Introduction
The second generation (G2; children of migrants) are most commonly defined as people born in a country to at least one foreign-born parent.1 This group represents one of the fastest-growing and most diverse young populations in migrant-receiving countries in the world today, due to the establishment and continuation of decades of international migration.2 In the European Union (EU) in 2021, 7% of the population aged 15–74 years were G2. These proportions were considerably higher in late adolescence and young adulthood (15–29 years; 11%).3 Sweden, the context of this study, has some of the highest proportions of G2 in all of the EU at 12% (15–74 years) and 19% (15–29 years).3 A recent review of mortality among the G2 in Europe describes an elevated risk of death in young to middle adulthood (between ages 15 and 64 years) compared with people born in a country to two parents born in that same country (a group henceforth referred to as the majority population).4 G2 people—especially men—with parent(s) born outside of Europe, particularly the Middle East, Northern Africa and sub-Saharan Africa, have elevated mortality risks between these ages compared with the majority population.4
Previous research has investigated this higher adult mortality risk of the G2 by focusing on its association with inequalities in their adult socioeconomic background (SEB)—particularly educational and labour market outcomes.5–9 These studies show that adult socioeconomic inequality among the G2 often explains a substantial part, but not all, of their higher mortality risk.4 However, this only explores a part of the lives of the G2. Less research has focused on the association of the higher adult mortality risks of the G2 with their childhood SEB,10 a critical, formative part of the life course. The ‘long arm’ of childhood SEB affects mortality well into adult life, independent of adult SEB.11–14 Migrants (the parents of the G2) typically experience downward social mobility after arriving in the host country. They are exposed to varying degrees (according to their country of birth) of socioeconomic inequality relative to the majority population.15–17 It is these conditions that directly inform the childhood SEB of the G2.
Additionally, there is evidence of upward intergenerational mobility among the G2 in Europe, particularly among children of migrants who have lower SEB origins.15–17 This means that parental SEB (and thus childhood SEB) might differ substantially from the adult SEB of those upwardly mobile G2.11 Thus, only focusing on associations between adult SEB and the higher mortality risks of the G2 might fail to capture the childhood socioeconomic disadvantage that the G2 were exposed to. Childhood SEB has an influence on mortality beyond its direct influence on adult SEB. This indicates that childhood SEB and adult SEB could have independent impacts on mortality among the G2 in addition to an impact through their effect on one another.12–14 We thus conceptualise that childhood SEB (captured by parental SEB) affects the mortality of the G2, both through its effect on adult SEB and through other mechanisms. This is in addition to any independent effects of adult SEB—which are not the focus of this analysis. We follow prior research in arguing that we expect these interrelationships to be contingent on a number of factors, including demographic factors such as sex, birth cohort and the parental country of birth, as well as the context in which our study takes place.
The aim of this article is to investigate the association of childhood SEB with variation in the young adult mortality risks of the G2 compared with the majority population in Sweden, with a focus on specific parental origins and causes of death.
Data & Methods
Data
We used the collection of Swedish register data Ageing Well at Stockholm University. The collection covers longitudinal, individual-level data from a range of administrative sources. We used the total population register, migration register, multigenerational register, cause-of-death register and the Longitudinal Integrated Database for Health Insurance and Labour Market Studies (LISA). Information was linked between the same people across the different register sources using a unique individual identifier. Ageing Well was generated and pseudo-anonymised for research purposes.
To be included in the study, subjects had to be (a) born in Sweden, (b) turn age 16 years between 1 January 1992 and 31 December 2016 and (c) have at least one living parent on turning 16 years old (to derive the parental information). A delayed entry design was imposed; entry into the study was conditional on survival to 16 years old.
The outcomes were all-cause mortality and mortality from specific causes. The age at death (derived from the exact date of death in the cause-of-death register) was used to identify whether someone had died. The cause of death was derived from the underlying cause-of-death variable in the cause-of-death register and categorised into two variables with differing detail levels. Variable one coded the cause of death into natural and external causes. Variable two coded the cause of death into 12 cause groups: cancer, circulatory diseases, respiratory diseases, endocrine, nutritional and metabolic diseases, diseases of the nervous system, other diseases and medical conditions, suicides, substance misuse (including alcohol), traffic accidents, other accidents and injuries, assault and unknown causes of mortality. The causes were coded using a combination of International Classification of Diseases 9th Revision (ICD-9) (1992–1996) and ICD-10 (1997—) codes. Online supplemental table S2 shows exactly how we categorised the causes of death.
Exposure was G2 status. People were classed as G2 if they were born in Sweden to at least one parent born abroad. G2 status was generated by linking parents to children (via the multigenerational register and the matching of parental and child identifications) and using information on individual and parental country of birth (derived from the total population register). The reference population in the models was always the majority population. We examined the association between childhood SEB and mortality at the generational level (ie, all G2 combined) and according to more granular parental origins. Specifically, those with at least one parent born in Finland, the other Nordic countries, other (non-Nordic) EU/European Economic Area (EEA) countries, former Yugoslavia, the rest of Europe, South America, sub-Saharan Africa, Northern Africa, Iran and Iraq, other Middle East and Northern Africa (MENA) and Asia.
Predictor variables included sex, birth cohort, family and living situation, highest level of parental education, parental disposable income and parental unemployment status. Family and living situation and the highest level of parental education were measured in the year the child turned 16 years old. Parental disposable income and unemployment status were measured in a 3-year period in which the child was 14, 15 and 16 years of age to counteract some of the volatility linked with single-year measures of these two predictors.
Sex was derived from the total population register and grouped into ‘male’ and ‘female’. Birth cohort was from the date of birth in the same register for years 1976 to 2000. Family situation was derived from LISA and grouped into ‘living at home with married parents’, ‘living at home with cohabiting parents’, ‘living at home with a single parent’ and ‘lives alone’.
The highest level of parental education (in LISA) was coded according to the International Standard Classification of Education into ‘primary’, ‘secondary’ and ‘post-secondary’ levels.
Parental disposable income quintile (from LISA) was based on a variable that records disposable income for a calendar year. To begin with, we estimated the average of both parents’ annual disposable incomes for each of the 3 years in which the child was 14, 15 and 16 years old. For each year, we then ranked the averages within centiles, generating a place in a distribution ranging from 0 to 100. The average of the centiles across the 3 years was taken and grouped into ‘lowest’, ‘lower’, ‘medium’, ‘higher’ and ‘highest’ quintiles.
Parental unemployment status (from LISA) was based on a variable that records the number of days of unemployment within a calendar year. For a given year, if a parent recorded 90 days or more of unemployment, they were coded as ‘unemployed’. The 90-day cut-off was adopted from a study examining unemployment persistence among the G2 in Sweden, which showed a good degree of consistency between this cut-off and official unemployment statistics from the Swedish Labour Market Survey.18 The total number of years unemployed in the 3-year period before the child turns age 16 years was then summed to a value between 0 (ie, 3 years in which a parent records less than 90 days of unemployment) and 3 (ie, 3 years in which a parent has equal to or more than 90 days of unemployment). An average of the values of the two parents was then taken and organised into unemployed for ‘0 years’, ‘1 year’, ‘2 years’ and ‘3 years’.
We note that there were 69 225 (or 2.86%) cases where a parent died before the child turned 16 years old. In these cases, we only considered the living parent’s education level and living parent’s unaveraged values for disposable income and unemployment status.
Methods
We implemented a multistate competing risk approach following the logic outlined in Putter et al.19 We fitted flexible parametric survival models, assuming a proportional hazard approach using ‘stpm2’ in Stata 18.20 21 The HRs were reported alongside 95% CI. In light of several of the well-known shortcomings associated with hazard ratios, several regression-standardised (ie, confounder-adjusted) measures of risk were also reported in online supplemental materials.
People joined the risk set on turning 16 years old during the period 1992–2016. They exited the risk set if they died, emigrated or were alive and residing in Sweden on 31 December 2016. The oldest age reached was 40 years old. Whether or not someone had emigrated was determined from registered emigration in the migration register and a residence indicator from the total population register updated at the end of each calendar year.
We conducted a complete case analysis—an analysis that includes only individuals for which there is no missing information in the exposure or any of the predictor variables. From a potentially eligible starting population of 2 368 459 individuals, 2 345 833 individuals (~99%) were retained for our final statistical analysis. 22 626 individuals were removed due to missing data in their family situation (17 763; 0.7%), parental educational level (1732; 0.1%), parental disposable income (3959; 0.2%) and parental unemployment status (3995; 0.2%). Note that these numbers do not sum to 22 626 due to an overlap in missing information across the variables for the same people in the data set.
Analytical strategy
We fitted two models, a ‘minimally adjusted’ model and a ‘fully adjusted’ model. The minimally adjusted model controlled for birth cohort (in single years, with 1976 as the reference group in the models), sex (with female as the reference group) and G2 status (with the majority population as the reference group). The fully adjusted model controlled for birth cohort, sex and G2 status and additionally controlled for the highest level of parental education (with tertiary level of education as the reference group), parental disposable income (with the highest income quintile as the reference group) and years of parental unemployment (with zero years of unemployment as the reference group). Figure 1 presents the minimally adjusted and fully adjusted HRs of all-cause and cause-specific mortality at the generational level using the 12-category cause-of-death variable. Figure 2 displays the minimally adjusted and fully adjusted HRs of all-cause and cause-specific mortality for the specific parental birth regions using the three category cause-of-death variable. When describing the results, we use ‘HR’ to refer to the minimally adjusted model and ‘aHR’ to refer to the fully adjusted model.
HRs of all-cause mortality and mortality from specific causes of death among the second generation aged 16–40 years in Sweden, 1992–2016, before and after adjusting for childhood socioeconomic background. Notes: Estimates in bold have CIs that do not overlap with 1. The red line represents the majority population reference category. ‘Min. Adj.’ refers to the minimally adjusted model, and ‘Ful. Adj.’ refers to the fully adjusted model. ‘HR’ refers to hazard ratio. Source: author calculations based on Swedish register data Ageing Well.
HRs of all-cause, natural and external mortality among the second generation aged 16–40 years in Sweden, 1992–2016, before and after adjusting for childhood socioeconomic background. Notes: Estimates in bold have CIs that do not overlap with 1. The red line represents the majority population reference category. ‘Min. Adj.’ refers to the minimally adjusted model, and ‘Ful. Adj.’ refers to the fully adjusted model. ‘HR’ refers to hazard ratio. Source: author calculations based on Swedish register data Ageing Well.
We conducted a large number of additional analyses to supplement and contextualise our main results. Online supplemental table S1 offers descriptive information about the sample (population sizes, time at risk, crude and age-standardised mortality rates) for the parental origin variable by natural and external causes of death. Online supplemental table S2 shows similar information to online supplemental table S1 but at the generational level and for the 12-category cause-of-death variable. Online supplemental table S3 shows how the distribution of the predictors varies between the majority population and the G2. Online supplemental table S4 reports the HRs of the predictor variables from the fully adjusted, all-cause mortality model. Online supplemental table S5 investigates sex differences in the HRs at the generational level. Online supplemental table S6 reports results for a more restrictive definition of the G2, limiting exposure only to people born in Sweden to two parents born abroad. For the regression sensitivity analyses, HRs for online supplemental tables S5 and S6 have also been investigated in one common model (ie, via the use of interaction terms). This did not change the value of the estimates or confidence limits. Online supplemental figure S1 reports annual entries into the population at risk broken down by parental region of birth so as to understand more how the parental origin composition of the G2 population changes over entry year. Online supplemental figures S2–S5 display a series of regression-standardised survival probabilities, failure probabilities, mortality rates and cumulative incidence functions (CIFs) from the generational-level models to allow us to interpret our main findings in terms of absolute risk.
