Introduction
Childhood trauma refers to adverse experiences and conditions during childhood and includes emotional, physical and sexual abuse as well as emotional and physical neglect (Morgan and Fisher, Reference Morgan and Fisher2007). Research suggests that childhood trauma has enduring neurobiological effects (Gunnar and Quevedo, Reference Gunnar and Quevedo2007; McCrory et al., Reference McCrory, De Brito and Viding2011) and is associated with a range of physical and mental illnesses in adolescence and adulthood (Kraan et al., Reference Kraan, Velthorst, Smit, de Haan and van der Gaag2015; Mandelli et al., Reference Mandelli, Petrelli and Serretti2015; Baumeister et al., Reference Baumeister, Akhtar, Ciufolini, Pariante and Mondelli2016). These include both clinical and subclinical levels of schizophrenia (Matheson et al., Reference Matheson, Shepherd, Pinchbeck, Laurens and Carr2012). Further, childhood trauma may impact the course of schizophrenia and affect symptom severity and prognosis (Matheson et al., Reference Matheson, Shepherd, Pinchbeck, Laurens and Carr2012; van Dam et al., Reference van Dam, van der Ven, Velthorst, Selten, Morgan and de Haan2012; Trotta et al., Reference Trotta, Murray and Fisher2015).
Two recent meta-analyses of observational studies found that experiences of childhood adversity were significantly higher among schizophrenia cases compared with controls (Matheson et al., Reference Matheson, Shepherd, Pinchbeck, Laurens and Carr2012) (Varese et al., Reference Varese, Smeets, Drukker, Lieverse, Lataster, Viechtbauer, Read, Van Os and Bentall2012). These analyses found an increased risk of psychosis or schizophrenia due to childhood trauma, irrespective of study designs or trauma measure.
Nearly all studies to date have examined these relationships using samples from high-income countries (Trotta et al., Reference Trotta, Di Forti, Mondelli, Dazzan, Pariante, David, Mulè, Ferraro, Formica, Murray and Fisher2013; Fisher et al., Reference Fisher, McGuffin, Boydell, Fearon, Craig, Dazzan, Morgan, Doody, Jones, Leff, Murray and Morgan2014; Dantchev et al., Reference Dantchev, Zammit and Wolke2018; Sideli et al., Reference Sideli, Fisher, Murray, Sallis, Russo, Stilo, Paparelli, Wiffen, O'Connor, Pintore, Ferraro, La Cascia, La Barbera, Morgan and Di Forti2018). Among the few studies conducted in low- and middle-income countries, South African studies found that childhood trauma was associated with premorbid adjustment and schizophrenia outcomes (Kilian et al., Reference Kilian, Asmal, Chiliza, Olivier, Phahladira, Scheffler, Seedat, Marder, Green and Emsley2017a, Reference Kilian, Burns, Seedat, Asmal, Chiliza, du Plessis, Olivier, Kidd and Emsley2017b; Asmal et al., Reference Asmal, Kilian, du Plessis, Scheffler, Chiliza, Fouche, Seedat, Dazzan and Emsley2018). Further, studies in LMICs have rarely considered the type of event or the number of cumulative traumas. It is important to study these relationships in LMICs in order to understand the context-specific schizophrenia risk and morbidity associated with childhood trauma, and to potentially inform appropriate interventions and policies to improve social and mental health outcomes for children. South Africa has a high prevalence of childhood trauma in the general population and more evidence is needed to understand the experience of trauma among people with severe mental illness (Jewkes et al., Reference Jewkes, Dunkle, Nduna, Jama and Puren2010; Denckla et al., Reference Denckla, Consedine, Spies, Cherner, Henderson, Koenen and Seedat2017).
The current paper seeks to contribute to the literature on childhood trauma and schizophrenia in LMICs through the following aims: (1) to describe the prevalence of an array of childhood traumas in cases and controls; (2) to examine how types of childhood trauma are associated with schizophrenia; and (3) to examine the associations between the cumulative number of traumas and schizophrenia.
Methods
Sample
This study utilized data from the Genomics of Schizophrenia in the South African Xhosa people (SAX) project. SAX is a case–control study designed to identify genetic variations and social exposures related to schizophrenia risk in a sample of the Xhosa population. Xhosa speaking people are one of the largest indigenous groups in South Africa and live mainly in the Eastern and Western Cape regions of South Africa (Koen et al., Reference Koen, Niehaus, Wright, Warnich, de Jong, Emsley and Mall2012; Campbell et al., Reference Campbell, Susser, Mall, Mqulwana, Mndini, Ntola, Nagdee, Zingela, Van Wyk and Stein2017a).
The recruitment strategies of the study have been described in previous publications (Campbell et al., Reference Campbell, Susser, Mall, Mqulwana, Mndini, Ntola, Nagdee, Zingela, Van Wyk and Stein2017a, Reference Campbell, Sibeko, Mall, Baldinger, Nagdee, Susser and Stein2017b). Initiated in January 2013, participants were Xhosa speaking people recruited from provincial hospitals and clinics in the Western and Eastern Cape regions of South Africa over a 5-year period. Xhosa language and ethnicity were self-reported.
Cases were defined using the following criteria: respondents who received a clinical diagnosis of schizophrenia or schizoaffective disorder for at least 2 years duration, and could give informed consent. Controls were defined as respondents who presented for treatment of non-psychiatric medical conditions at university-affiliated general medical hospitals and community health centers (CHC) that draw from similar catchment areas to the psychiatric hospitals and clinics where cases were recruited, matched by age group (e.g. 21–25) and gender.
The complete sample consists of 1420 cases (50.77%) and 1377 controls (49.23%) (n = 2797). Within the study sample, exposure data were completely missing for 326 respondents (11.6% of the total sample), and partially missing for 374 respondents (13.3% of the full sample). We attribute this to the inconsistent administering of the CTQ when the study began in 2013. To maximize the sample size, the main analysis imputed missing items for respondents with partial exposure data, using multiple imputations with 15 combined datasets, with corrected standard errors (Rubin, Reference Rubin1987). Imputation models included schizophrenia status, age, education, and all traumatic experience items. Those with missing responses for all traumatic experiences were excluded. Imputed model estimates were compared to unimputed estimates to examine the robustness of the analytic models to the degree of missing data. The final analytic sample comprised 2471 individuals (49.5% cases). Compared with the analytic sample, respondents with completely missing data were younger (age 34.9 v. 36.2, p = 0.0195) and more likely to be cases (60.7% v. 49.5%, p = 0.0002). There were no differences in education and gender between included and excluded samples.
Measures
The Structured Clinical Interview for Diagnosis for axis I disorders (DSM IV-TR version) (SCID-I), a widely used semi-structured interview tool (Spitzer et al., Reference Spitzer, Williams, Gibbon and First1992; First et al., Reference First, Spitzer, Gibbon and Williams1997) was administered to participants in Xhosa by one of the trained psychiatric nurses. The SCID-I has been found to have acceptable internal consistency, test–retest reliability as well as concurrent and predictive reliability (Spitzer et al., Reference Spitzer, Williams, Gibbon and First1992). The SCID-I collects data on psychopathology and physical illnesses including cardiovascular disease, diabetes, Tuberculosis and HIV/AIDS. For this study, interrater reliability for the translated Xhosa version of the SCID-I was obtained on a smaller sample of participants (N = 22). Reliability was good for the principle psychotic disorder diagnosis (κ = 0.74, p < 0.001). In addition to the patient interview, information was considered in the diagnostic process from referral notes, past and current clinical records, interviews with other members of the multidisciplinary teams and information from family members or other acquaintances of the patients. SCID-I interviews typically lasted 1.5–4.5 h.
The Childhood Trauma Questionnaire (CTQ) was administered to measure participants' childhood trauma experiences. The CTQ has been employed in several contexts, including in South Africa (Martin et al., Reference Martin, Viljoen, Kidd and Seedat2014; Kilian et al., Reference Kilian, Burns, Seedat, Asmal, Chiliza, du Plessis, Olivier, Kidd and Emsley2017b). The CTQ consists of 25 items designed for retrospective self-report of childhood trauma (Fink et al., Reference Fink, Bernstein, Handelsman, Foote and Lovejoy1995), and was originally designed to capture five broad trauma domains: emotional abuse, sexual abuse, physical abuse, emotional and physical neglect. In order to consider the cumulative effects of experiencing multiple traumas, each item was dichotomized as ever versus never experienced, consistent in several previous studies employing the CTQ to examine the onset of various psychopathologies (Gil et al., Reference Gil, Gama, de Jesus, Lobato, Zimmer and Belmonte-de-Abreu2009; Tucci et al., Reference Tucci, Kerr-Corrêa and Souza-Formigoni2010; Powers et al., Reference Powers, Fani, Cross, Ressler and Bradley2016). A full list of CTQ items can be found in online Supplementary Appendix 1. Respondents were asked to self-report how often a traumatic event occurred, from 1 (never) to 5 (often). Six items that refer to positive childhood experiences were reverse-coded so that a higher total score reflected a greater level of childhood trauma.
In addition, the CTQ includes a three-item minimization or denial scale used to assess respondents' potential for extreme reporting bias (Fink et al., Reference Fink, Bernstein, Handelsman, Foote and Lovejoy1995; Bernstein et al., Reference Bernstein, Stein, Newcomb, Walker, Pogge, Ahluvalia, Stokes, Handelsman, Medrano, Desmond and Zule2003; McGrath et al., Reference McGrath, Mitchell, Kim and Hough2010). However, there is disagreement regarding the validity of the scale as a measure of response bias (MacDonald et al., Reference MacDonald, Thomas, MacDonald and Sciolla2014), and it has been inconsistently used, including in several previous studies of South African samples (Kenny et al., Reference Kenny, Lennings and Nelson2007; Gerdner and Allgulander, Reference Gerdner and Allgulander2009; Lochner et al., Reference Lochner, Seedat, Allgulander, Kidd, Stein and Gerdner2010; Spies and Seedat, Reference Spies and Seedat2014; Spies et al., Reference Spies, Ahmed-Leitao, Fennema-Notestine, Cherner and Seedat2016). Therefore we did not remove respondents who met minimization/denial criteria.
Several covariates were considered to control for unmeasured confounding, based on previous studies of childhood trauma and schizophrenia (Powers et al., Reference Powers, Fani, Cross, Ressler and Bradley2016) as well as to remove the bias introduced by our matching strategy (Pearce, Reference Pearce2016). These included sex, continuous age (range: 21–54), education level (0–8th grade versus more than 8th grade), urbanicity (rural, urban, township), and region (Western Cape, Eastern Cape). A second set of models were run adjusting for additional covariates that represent potential comorbidities with schizophrenia, including lifetime depression (any v. none); lifetime alcohol use disorder (none v. abuse or dependence), and lifetime cannabis disorder (none v. abuse or dependence).
Those who were visually impaired or unable to read were assisted in completing the questionnaire by a nurse. All survey items were translated and back-translated into Xhosa in line with World Health Organization translation procedures (Menon et al., Reference Menon, Cherkil, Aswathy, Unnikrishnan and Rajani2012).
CTQ validation
There has been limited use of the CTQ in Xhosa-speaking samples and previous studies have not translated the instrument (Spies et al., Reference Spies, Ahmed-Leitao, Fennema-Notestine, Cherner and Seedat2016). Therefore, in a preliminary analysis, we examined validity and internal consistency of the translated instrument in our sample using factor analysis and ω coefficients (Dunn et al., Reference Dunn, Baguley and Brunsden2014). In the validation, an initial confirmatory factor analysis (CFA) was run to attempt to replicate the factor structure as originally designed. In the event of a poor CFA fit, we allowed for a newly derived factor structure, using exploratory factor analysis (EFA) in order to identify the best-fitting factor model of latent trauma domains. This exploratory approach was chosen over the use of CFA modification indices, to derive the most interpretable factor solution from the instrument. Factor loadings were estimated using weighted least squares means and variances, in order to handle categorical variables with non-normally distributed errors (Costello and Osborne, Reference Costello and Osborne2005). Goodness-of-fit was evaluated using the Root Mean Square Error of Approximation (RMSEA) and its 95% confidence interval (CI), and the Comparative Fit Index (CFI). Based on previous recommendations, we set cutoffs for acceptable fit at RMSEA < 0.10 and CFI > 0.90, and good fit at RMSEA < 0.05 and CFI > 0.95 (Hu and Bentler, Reference Hu and Bentler1999; Wall and Amemiya, Reference Wall and Amemiya2000). Additionally, in order to obtain an interpretable factor solution, we excluded items with low (<0.3) loadings on all factors, as well as statistically significant cross-loadings on multiple factors. Factor correlation was estimated using Pearson's r coefficient. Using the best-fitting factor solution, we then examined the internal consistency of each factor and the total CTQ scale with ω coefficients. Subsequent analyses were based on the best-fitting interpretable factor solution.
Data analysis
First, to better understand the degree of co-occurring traumatic events, we estimated the correlations between individual childhood trauma experiences. Second, we estimated the prevalence of individual traumas in the total sample, and compared the prevalence of each item among cases and controls using χ2 tests. Third, we estimated the prevalence of each trauma domain, dichotomized as ever versus never experienced, in the total sample and among cases and controls. Fourth, we estimated the odds ratios (ORs) of schizophrenia among those who experienced each trauma domain, compared with those who did not, in adjusted logistic regression models. Fifth, to examine the total burden of childhood traumatic experiences, we calculated the prevalence of experiencing cumulative individual traumas and estimated the odds of schizophrenia among those who experienced 1, 2, or 3 or more cumulative trauma domains compared with no trauma. Data were analyzed using SAS version 9.4 and Mplus version 7.
Ethical considerations
Ethical approval was granted by the Human Ethics Committee of the Faculty of Health Sciences, University of Cape Town, South Africa (HREC number: 049/2013) to conduct the study. Prior to data collection, our fieldworkers carefully explained in Xhosa the purpose and procedures of the study emphasizing that participation is entirely voluntary. He or she also explained that withdrawing from the study at any time is the participant's right and that data will be kept confidential by the study team. Individuals with schizophrenia may experience cognitive decline (Keefe and Harvey, Reference Keefe and Harvey2012; Bora, Reference Bora2015) so to ensure that informed consent was fully understood, all participants were screened for decisional capacity using the University of California, San Diego Brief Assessment of Capacity to Consent (UBACC) questionnaire (Seaman et al., Reference Seaman, Terhorst, Gentry, Hunsaker, Parker and Lingler2015; Campbell et al., Reference Campbell, Sibeko, Mall, Baldinger, Nagdee, Susser and Stein2017b). This is a well validated tool for screening for decisional capacity in participants with cognitive impairment.
Results
The initial EFA included the entire 25-item instrument. The model fit of a five-factor solution was good, however, there were several items with significant cross-loadings onto multiple factors. After removing four items, we identified a three-factor model with 21 items that evidenced good fit (CFI = 0.98; TLI = 0.97; RMSEA = 0.054), was interpretable, and was consistent with the original CTQ design. The ω coefficients for each factor domain indicated acceptable to excellent internal consistency. The three-factors included: (1) physical/emotional abuse (12 items; ω = 0.94); (2) neglect which comprises the lack of protective factors during childhood (4 items; ω = 0.85); and (3) sexual abuse (5 items; ω = 0.93). The total scale ω coefficient was 0.96 and the factor correlation coefficients ranged from 0.45 to 0.71. The final factor model is presented in online Supplementary Table S1. Together these findings suggest that the 21-item modified version of the CTQ has acceptable reliability and validity in this population.
The analytic sample comprised 88.1% men. The mean age of the sample was 36.1 years (s.d. = 9.13, range 21–54 years) and 71.9% had a minimum of primary school education. Overall, correlations between CTQ items were greater than 0.05 ranging from 0.16 to 0.70. The factor correlations were moderate, ranging from 0.21 to 0.42. All correlations are presented in online Supplementary Table S2.
The prevalence of individual childhood traumatic experiences overall and among cases and controls is presented in Table 1. Notably, the prevalence of experiencing any trauma was very high among cases (94.4%) as well as among controls (87.1%). All individual traumatic events were more prevalent among cases than controls. The prevalence of experiencing each trauma domain is presented in Table 2. Cases were significantly more likely to have experienced all trauma domains compared with controls. In the fully adjusted logistic regression models, the odds of schizophrenia was higher among those who experienced physical/emotional abuse (OR 1.63, 95% CI 1.28–2.1), neglect (OR 1.23, 95% CI 1.0–1.52) and sexual abuse (OR 1.22, 95% CI 1.0–1.48), compared with those who did not have these experiences. The odds of schizophrenia were 2.05 times higher among those who experienced any trauma than those who did not (95% CI 1.45–2.91). All model estimates are presented in Table 3.
Table 1. Prevalence of childhood traumatic events among those with and without schizophrenia
Table 2. Prevalence of childhood traumatic experiences by domain, among those with and without schizophrenia
Table 3. Unadjusted and adjusted ORs of schizophrenia, among those who ever experienced each traumatic event domain, compared with those who never experienced each domain
CI, confidence interval.
Note: Model 2 adjusted for sex, education, age, urbanicity, locality; Model 3 adjusted for model 2 covariates + lifetime depression, lifetime alcohol use disorder, and lifetime cannabis use disorder.
Figure 1 and online Supplementary Table S1 present the prevalence of cumulative traumas within each latent domain. A substantial proportion of the sample reported experiencing the maximum number of traumas within the domains of physical/emotional abuse (15.6%), neglect (26.7%), and sexual abuse (18.2%). In the fully adjusted logistic regression models, there was evidence of a dose–response relationship in the number of physical/emotional abuse experiences, increasing from 1.23 (95% CI 0.86–1.76) to 2.44 (95% CI 1.61–3.71). Those with two and three neglect-type traumas had 1.63 (95% CI 1.17–2.28) and 1.32 (95% CI 0.96–1.81) times the odds of schizophrenia, compared with those who experienced no neglect. There was no clear dose response for sexual abuse events. The odds of schizophrenia increased with increases in cumulative trauma experiences (range: 1.85–2.14). Compared with unimputed models, estimates were slightly attenuated after imputation, though were not substantively different. All model estimates are presented fully in Table 4.
Fig. 1. Cumulative proportion of traumatic events experienced for each domain.
Table 4. Adjusted ORs of schizophrenia, by a number of cumulative traumatic events experienced in each domain, compared with no traumatic experiences
CI, confidence interval.
Note: Model 2 adjusted for sex, education, age, urbanicity, locality; Model 3 adjusted for model 2 covariates + lifetime depression, lifetime alcohol use disorder, and lifetime cannabis use disorder.
Discussion
This study sought to examine the prevalence of childhood traumas, whether exposure to childhood trauma is associated with increased risk of schizophrenia, and whether that risk increased with cumulative trauma exposure in a South African, Xhosa-speaking population. First, our findings suggest that childhood traumatic events are very common among controls as well as cases. Second, all three latent domains of physical/emotional abuse, neglect, and sexual abuse are nonetheless associated with an increased risk of schizophrenia, most evident for those who had more than one traumatic event within a domain. Third, there is some evidence of a dose–response relationship between the number of physical/emotional abuse experiences and schizophrenia. There are several strengths to the methodology of the study, including the care taken to translate instruments into Xhosa.
Our findings that childhood traumatic events are high among cases and controls are consistent with several previous studies examining trauma and related conditions in South Africa (Jewkes et al., Reference Jewkes, Nduna, Jama-Shai, Chirwa and Dunkle2016; Gibbs et al., Reference Gibbs, Dunkle, Washington, Willan, Shai and Jewkes2018). Our results of the increased risk of schizophrenia among those who have experienced cumulative traumas are generally consistent with previous studies (Anda et al., Reference Anda, Felitti, Bremner, Walker, Whitfield, Perry, Dube and Giles2006; van Os et al., Reference van Os, Rutten and Poulton2008) including three systematic reviews of the relationship between childhood trauma and schizophrenia (Read et al., Reference Read, Van Os, Morrison and Ross2005; Matheson et al., Reference Matheson, Shepherd, Pinchbeck, Laurens and Carr2012; Varese et al., Reference Varese, Smeets, Drukker, Lieverse, Lataster, Viechtbauer, Read, Van Os and Bentall2012). We note that several studies of trauma in schizophrenia suggest that specific experiences of childhood trauma, e.g. physical or emotional neglect may have differential patterns of impact on various outcomes (i.e. premorbid adjustment, cognition and white matter abnormalities) (Kilian et al., Reference Kilian, Asmal, Chiliza, Olivier, Phahladira, Scheffler, Seedat, Marder, Green and Emsley2017a, Reference Kilian, Burns, Seedat, Asmal, Chiliza, du Plessis, Olivier, Kidd and Emsley2017b; Asmal et al., Reference Asmal, Kilian, du Plessis, Scheffler, Chiliza, Fouche, Seedat, Dazzan and Emsley2018). To our knowledge where previous studies have examined the relationship between childhood trauma and cognition in schizophrenia Kilian et al. (Reference Kilian, Burns, Seedat, Asmal, Chiliza, du Plessis, Olivier, Kidd and Emsley2017a), pre morbid adjustment (Kilian et al., Reference Koen, Niehaus, Wright, Warnich, de Jong, Emsley and Mall2017b) and white matter abnormalities (Asmal et al., Reference Asmal, Kilian, du Plessis, Scheffler, Chiliza, Fouche, Seedat, Dazzan and Emsley2018), no previous study has compared the relationship between childhood trauma and schizophrenia as outcome.
In modeling the exposure as cumulative traumatic events, we found evidence of potential threshold and dose–response effects. Experiencing a single physical/emotional abuse- and neglect-type event was not significantly associated with schizophrenia, and odds increased to significance among those who experienced 2 and 3+ events, in a dose–response manner for physical/emotional abuse. This may suggest that when experienced infrequently, children can overcome these forms of adversity without an increased risk of schizophrenia. In contrast, a single experience of sexual abuse was associated with elevated odds of disorder, emphasizing the severity of these particular experiences. Future research should further interrogate individual and accumulated traumas and the specific mechanisms underlying their potential differential effects.
These findings should be interpreted in light of several limitations. First, the case–control study design could be subject to recall bias where cases and controls could over or under report traumatic experiences. Also, the design limited our ability to determine the temporal ordering between each trauma and the onset of schizophrenia. It is plausible that traumatic experiences may have been a consequence of victimization after sub-clinical symptoms of schizophrenia (Beards et al., Reference Beards, Gayer-Anderson, Borges, Dewey, Fisher and Morgan2013; Tsigebrhan et al., Reference Tsigebrhan, Shibre, Medhin, Fekadu and Hanlon2014; Kraan et al., Reference Kraan, Velthorst, Smit, de Haan and van der Gaag2015), which might introduce bias due to reverse causation. Third, the psychological burden or social desirability of disclosing childhood trauma may have led to under-reporting of exposures (Susser and Widom, Reference Susser and Widom2012). This type of misclassification may be non-differential, or under-reporting may have been greater among the cases, though both types of misclassification would have biased the study estimates toward a null association, as found by Widom and colleagues in their 2015 study of intergenerational childhood trauma (Widom et al., Reference Widom, Czaja and DuMont2015). Fourth, the SAX survey recorded a limited set of covariates (sex, education, and age) which could be considered confounders in this study. As such, we consider the impact of unmeasured confounding as a source of bias. Fifth, we did not collect physician verified data of physical health conditions. Finally, the CTQ does not ask explicitly about the severity or duration of individual traumatic events. These characteristics should be considered in future research. We also suggest that future studies consider comorbid psychiatric conditions in schizophrenia, e.g. depression (Rössler et al., Reference Rössler, Joachim Salize, Van Os and Riecher-Rössler2005; Peralta and Cuesta, Reference Peralta and Cuesta2009; Herniman et al., Reference Herniman, Allott, Killackey, Hester and Cotton2017) or substance use as potential interaction variables of the relationship between childhood trauma and schizophrenia (Sideli et al., Reference Sideli, Fisher, Murray, Sallis, Russo, Stilo, Paparelli, Wiffen, O'Connor, Pintore, Ferraro, La Cascia, La Barbera, Morgan and Di Forti2018). Several studies suggest that childhood trauma is associated with mental disorders other than schizophrenia and further work is needed to determine whether there is specificity to these associations.
Conclusion
This study is a meaningful contribution to the literature examining childhood trauma in schizophrenia in low- and middle-income countries. The results strengthen the claim that childhood trauma is an important determinant for psychosis and schizophrenia and may be useful to consider when designing interventions to reduce psychosis incidence.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291719001703.
Acknowledgements
We wish to acknowledge the support of Dr Adele Pretorius, project manager of the SAX study and the data team for their assistance with data management, quality control and access.
Financial support
The SAX study receives funding from the National Institute of Mental Health (NIMH: Grant number 5UO1MH096754) and is a member of the Human Heredity and Health in Africa Consortium (H3 Africa) (http:h3africa.org/). SM has received support from the Harry Crossley and National Research Foundation of South Africa's post-doctoral research fellowships, a Claude Leon Foundation early career research award and a seed award from the School of Public Health, University of the Witwatersrand, Johannesburg, South Africa. SM has received funding from the Columbia University Southern African AIDS International Training/Research Program. JP has received support from National Institute of Mental Health (grant T32-MH1304346). DJS has received research grants and/or consultancy honoraria from Biocodex, Lundbeck, Servier, and Sun.
