Primary school students’ mental health in Uganda and its association with school violence, connectedness, and school characteristics: a cross-sectional study

This study used baseline data from the Good Schools Study, a cluster randomized controlled trial conducted in Luwero District in Uganda (registered at clinicaltrials.gov, NCT01678846) [23, 24].

The participants were selected in two stages. Firstly, after excluding 97 very small schools with less than 40 registered students in year 5 (Primary 5) and 20 schools with existing governance interventions, 42 schools were randomly selected from a list of the remaining 151 schools in the district. This sample represents 80 % of primary school students in the district. Secondly, a maximum of 130 students per school were selected from lists of all Primary 5, 6, and 7 students by simple random sampling.

Parents were informed about the study and the possibility to opt their child out from participating [23]. Informed written consent was sought from each selected student. The data were collected during individual interviews conducted in June and July 2012. Students were notified that there were no right or wrong answers and that all information would be kept private. Students disclosing serious maltreatment were referred to local child protection services [25]. Ethical approval for the study was obtained from the Ugandan National Council for Science and Technology (SS 2520) and the Ethics Committee of the London School of Hygiene and Tropical Medicine (#6183). The study protocol was published in detail elsewhere [23].

The following measures from the interview were used for this study. All measures were adapted and cognitively tested for understanding via an iterative process with a small group of primary school children and then pilot tested with 697 children attending Primary 5, 6, and 7 to explore item distributions before the main survey. The exact wording of the questions used for the creation of the individual-level exposure variables is displayed in Table 1.

Multilevel logistic regression was applied to examine the associations between school-related factors and mental health [32, 33]. This analysis method accounts for the clustered nature of the data, i.e. that students (level 1 units) are nested within schools (level 2 units), which is necessary for obtaining correct regression estimates. Furthermore, the approach enables the assessment of the role of the context on individual health outcomes [34]. A random effect for school was added to each fitted model to allow the intercept to vary across schools (random intercept model). A complete-case analysis was conducted, i.e. students with missing or non-substantive values (‘don’t know’, ‘no response’, ‘not applicable’) in one or more variables of interest were excluded from the analysis.

First, a model with the outcome but no exposure and confounding variables was fitted to estimate the intra-class correlation (ICC) and to assess evidence for clustering based on the provided likelihood ratio test (LRT). In order to have guidance for the multivariable analysis, a conceptual framework was created (Fig. 1) [35].

MODEL 1 contained all variables shown in the framework except school connectedness and connectedness to peers, in order to exclude variables that we hypothesized might lie on the causal pathway between violence and mental health and participation and mental health, respectively. MODEL 2 comprised all variables. Partial mediation was regarded as having occurred if the effect of the exposure variables on the outcome had decreased by at least 10 % in MODEL 2 compared to MODEL 1.

To quantify the school-level variation (i.e. investigate the role of the context), median odds ratios (MORs) were calculated [34]. This measure conveys how much the odds of mental health difficulties of a student would (in median) increase if he/she moved to a school where the odds of mental health difficulties was higher. The MOR can be directly compared to the ORs of individual (e.g. age, sex) or area (e.g. urban/rural) variables. The MOR is always larger than or equal to one, with the value one indicating no difference in the probability for mental health difficulties between schools. Lastly, LRTs were used to investigate whether the effect of one or more exposure variables on the outcome differed between boys and girls. A p-value below 0.05 was considered as statistically significant. All analyses were conducted using STATA 12 [36].

The participation rate of those sampled was 77 %. The reason for non-participation was mainly absenteeism from school during the period of data collection (19 %). Hence, 3,706 students participated in the study of which 52.3 % were girls (Table 2). The majority of students (81.4 %) were between 11 and 14 years old. Over half of students experienced violence from school staff in the past week. The prevalence of violence from peers in the past week was 21.1 %. Almost half of students were involved in making up rules in the past and most students (62.6 %) perceived that student’s views are considered all the time or most of the time. The majority felt close to other students all the time or most of the time (78.7 %). With a median of 8 and a range from 0 to 30 points, the Total Difficulties Score was highly skewed to the right. The 20.7 % that were categorized as children with mental health difficulties had scores of 14 and higher. The total number of students in Primary 5, 6, and 7 (school size) in each school varied between 73 and 734 (median = 135.5). The majority of schools was located in a rural area (64.3 %).

We did not find any evidence that school and peer connectedness mediated the association between violence and mental health, or between involvement in making up rules and mental health. Only the OR’s for consideration of students’ views were between 10 and 15 % lower in MODEL 2 compared to MODEL 1; indicating that connectedness partially mediated the association between consideration of views and mental health.

The MORs were 1.38 and 1.37 in MODEL 1 and MODEL 2, respectively (i.e. if a student moved to another school with a higher probability of mental health difficulties, the student’s odds of mental health difficulties would (in median) increase by around 1.4 times).

Our tests did not show any statistical evidence of a difference in effects of any of the individual-level exposure variables on mental health between boys and girls. The LRTs comparing models with an interaction term between each exposure variable and sex with a model without this term yielded p-values of 0.182 (different effect of ‘violence from peers in the past week’ on mental health for boys and girls in MODEL 1) and larger, indicating that the less complex model fitted better.

This study is one of very few rigorously conducted studies investigating the relationship between various factors related to school and mental health of students in a developing country, however, it is cross-sectional, and results should not be interpreted as causal.

While the proportion of sampled children who completed the survey within schools was quite high (77 %), selection bias cannot be completely ruled out. The main reason for non-participation was being absent during the data collection period. This may in turn be associated with working outside of school. Previously published results from this study indicated that the odds of experiencing different forms of violence was higher for those working one hour or more a day outside of school compared to those working less than one hour [21, 22]. Absenteeism may also be associated with poorer (mental) health status, which would mean our analyses may have underestimated the magnitude of the association between violence and mental health difficulties.

Questions about violence were taken from the ICAST-CI which has been widely tested and used internationally [26]. Students were repeatedly assured that teachers and staff members would not find out what they had answered to prevent underreporting of violence experiences. Nevertheless, underreporting might have occurred because of the shame and stigma which can be associated with some forms of abuse (e.g. sexual abuse) [27], or students may have had difficulty recalling some exposures. The main focus of the Good Schools Study was examining violence from school staff, hence more questions on this form of violence were included than on violence from other perpetrators. It is therefore possible that we failed to categorize some students as having experienced violence from others besides school staff because of the less detailed inquiry about violence experiences. For the assessment of mental health, the SDQ, a well-established and widely used instrument, was used. It should be noted that although having high levels of mental health difficulties on a screening tool such as the SDQ has a strong relationship with subthreshold and full-blown psychiatric disorders, a high SDQ score should not be interpreted as a psychiatric diagnosis [28, 29].

Lastly, the study was conducted only in one district of Uganda. However, the population in this district is demographically similar to the rest of the country and has both rural and urban areas [23]. Therefore, the results from this study are likely to be generalizable to other primary school students in Uganda.

Our results suggest that interventions for young people attending school should address their experiences of violence from school staff and from peers to improve mental health. Programs targeting bullying and other forms of violence in schools which have been developed in high - income countries and been proven to be effective should urgently be explored for their efficacy in low-income settings [37].

School connectedness is protective against student risk-taking behavior like substance use and positively impacts on educational outcomes [7, 40]. Fostering school connectedness is therefore a good strategy for strengthening mental health of students as well as for promoting other aspects of students’ health and academic achievements. Strategies for improving school connectedness have been integrated in existing intervention programs [40], and recommendations for increasing school connectedness have been developed by the Centers for Disease Control and Prevention that can serve as a practical guide for teachers [7].