Participants and procedures
The present study was based on data from a cross-sectional survey completed by adolescents in the city of Oslo, Norway. All junior and senior high schools (N = 91) in the city were asked to join the study and 75 (82%) of these schools agreed to participate. There was a geographically even distribution of non-attending schools in the city. All pupils in grades 9, 10, and 11 in the study schools were invited to participate, and a strategy for including those who were not attending the particular day of the survey in a second distribution were conducted. The overall response rate was 92.7, increased from 87% on the main day of the survey. The survey was anonymous, hence a license from the Data Inspectorate to process personal sensitive data was not required. Permission from the Ministry of Research and Education, the local school authorities and the school boards were obtained. Study participation was based on informed passive parental consent. The net sample comprised 11,440 participants with a mean age of 15.4 years (range, 14–17 years; 51.2% girls). Pupils completed a comprehensive questionnaire at school during two school hours.
Measures
The study was designed to allow participants to report episodes of SA and NSSH separately. Self-harming behaviour and suicidal behaviour was assessed with two questions:
“Have you ever taken an overdose of pills or otherwise tried to harm yourself on purpose?” (“no”, “yes, once”, and “yes, more than once”), a question derived from the CASE study and has been used in several other studies [
13,
20,
25,
26]. Suicide attempt was assessed with the question:
“Have you ever tried to kill yourself?” (“no”, “yes, once”, and “yes, more than once”), which has been used in previous Norwegian school surveys of adolescents [
27]. The adolescents were divided into four groups based on their responses to the two questions: 1) no SH, 2) NSSH, 3) SA, and 4) SA + NSSH. We assumed that those who reported an overdose/SH but not SA had no suicidal intent and their SH would belong to the NSSH category, and that those who confirmed SA but not SH had no behaviour belonging to the NSSH category. This is based on the logical assumptions that answering two different questions on self-harm and confirming one, but denying the other, is a way of reporting intent. Previous studies have categorized SH by intent based on a similar methodological approach [
4,
25].
Contact with CAPS was assessed through the question “Have you ever been in contact with or received help from child and adolescent psychiatric outpatient services?”
Questions on sociodemographics and psychosocial problems were asked to the adolescents in Norwegian, and measures and questions are used in previous Norwegian studies [
7,
17,
25,
28].
Socio-demographic variables included information on gender, age, social class, and immigrant background
. The
social class variable was an adaptation of Eriksson and Goldthorpe’s social-class categorization (EGP classes) based on the parents’ professional achievement. For the purpose of the current study, the variable was dichotomized into high or low socio-economic status. Dichotomization of information is widely done, and based on findings that parents’ education level is of importance regarding contact with help-services [
29].
The adolescents were asked about their own and their parents’ country of birth which formed the basis for a dichotomous variable on immigrant background. Adolescents were categorized as having a non-Western immigrant background if the adolescent and/or both of the parents were born in Asia/Africa.
Current suicidal ideation was assessed with one question from the Hopkins Symptom Checklist (SCL-90) [
30]. This has been found to be a valid approach [
31]. Participants were asked whether, during the previous week, they had had thoughts about ending their life, using a scale ranging from 1 to 4 (“not at all”, “a little”, “rather often”, and “very often”). For statistical analysis, this variable was dichotomized into “none or a little” and “rather often or very often”.
Substance use variables comprised information about drinking to intoxication and illicit drug use in the 12 preceding months. Because the distribution of these variables was skewed, they were dichotomized into whether or not the respondent had drunk to intoxication, and used other illicit drugs.
Depressive symptoms were assessed with six items from the Hopkins Symptom Checklist (SCL-90) [
30], using the previous week as a reference period. The shortened version are found to be valid and used in other publications [
32] as also the Norwegian translated version [
17]. The items were rated on a scale ranging from 1 to 4, with total scores ranging from 6 to 24 and a higher score indicating more depressive symptoms.
Eating problems were assessed using an eight-item Norwegian version of the Eating Attitudes Test, found to be a valid version [
33,
34]. The items were rated on a scale ranging from 0 to 3, with total scores ranging from 0 to 24.
Antisocial behaviour was assessed using 19 variables of criminality, rule breaking, and other types of antisocial behaviour that had occurred in the previous 12 months. The variables were derived from a Norwegian version of a questionnaire used originally in the National Youth longitudinal study [
35] and from the Olweus Scale for Antisocial Behaviour [
36]. Those who responded affirmatively were scored 1 on each of the items, with a sum score from 0 to 19 and a higher score indicating more antisocial behaviour.
Loneliness was assessed using the revised UCLA loneliness scale [
37] with five items scored on a scale ranging from 1 to 4, with a sum score from 5 to 20 and a higher score indicating more frequent feelings of loneliness.
Intimate friendship was examined using the question “Do you have one close friend you can talk to when you have personal problems?” The answers were “Yes” or “No”, with the latter labelled “No intimate friend to talk to”.
Self-perceived poor health was examined using a question on how they perceived their current general health status. The response categories were on a five-point ordinal scale ranging from “very good” to “very poor”. The distribution on this variable was highly skewed, and the responses were dichotomized into “good self-perceived health” versus “poor or very poor self-perceived health”.
Analytic strategy and statistical analyses
In the first step we compared the proportion who reported contact with CAPS between three SH-groups; those who reported both SA and NSSH; those who reported either SA or NSSH; and those who reported no SH. Two hypotheses were put to test. First, that adolescents with both SA and NSSH would be more likely to have been in contact with CAPS than other SH groups and second, that any differences in such contact, would largely be explained by differences in the severity of psychosocial problems. We therefore analysed the bivariate association between SH groups and CAPS contact, and we further analysed the bivariate associations between these two variables on the one hand and indicators of psychosocial problems on the other. To address the second hypothesis, we compared the results of bi-variate and multi-variate analyses where CAPS contact was regressed on SH groups. All these analyses were conducted using the entire data set (n = 10976). The rationale for combining those who reported either SA or NSSH, was that we in a previous study using the same data set [
7] found that these two SH groups were similar with respect to psycho-social problems.
In the next step we assessed in bi-variate and multi-variate models which psychosocial variables characterized the use of CAPS within a sub-sample, which comprised those with both SA and NSSH (n = 490).
All statistical analyses were conducted using SPSS, version 21SPSS (Inc., Chicago, Illinois). In the bi-variate analyses, we first applied cross-tables and chi-square tests for categorical variables, and analysis of variance and F-tests for continuous variables. Logistic regression models were employed to estimate unadjusted and adjusted odds ratios for predictors of CAPS contact in both steps of the analyses. In the multivariate logistic regression analyses, we applied a stepwise procedure based on model-fit criteria (log likelihood ratio). The covariates considered for inclusion in the multivariate models had demonstrated a bivariate association (p < 0.20) with the outcome variable (CAPS contact). Missing data were excluded list wise.