Socioeconomic position and self-harm among adolescents: a population-based cohort study in Stockholm, Sweden

The study population consisted of 169,262 adolescents comprising of seven birth cohorts, born between 1988 and 1994, who lived in Stockholm County at the age of 13, withdrawn from SYC. The study period extended from 2001 to 2011, with each of the seven birth cohorts being followed for 5 years, from age 13 to 17. Adolescents with missing values on at least one of the explanatory variables or the outcome variable (n = 3300) were excluded and the final study population consisted of 165,932 adolescents.

First-time self-harm, from here-on referred to as self-harm, was the main outcome of the study and was ascertained through individual record linkage to national administrative registers and regional health care registers, covering all pathways of diagnosis and care related to self-harm, except private clinics. The registers were: (1) the VAL database, a Stockholm County register on public health care services which includes out-patient, in-patient and primary care, (2) the Cause of Death register and (3) Pastill, a clinical database covering all visits to child and adolescent psychiatry in Stockholm. Self-harm was defined according to the tenth revision of the World Health Organization (WHO) Classification of Diseases (ICD-10) (Intentional self-harm X60–X84) in the VAL database, Cause of Death register and Pastill. In Pastill, self-harm was additionally defined by a diagnosis of suicidal attempt and by self-harm as a contact reason. Only the first episode of self-harm during age 13–17 was used.

Severity of self-harm was defined based on the level of care rendered to individuals: those who received inpatient care for self-harm were considered as severe cases and those who received outpatient care for self-harm were considered as less severe cases. The most common reasons to be hospitalized for self-harm in Stockholm County is suspected or identified suicidal attempt. It is also more common among those hospitalized to have substance related disorders and, to some extent, anxiety disorders as underlying psychiatric diagnoses, whereas psychosis and bipolar disorders, neurodevelopment disorders as well as disruptive, impulse-control and conduct disorders were less common in both groups. Depressive disorders and anxiety disorders are the most common comorbid psychiatric diagnoses in self-harm both with and without hospitalization. Hospitalization requiring admission for at least one night was considered as inpatient care.

Socioeconomic position (SEP), the main exposure, was measured the year the adolescent turned 12. SEP was measured in two ways, parental education and household disposable income. Information on SEP was extracted from the longitudinal integration database for health insurance and labor market studies (LISA). Level of education was categorized into three categories based on number of years of completed education: up to 9 years (primary education), 10–12 years (secondary education) and >12 years (tertiary education). The highest educational achievement of either parent was used to define parental education. Household disposable income was categorized into quintiles, with consideration of year of income determination in addition to the actual income to ensure that approximately equivalent income groups were compared over time. The first and fifth quintiles represented the lowest and highest household income categories respectively.

Demographic factors—age, gender and parental country of birth—were assessed using information from the Total population register. Age was used as a continuous variable. Parental country of birth was categorized in three groups: Sweden if a single parent or both parents were born in Sweden, outside Sweden if a single parent or both parents were born outside of Sweden, and mixed if one parent was born in Sweden and the other outside Sweden.

Social and economic factors used in this study were number of parents in the household and receipt of welfare benefit. A household was regarded as having received welfare benefit if anyone in the household received benefit, once or several times, during the year the adolescent turned 12; the data was extracted from LISA. History of mental disorder of biological parent was defined when a biological parent was hospitalized for at least one night due to any mental disorder. The information was obtained from the National Hospital Discharge register from 1964 until the adolescent turns 13 years old.

The characteristics of the cohort were described using descriptive statistics. Incidence rates for self-harm were calculated per 100,000 person-years. Proportionality of the hazard assumption was checked using log minus log graph. Analyses were performed using Cox proportional hazard regression to assess the association between self-harm, SEP and other relevant covariates and to estimate hazard ratios (HR) with corresponding 95% confidence intervals (CIs). Time under risk was calculated using the entry date defined as the date the adolescent turned 13 years of age, and the exit date as the date of the first-time diagnosis of self-harm, date of death of any cause, date of moving out of Stockholm County or the end of follow-up, whichever came first.

Stratified analyses were performed by severity of self-harm, to assess the role of severity of the self-harm; and by gender to address gender differences. We considered receipt of welfare benefits, parental country of birth, number of parents in the household and mental disorder of biological parent as potential confounders/mediators. SAS version 9.3 was used for all statistical analyses.

This population-based study with a large cohort of adolescents yielded high power with long follow-up time and full coverage of events of self-harm from almost all pathways of diagnoses and care to self-harm in Stockholm County. Since the health care system as well as the composition of the population is similar between the big cities of Sweden, the results of the study can be generalized to the population of those big cities and other populations within a similar context. We believe that we eluded some of the limitations confronted by previous studies—specifically, recall bias and loss to follow-up which could have led to selection bias. The longitudinal nature of the study gave us an opportunity to make conclusions about causality. Inclusion of non-hospitalized (less severe) cases of self-harm in this study helps to address this rarely studied portion of the self-harming population and to make more comprehensive conclusions. The gap in the data caused by missing information about the parental education, household income and other covariates were few ranging between 1.3 and 2.0% (n = 3300). And there was no significant difference found in risks of self-harm because of these missing values. Using multiple variables to assess SEP, which measures different aspects of the concept, helped to give a broader perspective as underlying pathways are multifaceted and complex. Literature suggested that variables which measure SEP should not be used interchangeably as they measure different aspects of socioeconomic positions and refer into different causal mechanisms [54, 55].

One limitation in this study lies in the use of health care registers and limits our analyses to cases of self-harm for which care has been sought. Compared to other recent population-based survey studies, the figures for self-harm are lower in this study which indicate that many adolescents who self-harm do not seek treatment [56]. The tendency to seek care may differ depending on method used, which could explain part of the differences between boys and girls. High priority is given to equity in health in Sweden [57] and the target of the Swedish Health Care Act is equity in opportunity to use healthcare depending on need [58]. However, studies show that health-care utilization is not always strictly linked to health status and need, several factors can impact whether ill-health status leads into utilization of healthcare [57], and several studies have revealed disproportionately lower utilization of healthcare services by people with low SES and ethnic minorities [59, 60]. In Sweden, lower socioeconomic groups refrain to a larger extent from seeking medical care they need [43, 44] and increment in these trends has been observed [45]. Though this is a somewhat lesser problem with regard to children, since most medical services are free for children [61], lack of time may also play a role. Hence, the increased risk found among adolescents with low SEP is likely an underestimation. On the other hand, parents of adolescents with higher SEP may choose to visit private psychiatric clinics, whose data was not included in this analysis, which would lead to a slight overestimation of our results. It is important to examine whether the degree of underreporting is comparable across SEP categories.

Another concern in this study was a possible non-differential misclassification of parental SEP and other social characteristics which could have occurred due to two reasons. First, only one household was recognizable for adolescents who passed equivalent or different amount of time residing in the homes of separated parents, as children in Sweden are registered at a single address [62]. Second, it was not possible to determine a second parent if he or she was not biological or adoptive parent, as the information on the second parent when non-biological/adoptive was differential due to housing conditions, and housing conditions are related to one’s socioeconomic position. Both by recognizing only one of two households and by excluding the second parent when non-biological/adoptive, some adolescents may have been classified to a lower SEP than they should. Such misclassifications would lead to underestimation of the effect.