Prospective associations between childhood externalising and internalising problems and adolescent alcohol and drug use

Participants

The sample comprised participants from the Bergen Child Study (BCS; for more information about the BCS and related publications, see uni.no/en/bergen-child-study), and the data stem from the first, second, and fourth waves of this study. The BCS is a longitudinal total population study of children in all public and private schools in the city of Bergen, Norway. The fourth wave of the BCS is nested within the youth@hordaland survey (Sivertsen, Harvey, Pallesen, & Hysing, 2017) covering the whole Hordaland county, in which Bergen is the largest city, as its target population. The BCS has been utilised in a range of previously published studies, and has also recently been used in a longitudinal design (Sivertsen et al., 2017).

The first wave of the BCS, conducted in autumn 2002, comprised a target population of 9430 primary-school children aged 7–9 years from the city of Bergen, and informed consent to participate was received from 7007 parents (74%) prior to inclusion in the study. The second wave was conducted four years later during spring 2006, and 5683 children aged 11–13 years participated (60% of the original target population). Six years later in winter/spring 2012, when the adolescents were 16–19 years of age, the target population was expanded to include the whole county of Hordaland, and 10,253 (53%) of the 19,439 invited adolescents participated. The three waves used in our study are labelled T1, T2, and T3.

A total of 2438 adolescents had participated in T1, T2, and T3 with valid parent or teacher responses on the Strengths and Difficulties Questionnaire (SDQ) at both T1 and T2, and therefore comprised the final sample. The mean age of the total sample at T3 was 17.4 years (standard deviation 0.8), and 53.7% of the sample were girls.

The study was approved by the Regional Committee for Medical and Health Research Ethics in Western Norway.

Explanatory variables: Childhood mental health problems

The variable of mental health problems was defined by scores on the Strengths and Difficulties Questionnaire (SDQ) (Goodman, 1997), which was completed separately by parents and teachers at T1 and T2. The SDQ is a screening questionnaire for children and adolescents aged 4–16 years, consisting of 25 items which describe positive and negative characteristics of children within five subscales: (1) emotional problems, (2) conduct problems, (3) hyperactivity/inattention problems, (4) peer/relationship problems, and (5) pro-social behaviour (not used in the current study). Each item is scored on a three-point scale – not true, somewhat true, and certainly true – with total subscale scores each ranging from 0 to 10.

The SDQ has been validated in several countries (Heiervang et al., 2007; Muris, Meesters, & van den Berg, 2003). A recent review found that the psychometric properties of the SDQ are strong, and recommended its use as a screening instrument (Stone, Otten, Engels, Vermulst, & Janssens, 2010). Importantly, the use of broader internalising and externalising scales from the SDQ is found to be acceptable in low-risk samples (Goodman, Lamping, & Ploubidis, 2010), has a good fit with the included subscales (Van Roy, Veenstra, & Clench-Aas, 2008), and the scales are relatively “uncontaminated” by one another (Goodman & Scott, 1999). Additionally, the prosocial scale has low correlation with the other four subscales (Goodman, 2001; Van Roy et al., 2008), supporting its exclusion from our analyses.

Accordingly, an externalising problems scale was constructed by merging the subscales of conduct problems and hyperactivity/inattention problems, while an internalising problems scale was constructed by merging the subscales of emotional problems and peer/relationship problems. We were interested in externalising/internalising symptoms that were present over time and across informants, and therefore added together the responses from teachers and parents from T1 and T2, and summed the scales to single, continuous variables. Moderate correlations between parent and teacher reports were found for both the externalising problems scale (R = 0.51) and the internalising problems scale (R = 0.47), which were evaluated as acceptable. The Cronbach’s α of the SDQ scales ranged from 0.81 to 0.85 in our sample.

Outcome variables: Alcohol- and drug-related problems

Self-reported measures of alcohol and drug use at T3 were our main dependent variables.

Included covariates

Age and gender for all participants were retrieved through personal identity numbers in the Norwegian National Population Registry. In addition, we measured self-reported perceived economic well-being by an item where participants rated their family’s economic situation either as (1) “Equal to others” (66.0%), (2) “Better than others” (28.8%), or (3) “Poorer than others” (5.2%). We also collected information on maternal and paternal educational attainment at T3 by two self-report items differentiating between primary school only, high school, or higher education. Perceived family economy, and maternal and paternal educational attainment were used as measures for socioeconomic status, and were entered separately into the regression analyses.

Statistical analysis

Strengths and limitations

Our study has several strengths. The sample consists of a well-defined population-based sample of children followed into adolescence, which is sufficiently large to enable a detailed investigation of longitudinal associations between childhood mental health problems and adolescent alcohol/drug use. An additional strength is the utilisation of repeated measures and multiple informants on the SDQ, providing a robust estimate for externalising/internalising symptoms.

The study has some limitations. First, although we employed a prospective design for the study with a temporal order of data collection, the findings of the study are not necessarily an expression of causality. The findings can also be explained, for example, by third factors (such as genetic predisposition, adverse life circumstances, and family characteristics), which may act as risk factors for both externalising/internalising problems and alcohol/drug use during adolescence. Additionally, we had available measures of alcohol/drug use only at T3, therefore failing to control for prior substance use. Second, complex mechanisms may be at work as mediators (e.g., timing of puberty, peer involvement, parental supervision, and parental problem drinking) between childhood externalising/internalising problems and subsequent alcohol/drug use (e.g., Dickson, Laursen, Stattin, & Kerr, 2015; Finan, Schulz, Gordon, & Ohannessian, 2015). This is beyond the scope of our study. Third, the information on alcohol/drug use was based on self-reports, while the information on childhood mental health problems was based on parent and teacher reports. The lack of clinical interviews thus adds as a further limitation to our study. However, whereas clinical diagnoses of either mental health or alcohol/drug-related problems are typically categorised as dichotomous measures, previous studies have also gained support for dimensional conceptualisations of common mental health problems (e.g., Andrews et al., 2007) and alcohol/drug-related problems (Beseler & Hasin, 2010; Krueger et al., 2004) that are not necessarily above a formal, clinical cutoff. Our study thus highlights how parent/teacher-reported externalising/internalising symptoms across a spectrum of severity are associated with self-reported alcohol/drug use. Future studies that also apply diagnostic levels of respectively externalising/internalising problems and alcohol/drug-related problems should therefore be encouraged. Fourth, although it might be preferable to have more detailed information on illicit drug use, only lifetime use was available in the youth@hordaland-survey. Illicit drug use is also particularly infrequent among Norwegian adolescents compared with other European countries (Kraus et al., 2016), and the inclusion of frequent illicit drug use could therefore be too strict a measure in the context of this study. Fifth, depression and conduct problems have previously been found to be the mental health symptoms most strongly associated with alcohol/drug use during adolescence (e.g., Armstrong & Costello, 2002; Hussong et al., 2017). In addition, associations with alcohol/drug-related problems tend to vary across subtypes of anxiety problems (Fröjd, Ranta, Kaltiala-Heino, & Marttunen, 2011; Ohannessian, 2014). However, the internalising measures of the SDQ, including its subscales, are generalised measures of internalising problems, which therefore do not separate between, for example, symptoms of anxiety and depression, and between subtypes of anxiety. Therefore, our findings may not be generalisable to all manifestations of internalising symptomatology. Sixth, only parent and teacher reports on SDQ were available at T1. Although teacher and parent versions of the SDQ have gained support in a review of psychometric studies (Stone et al., 2010), it should be noted that internalising problems are likely to be reported more accurately by children themselves (Ederer, 2004). The inclusion of self-reported externalising/internalising symptoms would add more strength to the measures of childhood internalising problems. However, there is no self-report data available on SDQ symptoms at T1 due to the young age of the children. Our externalising/internalising variables are constructed based on two time points, and thus including self-reports at one time only would not work. In addition, the use of parent/teacher reports on T1/T2 and self-reports on T3 may contribute to avoiding mono-informant bias. Finally, selective drop out is a well-known problem in longitudinal research (Wolke et al., 2009), and may not be fully ruled out in our study. However, the discrepancy between T1 (n = 7007) and the study sample (n = 2438) is not merely a result of dropout at each consecutive wave, but is also an administrative issue of the longitudinal survey design. More specifically, the BCS (comprising T1 and T2) had a different target population than the youth@hordaland (comprising T3). Therefore, it is not possible to calculate a strict attrition rate for the present study. We did, however, find some differences between the study sample and the full youth@hordaland sample, suggesting that the included individuals had somewhat higher SES than the non-included individuals. While this was the case, effect sizes for differences between the study sample and the full youth@hordaland sample were for the most part non-significant on alcohol/drug use, and were small on externalising/internalising problems. Hence, selective dropout is not likely to be an issue that would seriously bias our findings.