Data collected as part of the SA National Student Mental Health Survey were used to: (1) estimate 12-month prevalence of common mental health problems and self-harm; (2 estimate the proportion of students receiving treatments for the various mental health problems; (3) explore barriers to treatment; and (4) investigate sociodemographic predictors of treatment as mediated through barriers. The study is part of the ongoing work of the World Health Organization (WHO) World Mental Health Surveys International College Student Initiative (WMH-ICS) [21], which seeks to expand access to evidence-based treatments for mental disorders among students across the globe.
Procedure
All 26 public universities in SA were invited to participate in the survey, of which 17 agreed to be included. No reasons were given by the 9 non-participating universities, among which there were 2 HWIs, 3 HDIs and 4 UTs. Data were collected between April and October 2020. Participating universities distributed emails inviting all their undergraduate students to complete an anonymous online survey (N = 657,432). Two follow-up reminder invitations were sent to students, approximately a week apart. The study was initiated by Universities South Africa and funded by the South African Medical Research Council.
Procedures and measures
The survey was administered via Qualtrics (a web-based platform used for administering electronic surveys). Responses were self-administered by participating students and the following information was obtained:
Socio-demographic characteristics
Students reported their age, gender, population group, sexual orientation, parents’ education and whether they were full-time or part-time students. For population group we used the categories in government policies and the official population census (i.e., Black-African, Coloured, White, Asian, and “Other”) to explore disparities in mental health utilisation that may have resulted from the country’s history of racial segregation.
Mental health problems
Self-report information was collected to assess 11 common mental health problems, including 4 anxiety-based disorders (generalized, anxiety disorder (GAD), panic disorder, post-traumatic stress disorder (PTSD), social phobia), 2 mood disorders (major depressive episode (MDE), bipolar spectrum disorder), 3 disruptive behavior disorders (ADHD, eating disorder, intermittent explosive disorder) and 2 substance use disorders (alcohol use disorder, drug use disorder). We used the Composite International Diagnostic Interview Screening Scales (CIDI‐SC) [22, 23] to assess all disorders other than for alcohol use disorder, which we assessed using the Alcohol Use Disorders Identification Test (AUDIT) [24]. Previous cross-national research has documented good validity of these assessments compared to clinical evaluations [22, 23, 25].
Self-harm
Suicidal thoughts and behaviours were assessed using a modified version of the Columbia Suicidal Severity Rating Scale (C-SSRS), which has demonstrated good convergent and divergent validity with other multi-informant suicidal ideation and behavior scales used with adolescents, as well as showing high sensitivity and specificity for suicidal behavior classifications compared with other behavior scales and clinician evaluation [26]. Students were asked about passive suicidal ideation (i.e. wish you were dead or would go to sleep and never wake up), active suicidal ideation (i.e. thoughts of killing yourself), suicide plans (i.e. think about how you might kill yourself), suicide attempts (i.e. purposefully hurt yourself with at least some intent to die), and non-suicidal self-injury (NSSI) (i.e. do something to hurt yourself on purpose, without wanting to die, like cutting yourself, hitting yourself, or burning yourself)). Students who endorsed any of these items where then asked which of these problems occurred within the past 12-months.
Symptom severity
To measure level of impairment related to mental health problems (i.e. severity of symptoms) we used the Mental Component Score (MCS) of the Veterans RAND 12-Item Health Survey (VR-12) [27]. The VR-12 is a 12-item scale assessing 8 domains of health; namely, physical functioning, role limitation due to physical problems, bodily pain, general perception of health, social functioning, role impairment due to emotional problems, vitality, and mental health. The MCS was derived from the VR-12 questions assessing social functioning, role limitation due to emotional problems, vitality, and mental health. These items were then rescaled to yield a score ranging from 0 to 100, with higher scores indicating better health and less impairment [28]. The MCS has a mean of 50 and SD of 10 in the US population. Students who scored two standard deviations (SD) below the mean were defined as having severe symptoms while those who scored between one and two SDs below the mean were defined as having moderate symptoms and those who scored less than 1 SD below the mean were defined as having only mild symptoms.
Mental healthcare utilization
Students were asked if they had ever accessed treatment for an emotional or substance use problem and, if so, whether this occurred in the preceding 12-months. If treatment was received, the assessment asked separately if the treatment included psychological counselling, medication, or both.
Perceived need for treatment
We assessed perceived need for treatment by asking students who did not obtain treatment: Was there ever a time in the past 12-months when you felt that you might need psychological counseling, medication, or some other type of treatment for any emotional or substance use problems? Only students who answered affirmatively were queried about barriers to treatment.
Barriers to treatment utilisation
Students who did not receive treatment even though they screened positive for one or more of the 11 common mental health problems we assessed and/or self-harm and recognized a need for treatment were then asked about the importance of 9 barriers to treatment seeking commonly reported in prior student surveys (see footnote to Table 2).
Data processing
Standard calibration methods were used to weight the data within institutions to adjust for differences between survey respondents and the population on profiles defined by gender, population group, and year in school [29]. A second weight was then used to adjust for differences in survey response rates between institutions (Additional file 1: Table S1). Full descriptions of weighting procedures are reported elsewhere [4]. Multiple imputation (MI) across 30 MI replicates by chained equations was used to adjust for item-missing data [30].
Data analysis
We calculated 12-month prevalence estimates for mental health problems and self-harm, as well as gross associations with perceived need and treatment with cross-tabulations across the 30 multiple imputed datasets using Rubin’s rule [31]. MI-adjusted standard errors to adjust for the weighting and clustering of observations were obtained through the Taylor series linearization method [32]. We then used a data-driven method, random forests (RF) regression [33], to estimate the joint associations of the various groups of mental health problems (i.e. anxiety disorders, mood disorders, externalizing disorders, and self-harm) with probability of obtaining treatment. Given the computational complexity of RF using MI, the RF analysis was carried out at the person level among respondents who were imputed to have at least one condition in at least one imputation using counts of number of imputed with each condition imputed to be present. We retained the individual-level predicted probability of treatment based on this RF analysis as a control variable in subsequent prediction analyses described below.
Before carrying out other prediction analyses, though, we assessed the structure of reported barriers to treatment using principal axis factor analysis with oblique rotation to investigate the structure among responses to the questions about barriers (see the footnote to Table 2 for a full description of the 9 barriers). Missing values were imputed to the mean in carrying out this analysis. We then created summary dichotomous measures to describe whether each student reported one or more barriers within each factor to be either a very important or an important reason for not obtaining treatment. We generated a Venn diagram to examine the inter-correlations among these reports to define multivariable barrier profiles.
The prediction analyses used Poisson regression models with robust error variances [34] to estimate associations of sociodemographic factors and university type (i.e., HWI, HDI UT, DLU) with perceived need, barriers among students with perceived need, treatment, and treatment controlling for the RF predicted probability associated with disorder profiles. Poisson regression coefficients and ± two standard errors of these coefficients were exponentiated to create risk ratios (RRs) and 95% confidence intervals (95% CI).
We then decomposed the significant RRs of sociodemographic factors and university type with treatment by re-estimating the Poisson regression model in subsamples that excluded students with no perceived need and then successively excluded students with each type of barrier. This subsample analysis was used rather than control variable analysis (i.e., controlling for perceived need and barriers in a multivariable model) because control variable analysis is not possible when none of the people with the control variables received treatment. The importance of perceived need and barriers in explaining the RRs of the predictors with treatment was inferred in the subsample analysis by examining changes in RRs when we excluded respondents who lacked perceived need or reported various barriers.
Ethics
Ethical clearance was provided by the Health Science Research Ethics Committee of Stellenbosch University (Reference: N13/10/149). Institutional permission was obtained from all participating universities. Students provided informed consent electronically prior to data collection. Information about crisis and student counselling services was provided to all participants. Anonymised and de-identified data were securely stored on a password protected cloud-based server. The research was performed in accordance with the Declaration of Helsinki.
Add Comment