Life expectancy in the United States has been in decline for the first time in decades, and public health officials have identified a litany of potential causes, including inaccessible health care, rising drug addiction and rates of mental health disorders, and socio-economic factors.
But disentangling these variables and assessing their relative impact has been difficult.
Now, a multi-institution study led by the Yale School of Medicine and University of Alabama-Birmingham has attempted to tease out the relative impact of two variables most often linked to life expectancy — race and education — by combing through data about 5,114 black and white individuals in four U.S. cities.
The lives and deaths among this group of people — who were recruited for a longevity study approximately 30 years ago, when they were in their early 20s, and are now in their mid-50s — shows that the level of education, and not race, is the best predictor of who will live the longest, researchers report Feb. 20 in the American Journal of Public Health. The individuals were part of the Coronary Artery Risk Development in Young Adults (CARDIA) study.
Among the 5,114 people followed in the study, 395 had died.
“These deaths are occurring in working-age people, often with children, before the age of 60,” said Yale’s Brita Roy, assistant professor of medicine and epidemiology and corresponding author of the paper.
The rates of death among individuals in this group did clearly show racial differences, with approximately 9% of blacks dying at an early age compared to 6% of whites.
There were also differences in causes of death by race.
For instance, black men were significantly more likely to die by homicide and white men from AIDS.
The most common causes of death across all groups over time were cardiovascular disease and cancer.
But there were also notable differences in rates of death by education level.
Approximately 13% of participants with a high school degree or less education died compared with only approximately 5% of college graduates.
Strikingly, note the researchers, when looking at race and education at the same time, differences related to race all but disappeared: 13.5% of black subjects and 13.2% of white subjects with a high school degree or less died during the course of the study. By contrast, 5.9% of black subjects and 4.3% of whites with college degrees had died.
But there were also notable differences in rates of death by education level.
Approximately 13% of participants with a high school degree or less education died compared with only approximately 5% of college graduates.
To help account for differences in age-related mortality, the researchers used a measure called Years of Potential Life Lost (YPLL), calculated as projected life expectancy minus actual age at death.
This measure not only captures numbers of deaths, but also how untimely they were. For example, someone who dies at age 25 from homicide accrues more YPLL than someone who dies at age 50 from cardiovascular disease.
It would take two deaths at age 50 to equal the YPLL from a single death at age 25.
Even after accounting for the effects of other variables such as income, level of education was still the best predictor of YPLL. Each educational step obtained led to 1.37 fewer years of lost life expectancy, the study showed.
“These findings are powerful,” Roy said. “They suggest that improving equity in access to and quality of education is something tangible that can help reverse this troubling trend in reduction of life expectancy among middle-aged adults.”
A vast body of literature has investigated the association between mortality and health on the one hand and education on the other [1–11]. Previous research has confirmed that individuals with higher socioeconomic status (SES) have lower mortality risks regardless the SES measure (education, income or occupation) used.
The accumulated evidence demonstrates that the SES-health gradient applies not only to the working-age population, but also to the older individuals. The self-perceived health, mental well-being, and functional limitations of people at higher ages are affected by their education, income, and previous occupation [12–15].
In Germany, efforts to study mortality differentials by SES have been hampered by the lack of information on socioeconomic status on death certificates . Due to strict data protection rules, the linkage of individual records is not permitted in Germany.
Such linkages have, however, been performed in many other countries (e.g., for Austria [16–18]; for Switzerland [19–20]; for Finland [21–22]; for Norway  and for Finland and Norway ; for Lithuania [25–26]; and for Belgium ).
Despite these data availability constraints, several existing studies have assessed mortality differentials in Germany. They were based on either sample surveys or data from the German Federal Pension Fund, and used different SES indicators [5, 28–35].
Luy and colleagues  used data from the German Life Expectancy Survey to estimate the differences in length of life among West German citizens by education, household income, and occupation.
They highlighted the lack of national mortality data by socioeconomic status in Germany, and illustrated that there are alternative ways of estimating life expectancy using survey data with a mortality follow-up.
The authors were able to estimate life expectancy at ages 40 and 65, as well as the probability of surviving between both ages separately for each SES dimension. The results revealed substantial differences in life expectancies across all SES groups.
For example, the gap in life expectancy between the highest and the lowest educated men aged 65 or older was 3.7 years.
Kroll and Lampert  used the German Socioeconomic Panel (GSOEP) data for the estimation of differences in life expectancy, while taking income into account.
They used survival models with exponential baseline hazards to estimate relative mortality risks for different socioeconomic groups, and combined them with official life tables for Germany.
The individual’s age was introduced into the model as the covariate, in addition to five distribution-based income groups.
The mean net equivalent income was used for categorizing income, where the lowest category (less than 60% of the mean equivalent income) represented the group with a relatively high poverty risk, and the highest category (more than 150% of the mean equivalent income) represented the group with a relatively high level of material prosperity.
The analysis was done separately for men and women. Their results revealed a clear relationship between income and mortality, with higher income groups having higher life expectancy.
The life expectancy of men at age 65 in the highest income group was estimated to be 23.5 years, or 7.9 years higher than that of their counterparts in the lowest income group. Similarly, the life expectancy of women at age 65 with the lowest income was estimated to be 17.9 years, or 6.8 years lower than that of the women in the highest income category. However, the authors acknowledged the general problem that life expectancy based on the GSOEP data may be overestimated.
Doblhammer et al.  used the GSOEP data for 1991–2006 to evaluate the effect of family status, education, monthly income, professional position, household size, and health satisfaction on the life expectancy of men and women aged 50 or older.
The authors applied the Gompertz model to estimate the relative risks for all covariates, and then the age-specific mortality rates. In addition, they used the estimates for smoking, drinking, blood pressure, and diabetes from the available literature.
Their results on the differences in life expectancy by education for both men and women were similar to those of previous studies.
For many European countries, it has been demonstrated that relative inequalities in mortality have increased in recent decades [3,17,36–38]. In general, this increase appears to be more pronounced for men than for women.
The findings on the changes in absolute inequalities differ: some studies have reported no change , while others have suggested that these inequalities have been narrowing over time . By contrast, Schumacher and Vilpert  found that in Switzerland between 1990–1995 and 2000–2005, educational differentials remained almost constant in relative terms, but increased in absolute terms.
While the number of comparative studies for European countries on the trends in SES disparities in overall mortality and from specific causes has been growing [3,36,38,40–41], Germany is not included in any of them.
Most of these studies have explored either the social gradient in mortality or in health, but some have combined these two indicators into healthy life expectancy. Previous research has also confirmed the existence of a reverse relationship between education and poor health, as approximated by the ability to perform activities of daily living and self-rated levels of health disability [13,42].
Moreover, a systematic association between education and healthy life expectancy has been confirmed, whereby better educated people can expect to live longer and to have more years in good health than people with less education. It has also been shown that the SES gradient is much steeper for health expectancy than for life expectancy [22,24,43].
While it is difficult to estimate the magnitude of the social gradient in German mortality, it is even more problematic to explore the differences in levels of long-term care need in Germany by educational attainment; again due to the lack of appropriate data. Even the data collected by the census on mandatory LTC insurance [44–45] do not include information on the education of individuals.
Scholz  estimated long-term care-free life expectancy at birth and age 60 for German men and women using the official long-term care statistics and data from the Human Mortality Database (HMD) for 2013. The results indicated that, on average, men aged 60 or older can expect to live another 21.38 years, of which 19.43 years will be LTC-free and 1.95 years will be with care needs.
Kreft and Doblhammer  reviewed selected publications on the trends in LTC needs in Germany, and highlighted the inconsistency of the findings. They reported that some studies found evidence supporting the compression of LTC, while others found evidence supporting either a dynamic equilibrium scenario or the expansion in LTC in recent decades in Germany. The authors used administrative census data covering all German LTC insurance beneficiaries (2001–2009) to estimate CFLE and CLE by different care need levels (any or severe) for 412 counties.
Their results indicate that in the majority of these counties, there has been an expansion of any care needs, but a compression of life years with severe care needs. The authors also demonstrated that mortality, and not morbidity, has been the driving force in the absolute changes in CFLE and CLE.
Long-term care insurance in Germany
To measure care need, we use information on individuals receiving benefits from the mandatory German long-term care insurance system (LTCI, introduced in Germany in 1995), which also covers people with mandatory private health insurance (detailed descriptions of the LTCI scheme and its changes over time are available elsewhere [47–51]).
According to data from the Federal Ministry of Health , at the end of 2017 there were 3.302 million beneficiaries of mandatory social long-term care insurance, of whom 77.3% were aged 65 or older. The data further showed that 1.247 million recipients of LTC were men, 66.9% of whom were aged 65 or older (for women, the corresponding figures were 2.055 million recipients and 83.7%). In addition, there were an estimated at 0.189 million private LTCI recipients at the end of 2016.
In December 2015, about three-quarters (73%) of people in need of LTC were cared for at home, while the other 27% were living in a care home . Women represented 61% of those cared for at home and 72% of those living in a home for the older individuals. The oldest old, or those aged 85 or older, represented 32% of those receiving help at home and half of those living in a care home.
The Federal Ministry of Health  has projected that the number of LTC benefit recipients will increase to 5.32 million by 2050. It seems likely, however, that since LTCI provides only partial support, whereas additional care expenses must be paid privately [50–51], an individual’s socioeconomic status (SES) will have a large impact on his or her use of care.
The results of a previous analysis on the effect of education on the incidence and prevalence of LTC  confirmed that educational attainment matters for the utilization of care; i.e., that older adults with higher educational levels have a lower incidence and prevalence of LTC use than their less educated counterparts. The effect was found to be weaker for incidence than for prevalence.
Objectives of the study
The present work aims to examine the relationship between SES and life expectancy with and without LTC needs among older adults in Germany. It also evaluates the changes in the SES–LE/CFLE/CLE gradient over the 1997–2012 period.
As a proxy for socioeconomic status we use education, which is known to be a reliable predictor of health and mortality . Like earlier studies on mortality gradient in Germany [33–34,55], we focus here on estimating the results for men only. One of the reasons why we did not analyze the results for women (provided in S1 Appendix) is the lack of consistency in the mortality estimates from GSOEP for the second period (see the strengths and limitations section for more details).
Based on previous research, we hypothesize that in Germany, the share of life expectancy in which individuals require long-term care in overall life expectancy is rising over time (i.e., that there is an expansion of LTC). We also expect to observe a strong educational gradient in mortality and care needs that tends to increase over time.
Summary of findings
Based on three different individual-level datasets (GSOEP, Microcensus, and health insurance records), we assessed the relationship between educational attainment and long-term care-free life expectancy among older German men. We also investigated whether and, if so, how the education-LE/CFLE/CLE gradient changed between 2004 and 2012.
The available evidence suggests that there are substantial differences in mortality by socioeconomic status in Germany and elsewhere. To our knowledge, this is the first study that has provided estimates of long-term care-free life expectancy by educational level.
Our results indicate that while disparities in LE and CFLE by educational status are increasing, when life expectancy with LTC is considered, the number of years with care needs has been increasing over time, but the variation across the educational groups has not been pronounced.
This trend was found to be unchanged over time. Overall, the increase in the number of expected years of life without care and the decline in the health ratio of expected years of life without care to total expected years of life suggest that long-term care needs are expanding, regardless of the level of educational attainment.
Our estimates reveal that the education-LE and the education-CFLE gradients have been increasing over time: the difference in the number of years of CFLE between the lowest and highest educational groups increased by about 1.3 years (from about 3.3 years in 1997–2004 to almost 4.6 years in 2005–2012). The change in the education-CLE gradient was found to be insignificant.
The declining health ratio suggests, however, that while older German men are living longer, they are also becoming more likely to need LTC (expansion of LTC needs). This trend is occurring across educational levels. Among our more interesting findings is that the health ratio did not vary much between the educational groups. This means that there is no obvious and strong effect of education on the proportion of years with LTC needs, and that mortality does not differ much by educational group once people become dependent on care. It is the long-term care-free years of life that are influenced by SES.
Among the possible explanations for the diminishing educational gradient that can be observed as soon as older men start to need care is the postponement of morbidity (and disability) up to a certain age (e.g., 60 years) among men with higher SES. After these men grow older, their rates of health decline accelerate and their LTC needs increase more than is the case for men with lower SES.
Thus, the health differences between men in different SES groups become smaller at older ages [72–73]. Another potential explanation for this pattern is selective mortality: i.e., that the most disadvantaged people die at earlier ages, while relatively robust individuals survive .
Moreover, social conditions might contribute to the decrease in SES inequalities at older ages, as the interplay of retirement and welfare state policies may slow down the decline in health among the most disadvantaged older adults .
Yet another possible explanation for the diminishing effect of education (or any other SES proxy) on LTC use is that when a person is in need of care, impending death becomes the largest contributing factor (and not socio-demographic or health factors).
Gerstorf and colleagues , for instance, analyzed GSOEP data for people aged 70 years and above, and found that at higher ages, proximity to death is related to the loss of intellectual and sensory functioning, and that late-life changes in well-being are characterized by terminal decline. Given that people in need of care have limitations in physical and/or cognitive abilities, terminal decline might influence their use of LTC more than any of their socioeconomic characteristics.
In general, the results might be influenced by the indicators chosen for approximating SES (e.g., education, occupation, or income) and health (e.g., self-perceived health, cognitive impairment, disability, or long-term care need) . The effect of income-related differences on individuals’ self-perceived health might produce different results than the impact of educational differences on the use of LTC.
Our findings are in agreement with those of Hoffmann and Nachtmann , who used official LTC statistics (micro data on recipients of benefits) for Germany.
They found that between 1999 and 2005, there was an increase in the years spent in good health, but a decline in the share of healthy years relative to remaining life expectancy, which points to a relative expansion of morbidity. However, their analysis only looked at general trends in life expectancy, with no SES dimensions taken into account.
In another study on changes in LTC needs by small areas in Germany over the 2001–2009 period, Kreft and Doblhammer  split LTC into care levels. They found that there was an expansion of all care needs in the majority of the regions, but a compression of the most intensive care needs.
The expansion of morbidity hypothesis was supported in many countries by the findings of one of the earlier studies by Salomon and colleagues .
They used Global Burden of Disease data to estimate healthy life expectancy for 187 countries, and found that in most of these countries, an increase in overall life expectancy was followed by a rise in the number of healthy years lost to disability. In their discussion of the forces driving increasing healthy life expectancy, the authors highlighted reductions in child and adult mortality, rather than the decline in the number of years with disability.
When studying the SES-health association, it is important to determine whether the influence an individual’s SES has on health changes with age. The results of studies that have examined the development of the health-SES relationship over the life course, and particularly in later life, have been inconsistent.
Schöllgen and colleagues  reviewed previous studies and reported that research has provided evidence for three contradictory hypotheses: the cumulative disadvantage theory (which posits that the effect of SES on health status increases with age), the age-as-leveler hypothesis (which argues that the impact of SES declines as the person becomes older), and the continuity theory (which posits that the SES-health relationship remains stable or changes little with age).
The study that used data from the German Aging Survey has found that social inequalities in health in Germany have been either stable or increasing with age, depending on the SES and health indicators used.
Although our data do not allow us to conduct a cohort analysis, the issue of cohort trends in health should be given some attention. On the one hand, the cohorts in our study profited from the first educational expansion after World War I, and from having better career prospects after World War II . On the other hand, they were born during periods characterized by recession and war, which have been shown to be associated with long-term negative consequences for health [79–81].
Recent research for Germany on the education-health association and, on how this association changes across individual lives and cohorts , has pointed to a rapid widening of the educational gap in health with age, and thus supports the cumulative advantage hypothesis.
The results of this research also provide evidence of a divergence in health trajectories between educational groups in all of the analyzed cohorts, with the difference being more pronounced among the younger cohorts (referred to as the economic -wonder and baby -boom cohorts), and less pronounced among the older cohorts (referred to as pre-war, -war, -and–post-war cohorts).
Another important finding is that the steeper health decline observed among men with lower education is mainly attributable to the cross-cohort trend, and that no cross-cohort differences can be found among highly educated men. This analysis was based on self-rated health among people aged 23–84.
Whether the same conclusions can be drawn when LTC needs are used as the indicator of an individual’s health is uncertain. As we mentioned above, the results of such an analysis can depend on the measures chosen for approximating health and SES. Further studies are needed to explore how the association between education and the need for long-term care changes when age and cohort are taken into account.