To assess temporal trends in child mental health, we used scores from the teacher- rated version of the SDQ [
24] (Additional file
1) which indicates the likelihood of difficulties in five areas: emotional symptoms, conduct problems, hyperactivity/inattention, peer relationship problems, and prosocial behaviour. As well as five subscale scores, each case has a ‘total difficulties’ score based on the summing of scores on the first four subscales (i.e., all but prosocial behaviour, which is positively scored). Scores can range from 0 to 40, with a threshold of 16 and above on the total difficulties scale suggesting that a child may have a psychiatric diagnosis and/or may require further assessment and support. The SDQ was completed between February and April every year between 2010 and 2015 by preschool education staff for children in Glasgow City during their pre-school year (i.e., between the ages of 4.5 and 5.5 years) as part of routine transition documentation. This meant that the first wave of data collection occurred shortly before the launch of population-wide Triple P in Glasgow City.
Preschool staff were given training in groups by researchers (with top-up training available each year for new staff), as well as written guidance, on completing the SDQ. Whilst in some nurseries the child development officer (CDO, key worker) for the child completed the SDQ alone, in other nurseries the CDO completed the SDQ alongside the nursery head and/or other staff [
31]. Inter-rater reliability could not be assessed.
Response rates have grown year on year: from 67.4% of pupils in 2010 to 95.7% of pupils in 2014 and 89.8% in 2015, based on the Scottish Government pupil census. Overall, 224 preschools completed SDQs for at least one time point, while 46 preschool establishments completed data every year. In order to assess whether nurseries that returned data in the early years were systematically different from those who only returned data later, we completed analyses both on all available data and the limited sample of 46 nurseries, and demonstrated minimal differences in trends.
Each child had a SDQ Total Difficulties score and constituent domains recorded along with a code for the pre-school establishment that they attended. In 2010, data were completed using paper questionnaires, whilst in 2011 a mixture of paper and electronic forms were completed, moving to solely electronic completion by 2012. These data were linked to data held in the education services database comprising age, gender and deprivation status of the locality in which the child lived.
Two versions of the SDQ are available for this age group: a version for 3–4 year olds and one for 4–16 year olds. The 4–16 version was used in 2010 and 2011, but changed to the 3–4 version in 2012, following interviews with pre-school staff [
29]. The difference between the two versions is relatively small and involves the wording of two questions in the 4–16 conduct scale, about lying, cheating and stealing, which were changed to slightly ‘softer’ and more age appropriate questions about being argumentative with adults and being spiteful. Staff may have more readily answered positively to these softer questions in 2012 and thereafter [
29]. For this reason analyses conducted in this report use the shorter version of the conduct problems scale using only the same three questions which were asked every year. A score was then calculated in the usual way for SDQ, though using three answers rather than four or five: the three questions were used to give an average score and then multiplied by five in the usual way. Thus the same cut-offs for Conduct Problems were used. Cronbach’s alpha for the 5-item Conduct Problems scale in the present sample was 0.72. This fell slightly to 0.66 when it was reduced to three items. A score of 0.70 or above is usually deemed acceptable, although scores are partly related to the number of items in the scale as well as their internal consistency [
32].
Since information on the socioeconomic status of individual children (e.g. household income or occupational social class of head of the household) was not available, deprivation status was based on the postcode of the children’s home address and measured using the Scottish Index of Multiple Deprivation (SIMD). For each postcode, the publicly available SIMD was obtained. The SIMD is a continuous measure of compound social and material deprivation, calculated using income, employment, health, education, housing, geographic access to services and crime data [
33]. The measure is updated approximately every 4 years. The current analysis used the 2009 SIMD measure. SIMD is normally banded into quintiles. Due to the skew of the data in Glasgow City, whereby around 50% of people fall into the ‘most deprived’ SIMD quintile, Glasgow City have developed their own quintiles (GIMD quintiles), which allow for more in-depth analysis of deprivation within the city, and it is these which are used in these analyses [
33].
Analysis
Annual trends for SDQ total difficulties scale scores are presented, together with data trends by levels of deprivation of the children.
In order to reduce the risk of bias associated with different pre-school establishments (and thus different groups of staff) completing SDQ data in successive years, trends were plotted for the subgroup of these establishments which returned data in each year 2010–2015. Both continuous scores and banded scores were included in the analyses in recognition of the hypothesis that, although we are primarily interested in children moving out of the abnormal range, there may have been improvements made within the normal range in this whole population sample. Results for the conduct problems subscale are presented using only the 3-question version of the scale which was the same every year.
Tests for change over time were ANOVA tests and Post Hoc multiple comparisons in the form of Tukey tests. A multilevel approach (using child and nursery as levels) was explored but rejected due to the lack of difference between levels.
Linear regression models were fitted in order to investigate predictors of total difficulties scores. Variables entered into the models were cohort, sex, ethnic status, Looked After status, and GIMD.
Ethics
Formal ethical review board approval was not required for the present analysis of the anonymised dataset, but the ethical issues possibly raised by this study were considered by the research team. It was concluded that the project posed no harm to the participants, the schools or the different regions, as the anonymized data were collected by educational establishments as part of the routine documentation passed to primary schools for the benefit of teachers and pupils. A data sharing agreement was in place between Glasgow City Council and the University of Glasgow to facilitate the safe sharing of data.