Friends’ childhood adversity and long-term implications for substance misuse: a prospective Swedish cohort study
Abstract
Background and aims
Although an individual's childhood adversity is predictive of later substance misuse, the effect of adversity within an individual's friendship network has not been established. The current study aims to estimate the strength of the association between exposure to childhood adversity among individuals’ friends at the onset of adolescence, relative to individuals’ own exposure to childhood adversity, and hospitalization for substance misuse between young adulthood and retirement.
Design
Prospective cohort study.
Setting
Stockholm, Sweden.
Participants
Individuals born in 1953, living in Stockholm in 1963, and who nominated three best friends in the 6th grade school class (n = 7180; females = 3709, males = 3471), followed to 2016.
Measurements
The outcome was hospitalization with a main or secondary diagnosis attributed to substance misuse, reflected in Swedish inpatient records (ages 19–63 years). Five indicators of childhood adversity (ages 0–12 years) were operationalized into composite measures for individuals and their friends, respectively. Friendships were identified using sociometric data collected in the school class setting (age 13 years).
Findings
Individuals’ own childhood adversity does not predict childhood adversity among friends (P > 0.05). Childhood adversity among friends is independently associated with an increased risk of an individual's later substance misuse [hazard ratio (HR) = 1.17, 95% confidence interval (CI) = 1.09–1.24], independently of an individual's own childhood adversity (HR = 1.47, 95% CI = 1.34–1.61). However, childhood adversity among friends does not moderate the association between individuals’ own childhood adversity and later substance misuse.
Conclusions
Within a birth cohort of individuals born in 1950s Stockholm, Sweden, childhood adversity among an individual's friends appears to predict the individual's substance misuse in later life independently of an individual's own exposure to childhood adversity.
Introduction
Social relationships are complex determinants of health behavior that continuously evolve during the life-course [1-3]. These relationships are embedded in larger social, environmental and economic contexts, which also change over time to cumulatively influence health [2, 4]. During childhood, families are the primary source of socialization and influence on children's development and health behavior [5, 6]. Adverse family circumstances, or childhood adversity, is an established predictor of substance misuse, i.e. misuse of alcohol and illicit drugs, in later life [7-11]. Many indicators of childhood adversity have been identified as risk factors for substance misuse, including economic hardship [12-14]; household instability [9, 15, 16]; persistent [17] or unexpected [18] trauma; and parental alcohol and drug behavior, psychiatric disorder and criminality [7, 19-21].
Life-course approaches often focus exclusively on these family-related conditions as long-term predictors of behavioral health outcomes [8, 17]; however, the transition from childhood to adolescence is marked by an increasingly important social context within the school environment [2, 22]. The school class is a largely compulsory social setting in which adolescents spend a substantial proportion of their time, allowing strong relationships with peers to emerge and develop [23]. Adolescents must uniquely contend with the duality of the home and school contexts: the family decreases as the primary source of environmental influence [6] as adolescents’ autonomy grows and peers become increasingly salient sources of social interaction and support [24-26].
Exposure to adverse family circumstances may generate differences in health outcomes for both individuals and those with whom they have social relationships. Friendships with peers in adolescence may have the potential to buffer the impact of adverse circumstances in one's family environment or may introduce exposures not previously experienced in the individual's own social paradigm [27]. As such, childhood adversity among friends at the onset of adolescence may independently predict individuals’ substance misuse in later life, or it may interact with individuals’ own experiences of childhood adversity to strengthen the existing association.
- Does an individual's own childhood adversity predict childhood adversity among adolescent friends?
- Is childhood adversity among friends associated with later substance misuse?
- Does childhood adversity among friends moderate the association between an individual's own childhood adversity and later substance misuse?
Methods
The study's research questions and analytical plan were not pre-registered on a publicly available platform; all results should be considered exploratory.
Data
The data were derived from the Stockholm Metropolitan Study (SMS), a prospective cohort study comprised of children born in 1953 andliving in the Stockholm metropolitan area in 1963. The study was updated with register and survey data until 1986, when the SMS was de-identified [28, 29], after which the key that allowed additional follow-ups was deleted by Statistics Sweden in 2017. The SMS was probability matched to an updated data register (RELINK53), resulting in the Stockholm Birth Cohort Multigenerational Study (SBC Multigen, n = 14 608) [30]. The Stockholm Regional Ethical Review Board granted ethical permission to create the RELINK53 data register and probability match it to the SMS (reg. no. 2017/684–32); additional written informed consent was not required due to the anonymized nature of the data material.
Variables
Substance misuse-related hospitalization
The outcome captures the first hospitalization with a main or secondary diagnosis attributed to acute or chronic alcohol or drug use, reflected in inpatient care records from the National Patient Discharge Register collected by the Swedish National Board of Health and Welfare (1972–2016, ages 19–63 years) [31]. The measure includes diagnostic codes from the 8th, 9th and 10th versions of the International Classification of Diseases [32].
Measure of childhood adversity
Five indicators of childhood adversity were selected based on the aforementioned associations with substance misuse and available records in Swedish register data (ages 0-12 years): parental alcohol misuse, parental psychiatric disorder, receiving social assistance, residing in a single-parent household and father's criminality (data regarding the mother's criminality were not available).
Parental alcohol misuse includes any record indicating an alcohol-related offense, irrespective of whether the parent was subjected to institutional treatment or action by the temperance committee (temperance boards were established in Sweden in 1916 and contain records for individuals registered for alcohol-related offenses [33]). Parental psychiatric disorder reflects records of mental illness, psychiatric treatment or suicide and receiving social assistance indicates that the cohort member's family received financial benefits for at least 5 years, regardless of amount. Data for these indicators were derived from the Social Register from 1953 to 1965. Residing in a single-parent household was measured in the 1964 Register of Population and Income and denotes that the parent with whom the child resided was living alone, and father's criminality specifies that the father had at least one record in the National Crime Register from 1953 to 1959, inclusive of sentence [i.e. conditional (probation), unconditional (imprisonment) or exempt from punishment].
As the five indicators were derived from different registers, independent associations between each indicator and substance misuse were first estimated using Cox proportional hazards models. All associations were statistically significant, warranting inclusion in the final measure (Supporting information, Table S1). As childhood adversity indicators are typically correlated, the variance inflation factor (VIF) was used to test for multi-collinearity. VIFs ranged from 1.09 to1.29 (lower bound = 1), demonstrating low inflation due to linear dependence on other indicators. As such, all indicators were indexed: a count of 1 was issued for each indicator of childhood adversity with which an individual presented. The counts were then summed to create a childhood adversity composite score (CA score) with a possible value between 0 and 5.
Measure of childhood adversity among friends
The CA scores for the three friends nominated by each individual were summed to create a childhood adversity among friends composite score (CAAF score) with a possible value between 0 and 15. Friendships were identified through a sociometric test administered at age 13 during the School Study, conducted in 1966 under the SMS, which included all 6th-grade school classes in Stockholm (omitting classes comprised of children with learning disabilities). Individuals were instructed to nominate three members of their school class as friends using the question: ‘Who are your three best friends in class?'. Friendships are defined as outgoing nominations, whether symmetrical or asymmetrical, in the analysis.
Descriptive statistics for the study variables are presented in Table 1.
| n | % | |
|---|---|---|
| Substance misuse-related hospitalization | ||
| Neither alcohol nor drugs | 6704 | 93.4 |
| Only alcohol | 263 | 3.7 |
| Only drugs | 109 | 1.5 |
| Both alcohol and drugs | 104 | 1.5 |
| CA score (continuous) | Min = 0, max = 5, mean = 0.2 | |
| CA score (categorical) | ||
| 0 indicators | 5988 | 83.4 |
| 1 indicator | 787 | 11.0 |
| 2 indicators | 264 | 3.7 |
| 3+ indicators | 141 | 2.0 |
| CA indicatorsa | ||
| Parental alcohol misuse | 317 | 4.4 |
| Parental psychiatric disorder | 262 | 3.6 |
| Receiving social assistance | 1085 | 15.1 |
| Residing in a single-parent household | 634 | 8.8 |
| Father's criminality | 227 | 3.2 |
| CAAF score (continuous) | Min = 0, max = 11, mean = 0.7 | |
| CAAF score (categorical) | ||
| 0 indicators | 4291 | 59.8 |
| 1 indicator | 1629 | 22.7 |
| 2 indicators | 693 | 9.7 |
| 3 indicators | 308 | 4.3 |
| 4+ indicators | 259 | 3.6 |
| Sex | ||
| Female | 3709 | 51.7 |
| Male | 3471 | 48.3 |
- CA = childhood adversity;
- CA score = childhood adversity composite score;
- CAAF score = childhood adversity among friends composite score.
- a Prevalence of individual CA indicators.
Study population
Several exclusion criteria were applied to create the study population. Individuals who did not participate in the School Study (n = 2224) due to absence from school or not being in 6th grade were excluded, as were the 27 individuals who died prior to the follow-up period. As small school class size may influence the distribution of potential nominations, individuals in school classes of fewer than 10 students (n = 260) were removed [34]. Finally, to employ consistent metrics regarding childhood adversity among friends, individuals who nominated fewer than three best friends in the school class (n = 4917) were not included, reducing the final sample to 7180.
Data analysis
Regarding the first research aim, the association between individuals’ own childhood adversity and childhood adversity among friends was estimated using negative binomial regression. The CA and CAAF scores were included as the independent and dependent variables, respectively; results were presented as beta (β) coefficients and 95% confidence intervals (CIs). Due to the multi-level structure of the sociometric data, models 1–2 included school class-level random effects to account for any within-cluster non-proportionality of adversity or other unobserved characteristics; model 2 was also adjusted for sex.
The second and third research aims were examined by means of Cox proportional hazard models with shared frailty to account for any within-cluster homogeneity [35, 36]. The analyses estimate time to the first substance misuse-related hospitalization event, producing hazard ratios (HRs) and 95% CIs. All individuals entered the follow-up period on 1 January 1972 and were right-censored at time of death or at the end of follow-up (31 December 2016). Cox regression models were first used to estimate the associations between substance misuse-related hospitalization and the CA score (model 1) and the CAAF score (model 2); both scores were included as linear terms. Model 3 was mutually adjusted for both scores, whereas model 4 was further adjusted for sex. To estimate the moderating effect of the CAAF score on the association between the CA score and substance misuse-related hospitalization, an interaction term was included in model 5. This term was multiplicative, representing the product of both scores, creating a possible value between 0 and 75. All models included school class-level random effects to account for any unobserved homogeneity occurring within classrooms.
As considerable research has found a strong, graded relationship between exposure to multiple indicators of childhood adversity and risk for later substance misuse [7, 19, 37-39], a second set of hazard models with shared frailty were conducted in which each score was included as a categorical term to estimate the strength of the graded associations with substance misuse-related hospitalization. The CA score was recoded into four groups (‘0 indicators’, ‘1 indicator’, ‘2 indicators’ and ‘3+ indicators’), whereas the CAAF score was recoded into five groups (‘0 indicators’, ‘1 indicator’, ‘2 indicators’, ‘3 indicators’ and ‘4+ indicators’). The analytical procedure was nearly identical to the first set of hazard models; however, an interaction term was not included.
We tested the proportionality assumption by first plotting Kaplan–Meier curves for both independent variables and sex (figures not shown) and then conducting tests of proportionality by modeling Schoenfeld residuals. The global test for the final model indicates that the proportionality assumption holds, although there was some indication that the CA score violates the assumption. To address this, a sensitivity analysis was conducted in which the follow-up time was restricted to age 50 years (results not shown). These models showed similar, but generally stronger, associations between the CA and CAAF scores and substance misuse-related hospitalization. Another sensitivity analysis was conducted using a sample restricted to school classes where it was mathematically possible for an individual to nominate three friends with the same experience of (no) childhood adversity as her or himself (n = 5129). As the sensitivity analysis yielded nearly identical results (not shown), this restriction was not implemented.
Sex-separate analyses were performed early in this research. As no interaction effects were evident between sex and either independent variable, final analyses were adjusted for, rather than stratified by, sex. Similarly, separate analyses were initially conducted for alcohol- versus drug-related hospitalization, but as they yielded similar results alcohol and drug misuse were analyzed as a joint substance misuse outcome.
All statistical analysis was conducted using Stata version 15 [40].
Results
Table 2 presents results from negative binomial regression analyses estimating whether individuals’ own childhood adversity predicts childhood adversity among friends. The multi-level regression analysis showed a statistically non-significant decrease in the expected log count of indicators of childhood adversity among friends for each increase in individuals’ own adversity indicators (model 1); the results were unchanged after adjustment for sex (model 2). Descriptive statistics further detailing the relationship between the CA and CAAF scores can be found in the Supporting information, Table S2.
| CAAF score (continuous) | ||||
|---|---|---|---|---|
| Model 1a | Model 2b | |||
| β | 95% CI | β | 95% CI | |
| CA score (continuous) | −0.02 | −0.07−0.02 | −0.02 | −0.07−0.02 |
- CA score = childhood adversity composite score;
- CAAF score = childhood adversity among friends composite score;
- CI = confidence interval.
- a Model adjusted for school class-level random effects;
- b model 1 + adjusted for sex.
Measures of associations between substance misuse-related hospitalization and the CA and CAAF scores are shown in Table 3. For individuals, every unit increase in the CA score increased the hazard for substance misuse-related hospitalization by 50% (model 1). For every increase in the CAAF score, the hazard increased by nearly 20% (model 2). Both associations were statistically significant. Model 3, which was mutually adjusted for both scores, shows that the hazard for substance misuse-related hospitalization decreased marginally but remained statistically significant. These associations also held after the model was adjusted for sex (model 4). Results from model 5 suggested that the CAAF score is independently associated with later substance misuse-related hospitalization but does not moderate the association between the CA score and hospitalization; the interaction between the two scores was not statistically significant.
| Substance misuse-related hospitalization | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Model 1a | Model 2a | Model 3b | Model 4c | Model 5d | ||||||
| HR | 95% CI | HR | 95% CI | HR | 95% CI | HR | 95% CI | HR | 95% CI | |
| CA score (continuous) | 1.50*** | 1.33–1.64 | 1.46*** | 1.33–1.60 | 1.47*** | 1.34–1.61 | 1.49*** | 1.33–1.69 | ||
| CAAF score (continuous) | 1.19*** | 1.12–1.27 | 1.16*** | 1.09–1.24 | 1.17*** | 1.09–1.24 | 1.18* | 1.09–1.27 | ||
| Interaction | 0.99 | 0.94–1.04 | ||||||||
- CA score = childhood adversity composite score;
- CAAF score = childhood adversity among friends composite score;
- HR = hazard ratio;
- CI = confidence interval.
- Statistical significance:
- * P < 0.05;
- ** P < 0.01;
- *** P < 0.001.
- a Independent models with school class-level random effects;
- b mutually adjusted model with school class-level random effects;
- c model 3 + adjusted for sex;
- d model 4 + multiplicative interaction term (CA × CAAF) included.
To estimate the effects of co-occurring indicators of adversity on the hazard for substance misuse, the Cox regression analyses underlying Table 4 did not assume a linear association. For individuals, a graded relationship between the categorized measure of childhood adversity and substance misuse-related hospitalization was observed (model 1). Compared to those presenting with 0 indicators, a CA score of 1 had a 60% increased hazard for hospitalization, which increased with each additional indicator. Two indicators were associated with a nearly three times higher hazard, and three or more were associated with a nearly fourfold increased hazard.
| Substance misuse-related hospitalization | ||||||||
|---|---|---|---|---|---|---|---|---|
| Model 1a | Model 2a | Model 3b | Model 4c | |||||
| HR | 95% CI | HR | 95% CI | HR | 95% CI | HR | 95% CI | |
| CA score (categorical) | ||||||||
| 0 indicators (ref.) | 1.00 | 1.00 | 1.00 | |||||
| 1 indicator | 1.59*** | 1.23–2.06 | 1.56** | 1.20–2.02 | 1.57** | 1.21–2.03 | ||
| 2 indicators | 2.71*** | 1.94–3.80 | 2.56*** | 1.83–3.59 | 2.59*** | 1.85–3.63 | ||
| 3+ indicators | 3.69*** | 2.47–5.49 | 3.25*** | 2.17–4.87 | 3.23*** | 2.15–4.84 | ||
| CAAF score (categorical) | ||||||||
| 0 indicators (ref.) | 1.00 | 1.00 | 1.00 | |||||
| 1 indicator | 1.22 | 0.98–1.53 | 1.18 | 0.94–1.48 | 1.19 | 0.95–1.48 | ||
| 2 indicators | 1.16 | 0.85–1.59 | 1.09 | 0.79–1.49 | 1.12 | 0.81–1.53 | ||
| 3 indicators | 1.64* | 1.11–2.43 | 1.49* | 1.01–2.21 | 1.45 | 0.98–2.14 | ||
| 4+ indicators | 2.66*** | 1.88–3.77 | 2.27*** | 1.59–3.25 | 2.43*** | 1.70–3.47 | ||
- CA score = childhood adversity composite score;
- CAAF score = childhood adversity among friends composite score;
- HR = hazard ratio;
- CI = confidence interval.
- Statistical significance:
- * P < 0.05;
- ** P < 0.01;
- *** P < 0.001.
- a Independent models with school class-level random effects;
- b mutually adjusted model with school class-level random effects;
- c model 3 + adjusted for sex.
The hospitalization hazard was not significantly higher for individuals with a CAAF score of 1 or 2 (model 2). However, the hazard increased by 65% for individuals with a CAAF score of three, and nearly tripled among those for whom the CAAF score was four or higher. Both associations held in the mutually adjusted model (model 3), although with decreased hazard rates. In the fully adjusted model (model 4), which also controlled for sex, the graded association was still present for the individual, whereas the association among friends was significant only with a CAAF score of four or higher. As the linear analyses showed no evidence for interaction, a multiplicative interaction term was not included.
Discussion
This prospective cohort study estimated the strength of the associations between exposure to childhood adversity among individuals’ friends at the onset of adolescence, relative to individuals’ own exposure to childhood adversity, and hospitalization for substance misuse between young adulthood and retirement age. Childhood adversity among friends was independently associated with substance misuse, even after controlling for individuals’ own adversity. The categorical CA score estimated a graded relationship between childhood adversity and substance misuse, whereas a threshold effect was observed for the CAAF score; four or more indicators of adversity predicted an increased hazard for substance misuse. However, the two potential sources of adversity did not interact in a statistically significant way, nor did individuals’ childhood adversity predict adversity among best friends in the school class.
Shared adverse family circumstances do not appear to explain the independent association between childhood adversity among friends and individuals’ later substance misuse. At the onset of adolescence, individuals’ social environments expand and expose individuals to new social contexts and the family circumstances in which they originate through interactions with their peers. Adolescents’ increasing autonomy and susceptibility to external influences indicates that socio-demographic conditions may take on additional importance during this key developmental period [6]. Neighborhood effects have the potential to affect the health behaviors of individuals and their friends through shared access to social and economic resources, parenting norms and sources of social support [41]. Further, neighborhoods, and the schools located therein, dictate the peers to whom individuals have access [24]. However, prior studies using these data concluded that substantial demographic and socio-cultural heterogeneity existed within the Stockholm metropolitan area at the time [42, 43], indicating that neighborhood effects may not explain the association between friends’ childhood adversity and individuals’ later substance misuse.
Unobserved behavioral influence may be another explanation for this association. Studies show that peer influence is correlated with initiation or continuity of health risk behaviors [44-50], exposure to substance-using peers is associated with individuals’ onset or use of substances [51-55] and substance use initiated during adolescence has greater potential to become normative [2], with correlations between earlier onset and greater risk of later use [56, 57]. Not only is disentangling exposure to friends’ adverse family circumstances and friends’ behavior empirically challenging, but this study is also unable to estimate effects due to peers’ use of substances in adolescence.
Independent of friends’ childhood adversity, individuals' own adversity was strongly associated with later substance misuse. Considerable theoretical and empirical research has shown that the parental generation's living conditions tend to be transmitted to their offspring [58]. As such, heritable factors, particularly regarding substance misuse transmission, must also be considered. An additional sensitivity analysis was conducted to assess whether the association between substance misuse-related hospitalization and the CA score could be explained by biological factors, e.g. a genetic predisposition for alcoholism [59-61]. Here, the exclusion of individuals whose parents had an alcohol-related offense did not greatly reduce the association between exposure to childhood adversity and substance misuse-related hospitalization; the hazard ratios remained nearly identical to the results presented in Table 3 (results not shown). Therefore, exposure to any of these adverse conditions within the family environment, which tend to be strongly inter-related, is associated with a higher risk of substance misuse in adulthood.
Social relationships concurrently influence health risk behavior throughout the life-course; adolescence, being a unique developmental period, is marked by the concurrent and potentially contradictory influences of parents and peers. Growing evidence shows that early life experiences have long-lasting health effects, but the influence of friends’ family backgrounds and behaviors also have the potential to affect health risk behaviors well into adulthood.
Strengths and limitations
Regarding strengths, the unique data set used combines 44 years of register-based follow-up with original survey data, including the sociometric data collected during the School Study, which made it possible to identify individuals’ friends at age 13, and the extensive inpatient data through which substance misuse-related hospitalization could be identified. The childhood adversity measures collected via Swedish registers were comprised of parental behavior, family structure and socio-economic indicators.
This study also has several limitations. Sociometric data were only recorded for individuals in the 6th grade who were present in class on the day of the School Study. Those who were absent from class were still able to be nominated, and friendships included in the study could be symmetrical or asymmetrical. The observed association between friends’ adversity and later substance misuse could therefore be over- or underestimated, compared to studies where only symmetrical relationships are included [62]. Further, the analysis was restricted to those who nominated three best friends in the school class, reducing the study population to just under half the cohort. The reduced sample size also decreased the overall prevalence of childhood adversity, limiting comparisons to studies with similar exposures.
This study is unable to account for peer influence regarding the onset or use of substances in adolescence [63], nor does it include other delinquent behaviors among friends that could predict later substance misuse [63-65]. Further, it cannot adjust for the potential influence of other relationships taking place outside the school class. For example, relationships with older friends and siblings have been shown to increase adolescent substance misuse behavior [66]. By limiting nominations to friends in the 6th grade, excluding the potential influence of friendships with older individuals and not assessing other indications of problematic behavior, associations with later life substance misuse may be underestimated.
Although the use of population registers is a key strength of this study, the data are not without drawbacks. Identification of substance misuse events was limited to cases recorded in the National Patient Discharge Register, i.e. only cases severe enough to require inpatient care or cases where the individual sought treatment were included. Therefore, incidence of substance misuse may be underestimated. Substance misuse presents in a myriad of harmful patterns regarding the frequency, duration and severity of episodes [67]. As this study did not meaningfully measure these differences, all inpatient events related to acute or chronic use were included to encapsulate a more comprehensive measure of substance misuse-related hospitalization. Similarly, three indicators of childhood adversity were drawn from the Social Register. Those who received social assistance had a lower socio-economic position; therefore, psychiatric disorders and alcohol misuse may have been more often identified within this population compared with families with resources to cope without societal assistance, enabling them to bypass register records [68].
Finally, the generalizability of the findings is potentially affected by unmeasured period or cohort effects. The 1953 cohort grew up during the expansion of the Swedish welfare state, and some have even referred to them as a ‘golden generation’ [29]. However, childhood adversity existed within this cohort, although some indicators (e.g. residing in a single-parent household) may yield different effects in more contemporaneous cohorts. Moreover, it is likely that life-course patterns of alcohol and drug misuse behaviors have changed over time, as has the Swedish health-care system. However, there is nothing to suggest that childhood adversity, regardless of measurement, has become less important, nor that the role of social relationships for behavioral development has changed over time.
Implications for research
Childhood adversity among friends appears to predict substance misuse in later life relative to individuals’ own exposure to childhood adversity within this birth cohort of individuals born in 1950s Stockholm, Sweden. The social environments in which both families and peers operate are multi-faceted determinants of health risk behavior that concurrently and continuously evolve throughout the life-course. Additional high-quality longitudinal studies that estimate the effects of peers’ family conditions and substance use behaviors in adolescence on individuals’ later substance misuse are warranted to extend this research and ascertain the causal mechanisms behind this association.
Declaration of interests
This study was financially supported by the Swedish Research Council for Health, Working Life and Welfare (Grant Nos. 2016–07148 and 2019-00058). The funder had no role in the design of the study; collection, analyses, or interpretation of data; writing of the manuscript, or decision to publish the results.
Acknowledgements
We are very grateful to Anders Ledberg and Viveca Östberg for their valuable input to the manuscript. We would also like to thank the steering committee for the Stockholm Birth Cohort Multigenerational Study (SBC Multigen). This study was performed within the research programs Reproduction of Inequality through Linked Lives (RELINK) and Risk and Resilience: Pathways to (Ill) Health among Men and Women with Experiences of Childhood Adversity (RISE), both financially supported by the Swedish Research Council for Health, Working Life and Welfare (Forte), grant numbers 2016–07148 and 2019-00058.
Author Contributions
Lauren Bishop: Conceptualization; formal analysis; methodology. Ylva B Almquist: Conceptualization; formal analysis; methodology.
