mortality. The relationships among education, race/ethnicity, and socioeconomic status are complex, but we found that education accounts for most of the observed racial/ethnic disparities in our cohort. Taken together with the existing literature that education is a strong determinant of morbidity and mortality,³⁵ these findings have important implications for local policy-makers. City-level policies that support the achievement of equality in educational attainment—for example, increasing affordable housing options and access to high-quality early childhood education—are testable approaches to eliminating observed racial/ethnic disparities in health and longevity.⁵