Abstract
We expand on previous studies investigating the links between early health and later health by examining four distinct dimensions of early-life health and multiple life course outcomes, including the age of onset of serious cardiovascular diseases (CVDs) and several job-related health outcomes. The four dimensions of childhood health are mental, physical, self-reported general health, and severe headaches or migraines. The data set we use includes men and women in 21 countries from the Survey of Health, Ageing and Retirement in Europe. We find that the different dimensions of childhood health have unique ties to later outcomes. For men, early mental health problems play a stronger role for life course job-related health outcomes, but early poor/fair general health is more strongly linked to the spike in onset of CVDs in their late 40s. For women, these links between childhood health dimensions and life course outcomes are similar but are less clear-cut than for men. The spike in onset of CVDs in women's late 40s is driven by those with severe headaches or migraines, while those with early poor/fair general health or mental health problems do worse as captured by job-related outcomes. We also delve into and control for possible mediating factors. Exploring the links between several dimensions of childhood health and multiple health-related life course outcomes will enable a better understanding of how health inequalities originate and are shaped over the course of people's lives.
Introduction
Previous research has illustrated the multidimensional nature of health (Kalwij and Vermeulen 2008) and that both childhood-specific diseases and childhood health summary variables contain useful information about predicting adult health (Smith 2009a, 2009b). A recent review highlights that even mild health shocks early in life can have permanent adverse effects on later-life health (Almond et al. 2018). Here, we expand on earlier research on the links between early health and later health by exploring the influence that several distinct dimensions of early-life health have on multiple lifelong outcomes. These later outcomes include age of onset of specific health conditions, such as serious cardiovascular diseases (CVDs), permanent or temporary withdrawals from the labor force over the life course, and potential mediating factors. Quantifying how and which early-life health conditions predict these health outcomes—especially an individual's onset of CVDs—is important because they are the leading causes of death worldwide (World Health Organization 2021), including in high-income countries (Lozano et al. 2013). Because early detection is crucial to improve disease management and reduce morbidity and mortality among individuals with CVDs (Luo et al. 2019), such knowledge could have important life-improving benefits.
One of the main contributions of this article is, therefore, to reconstruct individuals' full lifelong (health) histories to investigate ties with multiple dimensions of early-life health. In particular, we will explore the link between four dimensions of early-life health and (1) the probability of the onset of one of six health conditions, including five cardiovascular diseases and chronic lung disease, over the life course in five-year intervals between the ages of 16 and 80; (2) the probability of experiencing at least one or at least two episodes of ill-health during adulthood; (3) the accumulated number of career gaps in employment in five-year intervals between the ages of 21 and 64; (4) the probability of leaving the labor force in five-year intervals between the ages of 21 and 64; and (5) the probability of having retired from a job because of ill-health. We will also explore several additional life events that may be influenced by early health, including educational attainment, having ever married, and ever having a child. In doing so, we will control for country of childhood fixed effects (FEs), cohort FEs, and their interactions, for whether the current country of residence is the same as the country of childhood, and for measures of childhood family socioeconomic status (SES). Our analysis sample includes men and women from 20 European countries and Israel. We limit the sample to those who survived until at least age 50; thus, those who may have been most dramatically influenced by early health problems are, unfortunately, likely not in our analysis.1
Literature Review
A growing literature demonstrates that the SES–health gradient in adulthood has its origins in an individual's early life (Case et al. 2002; Currie and Stabile 2003). Two chapters in the Handbook of Labor Economics (Black and Devereux 2011; Currie and Almond 2011), for instance, show that adverse health events in early life and parental SES have long-lasting effects on later-life health and SES-related outcomes, such as earnings and work effort. Furthermore, such adverse health events early in life do not need to be severe to have enduring effects on later-life health (Almond et al. 2018). The research includes literature tied to the fetal-origins hypothesis (e.g., Almond and Currie 2011; Barker 1995), which suggests a direct link from the prenatal period to adult health that may be independent of social class in adult life. It also includes work that explores the tie between illness and deprivation during childhood, unveiling possible long-term consequences for health during adulthood, both directly via a tie from early SES and early health to later health (Montez and Hayward 2014; Oi and Haas 2019) or indirectly by restricting educational achievement and life opportunities (e.g., Kuh and Wadsworth 1993). Other studies, such as Marmot et al. (2001) and Yang et al. (2017), found that the observed SES–health gradient in adulthood is only indirectly attributable to early-life events through later-life events. Later work, discussed in the following, also presents evidence that having good health during childhood and growing up in a more comfortable environment result in a higher level of education, as well as good health and higher economic status later in life. All these findings are important to policymakers because they may suggest that policies aimed at improving children's health and SES have long-lasting benefits for both the individual and society through increased human capital accumulation—hence better employment opportunities and better later life health (see also Marmot et al. 2012).2
In work that is slightly more closely tied to the question we explore here, several authors established a link between specific health conditions and adult health. For example, Currie and Stabile (2006, 2009), Fletcher and Wolfe (2008, 2009), and Currie et al. (2010) showed that childhood mental health problems have larger long-term effects on outcomes than those of specific physical health conditions. Currie et al. (2010), using data from Manitoba, Canada, for 50,000 children and their siblings linking health data to information on school performance and welfare use after age 18, examined the effects of having been diagnosed or treated for mental health problems in four age ranges (0–3, 4–8, 9–13, and 14–18) and compared the impact of mental health problems with those of having been diagnosed or treated for asthma and injuries. They found that mental health conditions have more serious consequences than the two physical health conditions.3
One of the missing aspects of most of these studies is that they focus on the fetal period and young children. As Currie (2020:10) pointed out, adolescence is quite understudied, and called it “the missing middle.” Yang et al. (2017:88) put this differently as “A dominant paradigm to explain the early- and later-life connection is the sensitive period model, which posits that exposures during sensitive periods of development (e.g., gestation, birth, childhood, and adolescence) induce enduring structural and functional changes in organisms through biological programming that are difficult to reverse and, in turn, affect later disease risk.” The authors further suggested that the focus on gestation, birth, and early childhood is based on a model that early-life conditions have stronger effects on adult outcomes than conditions experienced at subsequent time points. But do they? And at what age are these effects most commonly manifested? Yang et al. (2017) studied the influence of SES on two measures of health—inflammation and metabolic syndrome—among adults by age group. Their most relevant finding was that the link between SES and the health of older adults is through SES as an adult rather than as a child (however, SES as a child does influence SES as an adult).
The literature most closely tied to our research on lifelong health is far more limited.4 One early article is by Case et al. (2005), who used the National Child Development Study—which followed all children born in Great Britain in the week of March 3, 1958, from birth through to age 42—to study the link between childhood health and adult health, as well as education and SES. Their key early health measure was the number of physician-assessed chronic health conditions observed at ages seven and 16, and their key adult health measure was self-reported health (SRH) at ages 23, 33, and 42. They found that chronic health conditions at age 16 are tied to adult health at all three points, while the effects of conditions at age seven fade over time. When they looked at type of conditions, they found that physical impairments, mental and emotional conditions, and systems conditions all contribute approximately equally to poorer health at age 42.
A more recent article by Smith et al. (2012) used data similar to ours: the China Health and Retirement Longitudinal Study (CHARLS), which focused initially on two provinces in China and was modeled after the U.S. Health and Retirement Study (HRS) and the Survey of Health, Ageing and Retirement in Europe (SHARE). In the study, the authors explored the tie between childhood health (measured by SRH before age 16) and adult health (also measured by SRH), as well as doctor diagnoses of chronic illnesses, depression, lifestyle and health behaviors, and (instrumental) activities of daily living as an adult. They plotted the percentage of respondents reporting good or better health between ages 45 and 80 and found that those whose childhood health was good or better are four times more likely to have good or better health as an adult than those with poorer health as a child. The authors found strong statistical associations between childhood health and adult health outcomes but, interestingly, mainly only for Chinese women. Their main outcomes were measured in 10-year age intervals, with sensitivity checks for five-year and one-year intervals.5
Our contribution goes beyond these two studies in the following ways. First, we use four measures of childhood health before age 16: self-reported or general health, a measure of physical health, a measure of mental health, and a measure of severe headaches or migraines. These measures are more extensive than those used in other studies, although, as in the study of China, they are retrospective.6 Second, we measure multiple aspects of adult health with a focus on both job-related health and onset of specific health conditions. This is unique in our knowledge of the literature. Third, we measure outcomes over all of adult life in five-year intervals. This is similar to the approach used by Smith et al. (2012) for China, but more detailed. We also focus on the exact age at which a person is most likely influenced by each particular type of health problem we measure.
Data and Empirical Model
Data
We use individual-level data from SHARE, a multidisciplinary and representative cross-national panel of the European population aged 50 and older. Our measures of childhood health problems and later-life measures of health problems in each year are mainly retrospective. For life course health conditions and job-tied health measures, they include, correspondingly, data over ages 16–80 and 21–64 in one-year intervals, combined into five-year buckets.7
We use data from the first seven waves of SHARE, which were collected in approximately biannual periods between 2004/2005 and 2017 (Börsch-Supan et al. 2013).8 These include information on sociodemographic background characteristics, current health, and socioeconomic status. We construct our life course health measures using information from regular Waves 1, 2, 4, 5, and 6. These waves include a question on a battery of health conditions (PH006) and one on their age of onset (PH009). To reconstruct life course health profiles, we focus on six conditions, including five cardiovascular diseases and lung disease, which were asked about in all the waves.9 Most of our additional data are from two retrospective waves: Wave 3, also known as SHARELIFE (2008/09), and Wave 7 (2017), in which a SHARELIFE questionnaire was administered to all respondents who did not participate in Wave 3. SHARELIFE was, thereby, expanded to 14 additional countries that entered SHARE after Wave 3, as well as Israel, which had exited SHARE in Wave 3 and returned in Wave 5. SHARELIFE contains retrospective information on respondents' early-life circumstances and work careers. Interviewers in SHARELIFE used a life history calendar to guide respondents to answer questions as accurately as possible.10
Our analysis sample includes men and women from 21 SHARE countries.11 The initial sample consists of 69,373 respondents aged 50 and older in the interview year of SHARELIFE (in Wave 3 or 7).12 The final sample includes 93% of these respondents—28,266 men and 36,325 women. We drop those with missing values in childhood health variables and other childhood SES controls, reducing our sample by 1,285 and 2,954, respectively. We also exclude 331 respondents with missing values for regular dental care since childhood and 212 respondents with no information on their country of birth or on their age at which they moved to their current country of residence.
Our analysis was conducted separately for men and women. We did this to acknowledge that labor force participation—and career gaps, in particular—frequently differ for men and women, and did so for all outcomes to be consistent. The differences in these patterns convince us that, to understand the role of early health conditions, we should analyze men and women separately and that, for policy purposes, it is important to understand the patterns for both genders.
Measures of Early Health
We use retrospective self-assessed data on general, physical, and mental health and on severe headaches or migraines that refer to the period before an individual attained age 16. This categorization is fairly typical in studies of health status: physical health is based on reports of experiencing a set of illnesses that are primarily physical in nature, mental health is based on a set of severe mental illnesses, and general health is based on a more subjective overall assessment of health (Smith 2009a, 2009b). Our measures attempt to make full use of the rich data that were collected. Our measure of general health employs the commonly used self-reported five-point scale (excellent to poor), which we convert to a binary variable identifying the two lowest categories of health—poor or fair health—versus all others.13 Our measure of physical health is created from a polychoric principal component analysis (PCA) (Kolenikov and Angeles 2009) and includes count variables for respiratory problems (asthma, other respiratory problems, and allergies), infectious diseases (polio, severe diarrhea, meningitis/encephalitis, and other infectious diseases), cardiovascular diseases (diabetes or high blood sugar and heart troubles), and disorders of the sense organs (chronic ear problems, speech impairment, and difficulty in seeing even with eyeglasses), as well as two additional dummy variables (for appendicitis and other serious health conditions suffered before age 16). Polychoric PCA is applied to the pooled samples of men and women, and we keep the first principal component (PC), which explains the largest proportion of the total variance—26% for men and 28% for women (cf.Poterba et al. 2017). All the factor loadings on the first PC have an expected sign in determining (adverse) childhood physical health (see Table A2, shown in the online appendix, along with all other tables and figures designated with an “A”). We convert this into an index of good health and create a dummy variable indicating whether an individual is one standard deviation below the mean in the distribution of this first PC to create our index of childhood physical health. Our measure of mental or emotional health is based on responses to questions of whether respondents experienced emotional, nervous, or psychiatric problems, or epilepsy fits or seizures, before age 16. We also have information on severe headaches or migraines, but because it is not clear whether to designate these as physical health, mental health, or both, we instead use the response as our fourth measure of childhood health.14
The values for these health variables, along with background measures, are available for men and women in Table A3. Overall, 25.2% of men and 30.0% of women reported at least one of the four early health conditions, and 5.4% and 7.3%, respectively, reported two or more. As part of our research, we compare using a single indicator of adverse health, which measures whether an individual had at least one of these early health conditions—an approach closer to that of other researchers—with the use of all our four early health measures.
The four measures of health we use are largely independent. Overall, the correlations between any two of them are quite low, as shown in Table A4. The correlations for men are .06 for mental and physical health, .13 for mental and poor health, .20 for physical and poor health, and .09 between severe headaches and mental, poor, or physical health; for women, these correlations are .08, .13, .20, and .10–.12, respectively.
Measures of Life Course Health
Our life course health variables include two types of measures: a set of aggregate measures over working years or to the age of interview in SHARELIFE (Wave 3 or 7) and a set of measures reported by five-year intervals for job-tied health and health conditions (see footnote 7).
In the first set of aggregate measures, we include whether an individual (1) ever reported one period of poor health in adulthood, (2) reported two or more such periods, or (3) retired because of health problems. Next, we explore whether an individual (4) ever worked and (5) their number of career gaps, both by age 64 or the age of interview in SHARELIFE if below age 64 (means, standard deviations, and number of observations are provided in Table A3).
In terms of outcome measures reported over five-year intervals, we include whether an individual (6) had a gap in their employment. We use five-year intervals over ages 21–64 (or the age of interview in SHARELIFE if below age 64), where the responses are cumulative (see Figure 2). Related to the gap measure, we have whether an individual (7) had retired, another cumulative measure also over ages 21–64 (see Figure 3). Finally, and as a primary outcome of interest, we include (8) a measure created from reports of particular health conditions and measured in five-year intervals capturing first incidence, over ages 16–80 (see Figure 1). The aggregate conditions reported here include five CVDs: heart attack or other heart problems, hypertension (high blood pressure), high blood cholesterol,15 stroke or cerebrovascular disease, and diabetes. We also conduct sensitivity checks on single CVDs and stepwise add chronic lung disease and cancer to the overall set of CVDs (in Figure A1, we show the results for the five CVDs along with lung disease; all others are available upon request). The respondents were asked if they ever had each of these conditions and, if so, the age they first had the condition (see Data section). We think of these reports as a measure of incidence.
The full sample of 28,266 men and 36,325 women reports these outcome measures through age 50, but the sample then becomes smaller because some respondents are not yet in the older age brackets; mortality further reduces the sample (see, e.g., the bottom lines in Tables A6 and A7). In Table A3, we use the smallest sample for those who report they retired because of health—18,037 men and 18,870 women (this excludes those who have not yet retired).
We note that the pattern observed in Figure 1 of a rapid increase in incidence of a health condition–that is, a CVD—is likely dominated by hypertension, which is the most common of these diseases but tends to be rare until middle age or the mid-40s. McGrath et al. (2019) found that hypertension accounted for the greatest burden in terms of disability-adjusted life-years among middle- and older-age adults in the United States. Because of this importance, we highlight CVD in our analysis.16
Other Variables
Given the multicountry and multicohort nature of the data, in all estimates, we add FEs for 18 countries of childhood and for 7 birth cohort groups (using one as a reference in each set). To further absorb systematic differences in the life courses between individuals from different countries and cohorts, we include 17 × 6 interactions between the childhood country FEs and birth cohort FEs (for some outcome variables with limited or no variation across specific country-cohort combinations, we group the birth cohorts further and adjust the number of interaction terms accordingly).17 The mean ages of the samples are 67.2 for men and 66.9 for women. The oldest in the sample were born before World War I (0.07% of the sample). We also include a survey year dummy variable (for Wave 7) and various measures of the SES of the family when the individual was age 10, including the number of rooms per person, two dummy variables for limited bathroom or heating facilities and no books in the home, and another for having regular dental care since childhood. In addition, we control for whether the individual was born in an urban area and was born in or moved to the current country of residency during childhood.18
We also include a set of nonhealth later-life outcomes, which we explore in their own right and consider as potential mediators of the relationship between early health and later health. These include years in full-time education; whether final education in ISCED units was less than 3, 3–4, or 5-6; whether the individual ever married; and whether they ever had a natural child. The average number of years of schooling for our sample is 13.0 for men and 12.3 for women. Means, standard deviations, and number of observations for all these variables are reported in Table A3, by gender.
Empirical Model
For each of our eight life course health measures, y, we run a series of reduced-form equations of the following sort:
where i is each individual, and a represents an aggregated age interval (e.g., 20–64) for measures 1–5, or each five-year age group from 21 to 64 for the job-tied health measures 6 and 7, or from 16 to 80 for the health conditions measure 8 (see Measures of Life Course Health section). Mental, Physical, Poor, and Head are our core early-life health variables. Other is a vector with regular dental care since childhood and all other childhood SES indicators. Finally, Controls is a vector including country of childhood FEs, cohort FEs, interactions between childhood of country FEs and cohort FEs, one survey year FE, and a dummy variable for whether the respondent's current country of residence is the same as the country of childhood (see Other Variables section).
We also explore several potential mediating outcomes, such as education, (either years in full-time education or dummy variables for educational attainment), whether a person ever married, and whether a person ever had a natural child (see Table 2). Moreover, to test if the effects of our early health conditions operate through possible mediating factors, we run an additional set of regressions for our life course health measures 6–8 in which we add the years in full-time education and both ever having married and ever having a natural child in the corresponding age interval in the vector Mediators as additional control variables in Eq. (1):
The results from this latter analysis are shown in Figures A3–A5.19
Results
Descriptive Analysis
We begin by looking at age patterns of CVD health conditions and compare the age-specific incidence of those with early mental or physical health problems, severe headaches, or poor/fair health as a child to those without any of these four early conditions; we do so separately for men and women. In Figure 1, the graphs on the left show the first incidence of the included conditions and those on the right show the accumulated probabilities of having these conditions, or prevalence. All graphs cover ages 16–20 until ages 76–80.
The overall first incidence pattern for these combined serious conditions increases by age, reaching a maximum at ages 46–50, except for men and women with no childhood health problems and women who had poor/fair early health, who hit this maximum a decade later. For men, the highest incidence rate up to age 46–50 is for those who had a childhood mental illness. The pattern changes at that age, where mostly those with poor/fair general health as a child now show the greatest first incidence over the older years. Those with early mental illness tend to have the lowest incidence beginning at 56–60, perhaps because they are the most likely to have already experienced these conditions. The graphs on the right showing the cumulative probabilities (i.e., once a person reports they had an illness, they have a 1 for that indicator) suggest that those who had an early mental condition have the highest prevalence to the mid-50s; however, during their 60s and 70s, those who had poor/fair general health and severe headaches as a child gradually overtake them in terms of reports of these serious conditions. Males who did not report one of these conditions as a child continuously report the lowest prevalence rate up through their mid-70s.
Women show a similar pattern as men, but with slight differences. As noted, except for women with poor/fair health and those with no health problems as a child, the incidence rates are highest at ages 46–50, especially for those with severe headaches. Consistent with the results for men, after 46–50, the incidence rate is highest for those who reported early poor/fair health and lowest for those with early mental illness. The general pattern of cumulative probabilities is also similar to that of men, with those with no childhood health problems continuously reporting the lowest prevalence rate and those with severe headaches as a child exhibiting the highest cumulative probability of having these serious health conditions in the oldest age group (76–80).
We next look at those who report gaps in work over ages 25–64. In Figure 2, we see that for both genders, those who had an early mental health condition report the greatest number of gaps, but the pattern is far starker for men. Among men, by age 35, those who reported an early mental health condition report more than one gap, on average (1.2); this increases to 2.6 by age 45 and to nearly 5.0 by age 60. Those with poor/fair health and severe headaches as a child report the second and third highest numbers of gaps, but they are far below those reported by men with an early mental health condition. Men with a childhood physical health problem appear to not differ from those with no childhood health problem.
Turning to women, first we note the much larger number of gaps in working than for men. This difference may be tied to both norms and childbearing. At all ages, up until age 61–64, women who had an early mental health condition report the greatest number of work gaps (close to 6.0 by age 45 and nearly 12.0 by age 60). Interestingly, it appears that women with early physical health problems had fewer gaps than those with no early health condition at later ages, but the differences are small (less than half of a gap at age 61–64).
Our last measure of life course health is the pattern of retiring by age shown in Figure 3. For men, we find that by age 60, 50–55% of these men retire, but the rate at earlier ages is greatest for those who had an early mental health condition. The difference is starkest at age 50. The pattern for women is similar, although the rate of retirement is higher than for men at every age, and especially at ages 51–60.
Links to Later Health
The pattern of descriptive results suggests that, for all three health outcomes (CVDs, gaps, and retirement), men and women with early mental health problems had health problems as adults at younger ages than those who had other early conditions or who were healthy as children. Those who reported poor/fair health as children had more health issues than others their age, except for those with mental health issues. In most cases, for these individuals, their health problems came later in life than those with early mental health conditions. For retirement, differences remain until about age 55.
Next, to learn what is gained by using multiple dimensions of health, we estimate Eq. (1) and compare results including our four childhood health measures to those in which we include only a summary health dummy. The dummy measure (at least one) equals 1 for those who had at least one of the conditions and 0 otherwise. Results are shown in Table 1, separately for men and women.
We first look at the three more direct later health outcomes: reported one or more significant periods of ill-health during adulthood (column 1), two or more significant periods of ill-health during adulthood (column 3), and retired because of poor health (column 5). For men, the regressions show that those with early mental health conditions are most likely to report all three outcomes, followed by those with poor/fair health, except in the case of retired because of poor health in which those with severe headaches as a child are about as likely as those with poor/fair early health to retire early. All these results are statistically significant at the 5% level. For women, the pattern also shows a higher probability for all three of these later health outcomes among those who had early mental health conditions, followed by those who had early poor/fair health. Thus, all of these later health outcomes show a far larger link between early mental health problems and later health problems than do any of the three other early health measures.
Next, we look at those who ever worked (column 7). For men, we find a strong negative influence of early mental health (a rate that is 3.8 percentage points lower). Among women, somewhat unexpectedly, an early physical health problem is statistically significantly and positively linked to having ever worked. We then look at the number of career gaps (column 9). For men, again, those with mental health issues have the largest number of gaps (an increase of more than three gaps). Men with early poor/fair health have nearly one additional gap, and those with severe headaches as a child have an average of about a half gap more than those with no such health conditions. Among women, those who reported poor/fair health and those who reported early mental health conditions have an increased number of gaps (about 1.6 and 1.3, respectively).
Our comparison to the use of a single indicator of health (at least one) shows that the single indicator is significantly related to all five of these indicators of long-term health and five (four for women) in the expected direction of poorer long-term health (columns 2, 4, 6, 8, and 10). The exception among women is for having ever worked, for which the unexpected direction is positive. But, in every case, we learn more about the dimensions of health that have the greatest influence on long-term health and the R2 or pseudo R2 is nearly always greater for the estimates including the four indicators of early health problems. Thus, we conclude that there are important insights gained using the four indicators of early health as compared with one summary variable.
Links to Additional Aspects of Well-being and Human Capital
As part of our exploration of the influence of early health on later health, we also address the influence of early health on human capital accumulation (i.e., education) and on two measures of well-being—if the person ever married and ever had a natural child (Table 2). For the education measure, we include years in full-time schooling (columns 1 and 2) and three indicators of levels as used in many European countries: low or ISCED 0–2, medium or ISCED 3–4, and high or ISCED 5–6 (columns 3–8). For both genders, we find a similar—and what we view as very interesting—pattern: those who reported early physical health conditions attain more schooling than others even after controlling for all other right-hand-side variables in Eq. (1). This pattern remains if we instead use the three categories of completed schooling: men and women with early physical health conditions are less likely to have low education and more likely to have medium—and especially high—education. In the case of women, we find as well that those with poor/fair health and—to a lesser extent—those with early severe headaches are more likely to have low education and less likely to have a medium or high education level. Overall, then, we consistently find that those children who had an early physical health condition were likely to accumulate more human capital, on average, in the form of years of schooling than those who had other conditions or who were healthy as children.
Turning to our estimates for measures of well-being, links between early health and having ever been married (columns 9 and 10) or having had a natural child (columns 11 and 12) are stronger for men than for women. Women with early mental health problems are slightly less likely to have married (−0.02) or to have had a natural child (−0.04). For men, these coefficients are about 2–3 times as large in absolute terms. Men with early poor/fair health are also less likely to have married (−0.03) and to have had a natural child (−0.04); the latter difference is similar for women. No significant differences are found for men and women with severe headaches during childhood or with early physical health problems. Thus, for our two measures of well-being, the links to early life—while relatively small—are larger for men, particularly for those who had early mental health problems.
Finally, regarding the use of a single indicator of early health problems (at least one), we find that the single indicator is significantly related to only the well-being measures (and to just one of the two for women) in the expected direction of poorer long-term well-being (columns 10 and 12). This further underscores the importance of distinguishing between multiple dimensions of childhood health.
Links to Life Course Health
We now turn to our regression analysis in which each of our life course health outcomes serves as a dependent variable and our four early health variables are the right-hand-side variables of primary interest (see Eq. (1) for details). As a test of robustness, we also include the additional mediators found to be significantly influenced by our early health measures, namely, years of full-time schooling, having ever been married, and having had a natural child (see Eq. (2)).
Our focus here is on indicators of later health by five-year intervals beginning with the link to serious health conditions as an adult. We focus on the initial incidence of CVDs, which are shown in Figure 4 (the full set of estimates is given in Tables A6 and A7). We calculated heteroskedasticity-robust standard errors and show the 95% confidence intervals in the figures. Among men, the results suggest that those who had an overall poor/fair health problem as a child are slightly more likely (about one percentage point) to experience a CVD early in adulthood, from ages 16 to 40. For those with childhood mental health problems, there is limited evidence of a significant association with CVDs over men's life course, except for a significant increase in the early 30s (about two percentage points). However, virtually none of the subsequent estimates of mental health are statistically significant (see Table A6). Those who had severe headaches as a child had an elevated risk in their early 30s, early 50s, and again in their early 70s. Importantly, the spike in CVDs at ages 46–50 shown in Figure 1 is driven by those with poor/fair health as a child. The actual coefficient on poor/fair health is close to three percentage points and significant at the 1% level. The differences across these childhood health coefficients are statistically significant at the 5% level at several ages, including most years in early adulthood and then again in middle (46–50) and late adulthood (results available upon request). These results suggest a much higher likelihood of CVDs among middle-aged men who had poor/fair health followed by those with severe headaches as a child.
In contrast to men, women with a mental condition by age 15 are about two percentage points more likely to experience an initial CVD at ages 16–20, 21–25, and 31–40 (see Figure 4). The first two age periods are significantly different from those who had early physical health conditions, severe headaches, or overall poor/fair health (results available upon request). Another interesting pattern—and one similar to that for men—is the higher probability of women who experienced early severe headaches having an initial CVD at ages 46–50 (with a coefficient of 0.025) continuing throughout their 50s. Overall, we find quite different patterns by gender, where men with early poor/fair health are at increased risk of a CVD especially at ages 46–50, while women with early mental health problems are at greater risk in their late teens, early 20s, and early 30s. Women with severe headaches are more likely to experience an initial CVD at ages 46–50, the age at which men with poor health appeared at greater risk. Both men and women with early severe headaches are at greater risk at ages 51–55.
In Figure A2, we show a similar analysis in which we extend the health conditions studied to include lung conditions. The pattern essentially replicates the analysis just described. Then in Figure A3, we extend the right-hand-side variables to include years of full-time schooling, having ever been married, and having ever had a natural child. The inclusion of these variables does not change the results for men or women. Thus, these two tests of robustness provide additional evidence of the pattern by which our four early health conditions are linked to the future pattern of CVDs, or important and common health conditions over the life course.20
Our next later health indicator that we measure is accumulated gaps in one's career, measured only for those who ever worked and shown in Figure 5 (the full set of estimates is given in Tables A8 for men and A9 for women). We measure this outcome only over ages tied most clearly to work: 21–25 to 61–64. Men show a very clear pattern in which those who had early mental health problems have a higher number of gaps at every age beginning at 21–25 and continuing through 61–64. On average, the additional number of gaps reaches a maximum of 2.6 for these men by the time they reach 51–55. The second group of men with higher numbers of gaps are those who had poor/fair health as a child, but this pattern does not really begin until the early 30s and only reaches nearly one gap by 61–64. The pattern for those with early severe headaches is similar to that of men with early poor health but smaller. These differences in estimates are statistically significant when comparing mental health to any of the other early health variables, and particularly so across all four of these early health conditions for all our included ages (results available upon request). When we add years of full-time schooling, having ever been married, and having ever had a child to the right-hand-side variables, the patterns remain (see Figure A4).
For women, we see a quite different pattern in which those with poor/fair health as a child have more career gaps beginning at age 21–25, eventually reaching 1.2 additional gaps by 61–64. Women who experienced early mental health problems show a slight increase in gaps (about 0.7–0.8), but they only begin at 46–50 and then fade after 51–55. Women with early severe headaches or those who experienced early physical problems appear largely unaffected in terms of career gaps. For women, the comparison of the coefficients on poor/fair health to severe headaches or physical health, and that of all four coefficients, finds significant differences at most ages from 26–30 onward (results available upon request). The pattern remains robust when we add schooling, marriage, and birth of a natural child as right-hand-side controls (Figure A4).
Our final later life course health outcome is age of retirement. As shown in Figure 6, the patterns for men and women are largely similar (the full set of estimates is given in Tables A10 for men and A11 for women). Men with early mental health problems are more likely than those without such health problems to retire early. By ages 36–40, the gap in probability of a man who had an early mental health problem is 0.05, which increases to about 0.08 at 46–50. In contrast, men with poor/fair health as a child are also more likely than those with at least good childhood health to retire early, but the difference is less than half that of men with a mental health condition. Their pattern increases as they age, reaching a maximum at ages 56–60. These observed differences are statistically significant when comparing men with the four early health conditions over ages 36–40 through 56–60 (results available upon request). The pattern remains when we add schooling, marriage, and birth of a natural child to the control variables (Figure A5).
Women who had an early mental health condition and—to a lesser extent—those with early poor/fair health are also more likely than those with no such health conditions to retire at almost every age studied. This pattern is replicated when we add schooling, marriage, and birth of a natural child as controls (Figure A5).
Discussion and Conclusion
In this article, we explored the influence of four distinct dimensions of early-life health on the onset of specific health conditions and permanent or temporary withdrawals from the labor market over the life course. Our main results for both sexes were robust to—among other checks—the inclusion of possible mediating factors (schooling, marriage, and birth of a natural child). Nevertheless, this study has limitations. Most notable among these, our early health variables are limited by the questions asked of the respondents and by possible recall bias. While such bias could result in an underreporting in some of the health conditions and possible reporting bias in childhood SRH (Smith 2009b), given the previous lack of evidence of justification bias, we would expect this to result in an attenuation bias. Other typical problems in this context—such as differential mortality, which causes individuals from disadvantaged social backgrounds to be less likely to reach the age of 50 (and be in our panel), or parental efforts to compensate for the negative effects of childhood health problems—may also attenuate our estimates of childhood health toward zero (Currie 2009; Palloni et al. 2009). As argued in the following, according to our findings, parental compensating behavior may have even reversed the (expected) negative impact of childhood physical health problems on later-life outcomes that are more prone to parental influence, such as education.
We see the primary contributions of this study as its focus on the links between distinct dimensions of childhood health and multiple life course health outcomes, including age of onset of serious cardiovascular conditions and job-related health outcomes. We found that the different dimensions of childhood health have distinctive links to later outcomes. We briefly compared our analysis using a single indicator of early-life health problems and found that we gained considerable insight and improved the fit of our equations when using the four distinct early health problems.
For men, early mental problems play a stronger role for all life course job-related health outcomes, but early poor or fair general health and severe headaches are more strongly linked to the onset of CVDs. In particular, the spike in CVDs found in the late 40s is driven by men with poor or fair general health as a child. For the included work-related measures of life course health, it is men with early mental health problems who do worst. They have the lowest probability of having ever worked (about four percentage points less likely), and when they work, they had the greatest number of gaps; the average number of gaps for them rose to two by age 50 and to 2.5 by age 64. Finally, men with early mental health conditions were also the most likely to retire early because of health (five percentage points more likely). These differences were often only about one third as large for those with early poor or fair general health and mostly insignificant for those with early physical health problems or severe headaches as a child.
For women, the links between the different dimensions of childhood health and the multiple life course outcomes are less clear-cut than for men. For instance, women who had a mental health condition early in life are more likely than others to experience an onset of a CVD in their early to middle adult years but not thereafter, and the initial and largest spike in the onset of a CVD at ages 46–50 is driven by women who had severe headaches by age 15. For life course job-tied measures, we first found a seemingly surprising outcome: early physical health problems are, in some instances, tied to better work-related outcomes. Women with early physical problems are more likely than other women to work (an increase of 1.9 percentage points); however, when they work, we find no evidence of an impact on gaps compared with women without physical health problems. Instead, women who reported poor/fair health and early mental health conditions have a greater number of gaps—nearly 1.2 for those with poor early health and 0.8 for those for early mental health issues by age 55, even though, for the latter, this increase fades out afterward. For our last measure of job-tied health—age at retirement—we again find no negative link to physical health. Instead, women who had an early mental health condition are also the most likely to retire at nearly every age studied, followed by those who had early poor/fair health and, to some extent, those with severe headaches; however, the differences are less than half that of women with an early mental health problem.
What might explain this seeming puzzle? We looked at intermediary outcomes, such as schooling, as well as whether an individual ever married or had a natural child. For both men and women, we find that those who reported early physical health conditions attain more schooling than others, after controlling for family SES, cohort FEs, childhood country FEs, and interactions between cohort FEs and childhood country FEs. This suggests that parents may invest more in education for children with physical health problems, perhaps with a view that they require additional schooling to succeed in the labor market. The links between early health and having ever married or having a natural child are—somewhat surprisingly—larger for men than for women. Women with early mental health problems are slightly less likely to ever marry or have a natural child (by two percentage points). For men, these coefficients are about three times as large in absolute terms. Men with early poor/fair health are also slightly less likely to marry. For both men and women, we find no such negative associations for those with early physical health problems.
The findings from our analyses on intermediary outcomes provide both a partial explanation for our life course results and evidence that it is important to distinguish early health problem by type (and gender) to predict (and counter) life cycle health consequences. More generally, our study shows that investigating the links between several dimensions of childhood health and multiple health-related life course outcomes enables a better understanding of how health inequalities originate and are shaped over the course of people's lives.
Acknowledgments
This research used data from SHARE Waves 1–7 (DOIs: 10.6103/SHARE.w1.710, 10.6103/SHARE.w2.710, 10.6103/SHARE.w3.710, 10.6103/SHARE.w4.710, 10.6103/SHARE.w5.710, 10.6103/SHARE.w6.710, and 10.6103/SHARE.w7.710), as well as data from the generated Job Episodes Panel (DOI: 10.6103/SHARE.jep.710). See Börsch-Supan et al. (2013) and Brugiavini et al. (2019) for methodological details. The SHARE data collection has been funded by the European Commission through FP5 (QLK6-CT-2001-00360), FP6 (SHARE-I3: RII-CT-2006-062193, COMPARE: CIT5-CT-2005-028857, SHARELIFE: CIT4-CT-2006-028812), FP7 (SHARE-PREP: GA 211909, SHARE-LEAP: GA 227822, SHARE M4: GA 261982), and Horizon 2020 (SHARE-DEV3: GA 676536, SERISS: GA 654221), and by DG Employment, Social Affairs & Inclusion. Additional funding from the German Ministry of Education and Research, the Max Planck Society for the Advancement of Science, the U.S. National Institute on Aging (U01_AG09740-13S2, P01_AG005842, P01_AG08291, P30_AG12815, R21_AG025169, Y1-AG-4553-01, IAG_BSR06-11, OGHA_04-064, HHSN271201300071C), and various national funding sources is gratefully acknowledged (see www.share-project.org). The authors wish to thank the editors and two anonymous referees of this journal for helpful comments.
Notes
Numbers from the youngest cohort covered in the Human Mortality Database (https://www.mortality.org/) born in the 1920s show that across the 11 countries included, an average of 77% of men and 83% of women survive to age 50. Life expectancy at birth for these cohorts is, however, fairly low (62.3 and 70.4 years, respectively). Only about 6% of our respondents were born in the 1920s or earlier, and nearly 75% were born in the 1940s or later. Given the sharp decline in infant mortality rates during this period, we would expect survival rates to age 50 to be substantially higher than 80% in most of our sample cohorts. For instance, more recent data from the United States show that between 2006 and 2016, the cumulative death rate per 100,000 persons aged 25–44 ranged from 136.7 to 159.0, with the highest rate in 2016 (National Center for Health Statistics 2018).
Several recent studies find causal effects of very specific exogenous early-life events on later-life outcomes. Two recent reviews by Currie and Almond (2011) and Almond et al. (2018) that focus on the experience of severe and mild health shocks in early childhood, respectively, are quite helpful, so we largely skip this literature.
Several studies using data that are similar to ours from the Survey of Health, Ageing and Retirement in Europe have also explored the associations between childhood health and later-life health (e.g., Arpino et al. 2018; Flores and Kalwij 2014; Howdon et al. 2019; Pakpahan et al. 2017). Besides focusing on health at a single age later in life, these studies typically use childhood health summary variables, such as childhood self-reported health, or summary indicators for the presence of chronic conditions during childhood. That is, they do not distinguish between different types of childhood health dimensions.
There is a broader literature that has explored the associations between childhood health and labor market outcomes over the life course (e.g., Flores et al. 2020; Goodman et al. 2011; Haas et al. 2011; Smith 2009b, Smith and Smith 2010), with some of them distinguishing between multiple dimensions of childhood health (Goodman et al. 2011; Smith and Smith 2010).
Hoffmann et al. (2018) also used data similar to ours from SHARE and found an association between childhood health (measured by SRH before age 16) and adult health; however, they did not distinguish between different dimensions of early-life health and focused on (just) two periods in adulthood (30–50 and 50–90). Moreover, because they used an older wave of SHARE, they included just 10 countries.
To the best of our knowledge, health registers covering a cohort’s full life span, which would avoid recall bias, are not available (yet). For instance, an article by Andersen (2021) used Danish health register data with very detailed health information on general practitioner visits and inpatient and outpatient hospital care, which are available for a maximum of 24 and 37 years, respectively. Hopefully in the future, the use of registry data will be possible to test and extend our findings.
Our analysis here includes what we call “job-tied” health measures, including gaps in work or retirement. Although these can occur for multiple reasons, our discussion assumes health-related reasons are a primary factor, but the reader may wish to skip these measures and focus on those that more directly measure later health.
SHARE uses refreshment samples to enhance the representativeness of the sample. Survey participation (i.e., panel retention) in SHARE is quite high. For instance, the average response rate in every wave from Wave 1 to Wave 7 ranged typically between 60% and 90% (see Bergmann et al. 2017).
The regular panel of Wave 7 includes question PH006 but not question PH009. This prevents us from using the eight newly added countries in this wave (Finland, Lithuania, Latvia, Slovakia, Romania, Bulgaria, Malta, and Cyprus) in our analysis.
The retrospective data on our job-tied health measures are taken from the third release of the retrospective SHARE Job Episodes Panel (JEP) (Brugiavini et al. 2019). JEP contains cleaned information on the start and end dates of all job spells that SHARELIFE respondents (from Wave 3 or 7) had during their working life, as well as information on their year of retirement and nonemployment spells. Accuracy might be an issue for self-recalled measures in SHARE, such as wages, especially when they were earned a long time ago. Our life course analyses focus on measures of health tied to employment, which are likely less subject to recall problems than earnings measures.
The list of countries with gender-specific sample sizes is shown in Table A1 of the online appendix.
In JEP, from the initial 91,743 respondents, we lose 21,485 respondents when retrieving health histories from Waves 1, 2, 4, 5, and 6 for all respondents in JEP. Approximately 70% of the respondents we lose are from the newly added eight SHARELIFE countries in Wave 7, which we cannot use in our analysis (see footnote 9). We also drop 885 respondents younger than 50.
Childhood SRH in SHARELIFE contains a sixth category “changing,” which includes 0.4% of the sample and is recorded only if the respondent spontaneously answers that their childhood health varied a great deal. We also include this group in our measure of poor/fair health.
While readers might be concerned that retrospective questions on health might lead to justification of poor labor market outcomes, Fletcher (2014) and Smith (2009b) did not find evidence of justification bias in retrospective reports of childhood ADHD and childhood SRH, respectively. In addition, Havari and Mazzonna (2015) did not find evidence of recall error in the childhood disease variables in SHARELIFE.
Routine testing of cholesterol did not occur before 1950 (Kuijpers 2021), so older persons in SHARE were unlikely to know they had high cholesterol prior to that date. Today, according to the Centers for Disease Control and Prevention (CDC), about one in five adolescents have unhealthy cholesterol levels and nearly 93 million adults 20 or older have high cholesterol. The CDC suggests cholesterol should be checked beginning with children and adolescents (https://www.cdc.gov/cholesterol/checked.htm).
For example, Hardy, Lawlor and Kuh, in a frequently referenced article, stated “coronary heart disease (CHD) and stroke are rare until middle age, but the pathophysiological process of atherosclerosis, which ultimately leads to cardiovascular disease (CVD), is initiated in early life” (Hardy et al. 2015:101). This life course perspective is fully consistent with our exploration.
The 18 childhood countries that we consider are Austria, Belgium/Luxembourg, Czechoslovakia/Hungary, Denmark, Estonia, France, West Germany, Greece, Italy, the Netherlands, Poland, Spain, Sweden, Switzerland, former SFRY (Socialist Federal Republic of Yugoslavia), other former USSR (Union of Soviet Socialist Republics), other EU (European Union), and other non-EU countries. The seven cohort groups are born before 1930, in 1930–1934, in 1935–1939, in 1940–1944, in 1945–1949, in 1950–1953, and after 1953.
The definitions of our childhood SES measures are given in Table A3. We attempt to include as many of the detailed childhood SES variables as were collected in both SHARELIFE interviews (Waves 3 and 7). For instance, we do not use the ISCO code with the occupation of the main breadwinner of the household when the respondent was age 10 because this variable has not yet been coded for approximately 20% of the SHARELIFE respondents in Wave 7 (at least in Release 7.1.0, which is the one we use here).
In a robustness check, we included in Eq. (1) 21 current country FEs and interactions between the survey year FEs and the 12 current countries that participated in both SHARELIFE waves, Waves 3 and 7 (we use one as a reference in each set and excluded here the dummy variable for being born in or moved to the current country of residence during childhood). We did this to account for any unobserved systematic differences over time between countries (e.g., in labor market policies and social insurance schemes) that may affect an individual’s health or labor market outcomes at later ages. Reassuringly, the estimates of our childhood health variables remained largely unchanged (results available upon request).
Because there was no routine testing of cholesterol before 1950, so that older persons in SHARE were unlikely to know they had high cholesterol prior to that date, we also excluded cholesterol from the set of CVDs. Doing so leaves our main results mostly unchanged (results available upon request).