Participants
Participants were adolescent and young adult males, aged 12 to 25 years (M = 16.65; SD = 1.43). Most were 15–17 years (80.5%), while the remainder being 12–14 (8.5%) and 18–25 (11.0%). They were in custody in two large youth detention centers (YDCs) in urban areas in the Netherlands, where the MAYSI-2 and SDQ were given as a routine part of the detention centers’ intake processes, to all entering youth consecutively between May 2008 and December 2012 (for details, see for example: [25, 26]). For the current study, data were used from 1259 detained male adolescents who completed the mental health screening and assessment protocols and for whom official criminal records were available. Regarding ethnicity (as defined below), 22.6% of the boys were of Dutch origin, 25.5% of Moroccan origin, 21.1% of Surinamese/Antillean origin, and 30.0% included a wide variety of ethnic or national origins. These percentages are consistent with those presented in prior work with detained boys in the Netherlands (e.g., [18]). For 10 boys (0.8%), information to determine ethnicity was lacking, and these boys were excluded from the study, resulting in total sample of 1249 boys.Footnote 1
Measures
Massachusetts Youth Screening Instrument-Second Version (MAYSI-2 [12])
The MAYSI-2 is a 52-item screening tool in which youth answer questions (yes/no) that sample the presence or absence of symptoms or behaviors related to several areas of emotional, behavioral, and psychological disturbances. The MAYSI-2 was specifically designed and normed for use among youth entering a juvenile justice setting, and can be administered in about 15 min by computer or paper and pencil self-report. Factor analyses indicated that the items produce scores on six clinical scales: Alcohol-Drug Use (8 items), Angry-Irritable (9 items), Depressed-Anxious (9 items), Somatic Complaints (6 items), Suicide Ideation (5 items), and Thought Disturbance (for boys only; 5 items); and one non-clinical scale (Traumatic Experiences; 5 items). There is no MAYSI-2 total score as the test was not intended to measure a broader construct such as mental distress or emotional disturbance [12]. None of the scales were intended to be diagnostic of DSM-5 mental disorders, merely to identify symptoms suggesting the need for further assessment (e.g. [27]). Each clinical MAYSI-2 scale has a “Caution” cutoff empirically developed to identify youth who might be in need of clinical attention [28]. Each clinical scale also has a “Warning” cutoff identifying scores obtained by the top 10% of youth in the original Massachusetts normative sample [12], flagging youth who are most in need of clinical attention.
The present study used the official Dutch version of the MAYSI-2 [29] which was developed using translation and back-translation procedures. The Dutch MAYSI-2 has been shown to have good psychometric properties in terms of factor structure, internal consistency, and construct validity [5, 25, 27] in youth being detained in the Netherlands, including detained youth from Dutch, Moroccan, Surinamese/Antillean, and Mixed ethnicity.Footnote 2 The MAYSI-2 was introduced in various European countries in the past eight years, including the Netherlands (see: http://www.inforsana.eu). Pending further information being developed in Europe, clinicians are guided to use the cut-off scores developed for use in the U.S. [12, 30]. The current study relied on the six raw clinical MAYSI-2 scale scores and U.S. based Caution cut-offs (unless otherwise stated).
The Cronbach’s alpha (α) and mean inter-item correlation (MIC) for the six clinical MAYSI-2 scales in the total sample (N = 1249) were as follows: Alcohol/Drug Use (α = 0.84; MIC = 0.40); Angry-Irritable (α = 0.76; MIC = 0.27); Depressed-Anxious (α = 0.67; MIC = 0.19); Somatic Complaints (α = 0.58; MIC = 0.19); Thought Disturbance (α = 0.50; MIC = 0.17); and Suicide Ideation (α = 0.77; MIC = 0.41). Of note, α can be interpreted as follows: < 0.60 = insufficient; 0.60–0.69 = marginal; 0.70–0.79 = acceptable; 0.80–0.89 = good, and 0.90 or higher = excellent [31]. Because α penalizes shorter scales, [32] we also presented MIC values, which is considered to be a more straightforward indicator of the internal consistency of a scale than α, and should be at minimum in the range of 0.15 to 0.50 to be considered adequate [33]. Additional file 1: Part 1, presents α and MIC values for the six MAYSI-2 scales across the four ethnic groups.
The Strength and Difficulties Questionnaire self-report version (SDQ [17])
The SDQ is a self-report and third-party informant (parent and teacher) screening instrument for psychosocial functioning of children and adolescents. The current study used the self-report version. The SDQ has five subscales, each with five items offering three response categories (Not true = 0, Somewhat true = 1, Certainly true = 2), has been used with detained youth in prior research (e.g., [18, 26]), and is used internationally (e.g., [34,35,36]). The present study used two SDQ scales—Conduct Problems, and Hyperactivity—that are not covered by the MAYSI-2. “Borderline Cut-off” scores for these two scales are 4 and 6, respectively [37]. The current study used the raw scores and borderline cut-offs unless otherwise specified. The α and MIC for the two SDQ scales in the total sample (N = 1249) were as follows: Conduct Problems (α = 0.55; MIC = 0.22) and Hyperactivity (α = 0.79; MIC = 0.43). Of note, prior work revealed that αs for these latter two scales ranged from 0.47 to 0.60 (Conduct Problems), and from 0.66 to 0.67 (Hyperactivity) in epidemiological sample of British adolescents [38] and a community sample of Dutch adolescents [39]. Additional file 1: Part 1, presents α and MIC values for these two SDQ scales across the four ethnic groups.
Omnibus variable
Using the MAYSI-2 and SDQ, we also created an “omnibus variable” that reflects the number of times participants were at or above the Caution (MAYSI-2) or Borderline Cut-off (SDQ) on the eight scales being used to measure eight different types of mental health problems (i.e., six MAYSI-2 and two SDQ scales). This omnibus variable, from here onwards referred to as “Omnibus Mental Health Problems” (theoretical range 0–8), was intended to be indicative of the severity or multiplicity of mental health problems. The percentages of boys at or above various cut-off scores can be retrieved from Additional file 2: Part 2.
Violent criminality
Violent arrest was defined as any offense involving physical harm to another person (e.g., manslaughter, theft with violence, and sex offenses). Data were gathered based on the General Documentation Registry (GDR) of the Ministry of Justice Court Documentation Service of the Netherlands. The Registry contains information on the number, time, and nature of all criminal cases registered at the Public Prosecutor’s Office, including their adjudication. We used all registered cases, regardless of their adjudication. Specifically, in addition to cases that ended in a guilty ruling, cases that ended in a prosecutorial waiver or an acquittal were also included when reconstructing the respondents’ criminal career. Data include all such information from age 12, which is the minimum age of legal responsibility in the Netherlands, to the respondents’ age on June 30th 2013, which represents the end of the follow-up period for this study. The variable Past Violent Arrests refers to the number of violent arrests before the completion of screening (i.e. shortly after detention intake, see Procedure). The variable Future Violent Arrests refers to the number of violent arrests in the follow-up period, that is the weeks between completion of screening and June 30th 2013.Footnote 3 The percentage of youth with at least one prior violent arrest was 76.1% for Dutch boys, 74.1% for Moroccan boys, 86.1% for Surinamese/Antillean boys, and 79.1% of Mixed Origin boys. For future violent arrest these percentages were 27.5% (Dutch), 34.9% (Moroccan), 41.4% (Surinamese/Antillean), and 32.8% (Mixed Origin).
Ethnic background
Based on the Dutch standard classification of ethnic groups [40] and in line with prior work from the Netherlands (e.g., [5]), a participant was categorized as “Moroccan” or “Surinamese/Antillean” when the adolescent himself and/or at least one parent had been born in Morocco or Surinam/Dutch Antilles, respectively. When both parents were of different non-Dutch origin, we used the mother’s country of birth to determine the child’s ethnicity. Participants were classified as Dutch when both parents and the child were born in the Netherlands. All other participants were assigned to the “Mixed Origin” group, implying not “mixed identity” for any one participant, but simply a group comprised of mixed ethnic origins.
Procedure
The MAYSI-2 and SDQ were administered on a standalone computer within a few days after detention entry (Mean number of days = 3.3, SD = 5.6) in the presence of non-clinical personnel, to all youth entering YDCs. Assistance was available at request (e.g., if the youth did not understand a question). When reading abilities were insufficient, the questionnaires were read to the youth. Youth were made aware that the mental health screening and assessment were part of the YDCs’ clinical protocol and that all the outcomes from this protocol were available to YDCs personnel (e.g., clinicians) and could be included in their file. Through standardized oral and written information provided by the YDCs upon start of detention, youth and their parents/care-takers were informed that the mental health screening and assessment outcomes would be used for scientific research, unless they declined (passive informed consent). They were also informed that, if they did not decline, their information would be transferred anonymously to the researchers, so that information could not be traced back to them. The Medical Ethical Review Board of the Leiden University Medical Center deemed study protocols to be exempt from review because data were collected by the YDCs as part of a clinical protocol and for clinical purposes.
Data-analyses
Multivariate Poisson regression analyses (with 95% confidence intervals [CI]) were conducted to examine the relation between mental health problems and future violent arrests. These analyses were performed in two ways. First, we examined the relation of each MAYSI-2 and SDQ scale score to violent arrests (called the “bivariate model”). Second, we examined each scale’s relation to violent arrests when all other scales were added to the analysis, together with three control variables, being: age (at detention entry), number of past violent offenses, and follow-up time (called the “multivariate model”). These control variables are important to consider because age is inversely related to criminal recidivism (e.g., [41]), because past violent offending is a robust predictor of future violence (e.g., [42]), and because some research has suggested that mental health problems may lose their value for predicting future violent offending after controlling for prior violent offending (e.g., [7]). It is also important to account for differences in the time participants had to commit new violent crimes. Therefore, follow-up time was used as a control variable as well. To avoid finding significant differences due simply to random error when computing large numbers of tests, we discounted any significant relations as “uninterpretable” (nullified) if 20% or fewer significant relations were revealed within an ethnic group. Specifically, this implies that when running nine tests in one ethnic group (i.e., eight single scale models plus one control model) at least 2 or more significant effects must be revealed. This is a conservative criterion, as “chance” findings of significance by random error in multiple comparisons usually are interpreted as 1 in 20 (5% of comparisons) (e.g., [43]).
Next, the aforementioned analyses were repeated using the Omnibus Mental Health Problems variable instead of the raw MAYSI-2 and SDQ scores. This omnibus variable (i.e. number of times at or above MAYSI-2 and SDQ cut-offs) may be appealing for clinicians who want to identify youth with comorbid mental health problems for decision making related to screening, and may prefer to use dichotomies rather than dimensional scores [44]. However, these cut-off scores derived in the U.S. (MAYSI-2) or Britain (SDQ) might not be optimal to identify detained youth in the Netherlands with elevated mental health problems.
To circumvent the potential problem that our Omnibus variable is based on a less-than-optimal cut-off score, we also performed latent profile analyses (LPA) using Mplus 6.1 [45] to identify distinct subgroups based on their permutations of raw MAYSI-2 and SDQ scale scores. LPA is a data-driven, person-oriented, model-based clustering technique to assign youth to mutually exclusive subgroups and uses statistical criteria to compare models to identify the optimal number of groups to retain [46]. Technical details for LPA are provided in Additional file 3: Part 3. In this study, the six raw MAYSI-2 and two raw SDQ scale scores were used as the clustering variables in LPA. The outcome of these LPA will be used for comparison and predictive purposes. All analyses were performed separately for each ethnic group. SPSS 23.0 was used, unless otherwise specified, with p < 0.05 as an indicator of statistical significance.