Mental health of people detained within the justice system in Africa: systematic review and meta-analysis

Worldwide, people with mental disorders are detained within the justice system at higher rates than the general population and often suffer human rights abuses. This review sought to understand the state of knowledge on the mental health of people detained in the justice system in Africa, including epidemiology, conditions of detention, and interventions. We included all primary research studies examining mental disorders or mental health policy related to detention within the justice system in Africa. 80 met inclusion criteria. 67% were prevalence studies and meta-analysis of these studies revealed pooled prevalence as follows: substance use 38% (95% CI 26–50%), mood disorders 22% (95% CI 16–28%), and psychotic disorders 33% (95% CI 28–37%). There were only three studies of interventions. Studies examined prisons (46%), forensic hospital settings (37%), youth institutions (13%), or the health system (4%). In 36% of studies, the majority of participants had not been convicted of a crime. Given the high heterogeneity in subpopulations identified in this review, future research should examine context and population-specific interventions for people with mental disorders. Electronic supplementary material The online version of this article (10.1186/s13033-019-0273-z) contains supplementary material, which is available to authorized users.


Introduction
Investments in mental health care and research are critical: globally, psychiatric and neurological disorders comprise approximately 13% of the disability-adjusted life years and nearly one-third of years lived with disability [1]. The prevalence of mental illness is higher among people detained within the justice system (PDJS) compared to the general population [2][3][4][5][6][7][8][9]. In this review, PDJS are defined as people detained or incarcerated in prisons, jails, youth institutions, or forensic inpatient units of hospitals. In accordance with the World Health Organization (WHO)'s definition, a forensic inpatient unit is "exclusively maintained for the evaluation or treatment of people with mental disorders who are involved with the justice system. These units can be located in mental hospitals, general hospitals, or elsewhere. " [10]. This review uses an inclusive definition of detention to respond to the scarcity of research on PDJS's mental health in Africa, particularly for people detained outside of prisons. International systematic reviews on the mental health of PDJS show that populations in prisons are multiple times more likely to have several major mental disorders [2,4,5,9] and have a three to sixfold higher risk of death by suicide [3]. More than 10 million people are held in penal institutions worldwide [11], and have increased risk of adverse outcomes such as all-cause mortality, suicide, self-harm, violence, and victimization [3]. The prevalence of mental disorders in African countries is of particular concern due to resource constraints for mental health and expansion of the movement to shift resources from institutions to community-based care in low-and middle-income countries (LMICs) [12,13]. In 2017, there were 2.5 total mental health beds per 100,000 population in African countries as a whole, 80% of which were in psychiatric hospitals [10], illustrating a contrast between where most existing mental health resources go (hospitals) and where services may be needed and, in reality, delivered (the community). There is, however, an international movement, including

Open Access
International Journal of Mental Health Systems *Correspondence: kkidia@gmail.com 3 13:31 in Africa, to shift resources from a country's psychiatric hospital or hospitals to other forms of mental health services [14,15]. This movement recognizes that most people in low-resource settings have been receiving care in the community (if at all). However, in contexts with inadequate community health care, people previously in institutions may face increased risk of diversion to the justice system [15][16][17][18]. As health resources shift to better support mental health care in primary care and outpatient services in some African countries, research is needed on mental health interventions for justice-involved populations with elevated prevalence rates. However, there has been minimal attention to the mental health of PDJS in international data collection and guidelines [19]. We surveyed United Nations (UN) and WHO guidelines on mental health and detained populations from the past 15 years ( Table 1) and found that the mental health of PDJS has not been present in the majority of publications. The exclusion of the justice system from research or policy priorities contrasts to consensus in international prison literature which explicitly requires extensive mental health care services (see Table 1).
Despite the burden of mental disorders among PDJS in Africa, research in this area is sparse [3,19]. Existing international systematic reviews on the mental health of PDJS have either included studies in only one African country, Nigeria [2], or none at all [4,5,20,21]. Moreover, these reviews do not report methodological bias or ethics procedures data of included studies, and the search criteria of most do not include institutions that detain justiceinvolved youth or inpatients in forensic psychiatry units [2,4,5,21], even though this is where many people with mental illness may be detained and are similar to prisons in some countries [22]. A recent systematic review on the influence of prison climate on mental health resulted in studies from only high-income countries [21]. Similarly, a review of psychological therapies for PDJS internationally did not result in studies from LMIC countries other than India, Iran, and China [20]. Importantly, however, its search criteria included youth and participants in secure hospitals, while the other systematic reviews include only those in prisons and jails [2,4,5,21].
Given the paucity of research focused on Africa and the resource constraints of these countries, this review aims to understand the scope of knowledge on the mental health of PDJS in Africa and to identify gaps in the literature, in order to inform future research, interventions, harm reduction, efforts, and policy. Given our study aims, we did not limit our search and selection to a single study design or outcome measure. In contrast to previous reviews investigating only prevalence, specific study types, or restricted to prisons, the broad scope of this review responds to the scarcity of research in African contexts and the systems-wide nature of detention and issues surrounding the mental health of PDJS.

Methods
The review followed guidelines of the Preferred Reporting Items for Systematic Review and Meta-analyses (PRISMA) [23], and the meta-analysis of observational studies in epidemiology (MOOSE) [24], both of which are found in Additional file 1: Appendices S1 and S2. The protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO). The Registration Number is CRD42018098852.

Search strategy and selection criteria
We searched PubMed, PsycINFO, Embase, Web of Science, and Africa Index Medicus for studies dating from the inception of the database to 16 November 2017, and published in English or French, major research languages on the African continent. Because the inception of databases are constantly updated as new articles are digitized, searches were run without date limits. These databases represent major features of this paper: international health databases (PubMed, Embase, Web of Science), a psychiatry-specific database (PsycINFO), and an Africa-specific database (African Index Medicus). Africa Index Medicus is managed by the WHO, and is one of two major African health databases. The other is African Journals Online, which is not health specific. Our search strategy included terms related to the following ideas: (1) Africa (defined as the WHO Africa Region [25]), (2) detention or incarceration, and (3) mental health. Search syntax and controlled vocabulary words were modified for each database. Complete search terms for each database are in Additional file 1: Appendix S3.
Only primary research studies published in peer reviewed journals in English or French that met the following PICOS criteria were included. Participants included any people detained by the state, regardless of whether they had been sentenced to a crime. We excluded people formerly detained if data was collected after their release, and excluded people detained for immigration (similar to prior research on detained populations) [3]. We included studies that related to a mental illness as defined according to the DSM-V, but included studies that used other diagnostic criteria. Due to the scarcity of systematic research in this area, we included both qualitative and quantitative studies, including systematically collected data on policy, health systems, and conditions of detention, such as availability of physical or human resources, services, and the legal process. We included studies in the WHO Africa Region [25] and in any setting in which people were involuntarily held by the government for being accused or convicted of Report  committing a crime. We excluded the settings of house arrest or non-forensic psychiatric hospitals in which people are involuntarily committed for disease alone.

Data collection
After the search was conducted and duplicates removed, two researchers (AL, HK-trained undergraduate students) separately screened the titles, abstracts, and full texts of studies to assess whether they met the inclusion criteria, reconciling differences between each step through discussion. A third reviewer (HJ) acted as a tie breaker if the other screeners could not come to consensus. Following the full text screen, a backward search was conducted on all articles selected for inclusion. The results of the backward search were screened for inclusion following the same protocol as the initial search results (Fig. 1). If a paper identified could not be located using multiple university library systems, we attempted to contact the study author. If the author could not be contacted, we excluded this paper as it was not possible to access the full text. We did not contact authors to obtain missing data or details on methods.

Risk of bias assessment
Risk of bias was assessed using the following separate tools for each study type. Pre-post studies: Quality assessment tool for quantitative studies [26]; RCTs: Cochrane Collaboration's tool for assessing risk of bias [27]; qualitative: Critical Appraisal Skills Program (CASP) Checklist for Qualitative Studies [28]; prevalence: Prevalence Critical Appraisal Checklist [29]. We did not assess risk of bias for validation studies or structured reviews of health systems. The small number of these studies included in the review and the methodologic heterogeneity of structured reviews or descriptions of health systems make them difficult to compare with a single tool. We did not report a single overall risk of bias score for each study, as these scores typically involve arbitrarily weighting different domains of risk of bias [30]. Rather, we reported grouped studies into categories of high, medium, and low risk of bias based on the overall assessment generated by the methods assessment instruments. The methodological assessment was used to exclude studies that had either (1) insufficient information on methods or (2) a clear methodological flaw or inconsistency.

Data extraction
Data on study characteristics, sample characteristics, and study outcomes were extracted from each study (Tables 2, 3). Systematic reviews have the potential to elevate studies with poor ethics standards [31]; therefore, we extracted data on the presence or absence of documentation of an ethics committee review and informed consent procedures. If institution conditions or the laws or policy of the health or justice system were described, we extracted these descriptions and categorized the data into common findings (Additional file 1: Appendix S9).

Data analysis
Because of the heterogeneity of the study designs and outcomes, we conducted narrative analysis, as an overall meta-analysis was not possible. To facilitate comparison between like studies, we structured our presentation of results by study design.

Meta-analysis
The large number of prevalence studies enabled us to conduct a meta-analysis of the prevalence of mental disorders. Notably, this analysis was added post hoc, as we did not anticipate sufficient study design homogeneity for meta-analysis at the outset of the study. We generated pooled prevalence estimates with 95% confidence intervals [32] for key disease categories: mental ill health (including measures of psychological distress or an unspecified mental disorder, often assessed using a screening tool, such as the General Health Questionnaire (GHQ) [33], but excluding results of disorder-specific instruments), mood disorders, psychotic disorders, and substance use. Heterogeneity between studies within each category was assessed using Chi squared and I 2 (> 50% is considered heterogeneous [34]). A random effects model was used to estimate the pooled prevalence, as all groups demonstrated significant heterogeneity. The random effects model weights included studies to account for both sample size and between-study variance with the between-study variance term dominating the weighting when studies are heterogeneous [35], as it assumes that studies come from different distributions. In order to identify sources of heterogeneity, we conducted subgroup analysis, grouping according to youth and adults; then, within studies of adults, location (prison or forensic ward) and data collection method (instrument or clinical record). In the subgroup analysis, we determined the pooled prevalence for each subgroup and its heterogeneity. Statistical analysis was done using STATA/ SE 15.0 [36]. Additional details of the data extraction and statistical analysis are shown in Additional file 1: Appendices S4, S6 and S7.

Search results
After removing duplicates, our search yielded 1240 results in the initial database search, of which 73 met inclusion criteria. We excluded three studies based on the methods screen results and added eight from a backward search. We were not able to access the texts or abstracts of three studies from the backward search title results because they were not available for loan from multiple university library systems. A third researcher (HJ) reviewed 39 full texts about which the other two reviewers could not reach agreement and chose to include 12 of them (31%). We added two articles identified through expert consultation. This yielded 80 papers for data extraction (Fig. 1), including 17 papers that were written on the same sample and as part of the same study as another paper in our dataset, but with different methods details or outcomes reported. In this review, the terms "independent studies" or "samples" refer to the number of independent studies, and the term "papers" refers to all papers in the dataset. There were 70 independent studies in our data. All papers are described in Tables 2 and 3.

Study characteristics
Prevalence was the dominant study type (67%), with small numbers of other study types including ten nonprevalence cross-sectional studies, four qualitative studies, four tool validations, two structured health systems reviews, two pre-post studies, and one randomized control trial. The majority of studies took place in either Nigeria (30; 43%) or South Africa (21; 30%). Only five of 23 non-prevalence studies took place outside of Nigeria or South Africa: two health systems studies in Zimbabwe, one qualitative study in Zambia, one cross-sectional study in Rwanda, and one international comparative study. Sixty-nine percent of studies collected primary data, or data resulting from diagnostic or screening tools, and 37% collected secondary data, defined as data extracted from available records. Four studies collected both primary and secondary data. Thirty-two studies (46%) examined prison settings, of which all but one study collected primary data. Twenty-six studies (37%) examined forensic hospital settings (included only if they were settings in which a justice-involved population was detained). Among these, 88% were based on secondary data, and the majority took place in either South Africa (54%), Nigeria (15%), or Zimbabwe (12%). Nine studies (13%) examined youth institutions, of which all were conducted in Nigeria. Three studies (4%), two conducted in Zimbabwe and one international, examined the mental health system.

Risk of bias
Most of the papers assessed fell into the low (36; 49%) or medium (33; 45%) methods risk of bias categories. A number of medium risk papers were given this judgement based on lack of detail in methods reporting, meaning that it was unclear if the studies had high risk of bias or if methods were not reported well. About half of the medium risk papers reported diagnoses from secondary psychiatric records (17), for which it was often unclear how the diagnoses were made, who made them, or if they were based on standard criteria, such as the ICD. Of the 57 psychiatric prevalence papers, 44% (25) fell into low risk, 47% (27) fell into medium, and 9% (5) fell into high risk of bias categories. All "high risk" studies were prevalence papers. See Additional file 1: Appendix S5 for results of the methodologic review of prevalence studies, which includes evaluation of sampling technique using the Prevalence Critical Appraisal Checklist.

Participants
In 69% of samples, more than 85% of participants were male. In 10 samples, "male" was an inclusion criterion. The mean age of participants among samples listing mean age was 28.8, and there was a broad distribution of sample size, from 18% of participant samples with 50 people or less (including all three interventions), to 13% with over 400 participants. In 36% of samples, the majority of participants had not been convicted of any crime (the majority of participants were either awaiting trial or detained without trial). In another 41% of samples, trial status was not stated or unclear, or participants were justice-involved youth (in which trial status was ambiguous). In only 20% of samples, the majority of participants had been convicted.

Ethics characteristics
Of all papers, 35% neither documented ethics committee approval nor described an informed consent procedure, and 41% described both. Ten percent reported ethics committee approval but not informed consent, and 14% reported informed consent but not ethics committee approval.

Outcomes
To facilitate comparison between like outcomes, we have presented study outcomes by study design: prevalence, non-prevalence cross-sectional, intervention testing, qualitative, structured health system review, and tool validation. Of the 57 prevalence papers reporting mental health diagnoses or screening results, 44 (77%) reported diagnoses of psychiatric conditions, of which 14 papers also used instruments designed to screen for psychiatric Lovett                         morbidity but not diagnose. Thirteen papers (23%) reported outcomes obtained with screening tools alone, not diagnostic instruments. Reported factors associated with a psychiatric outcome and secondary outcomes are displayed in Additional file 1: Appendices S10 and S11.

Cross-sectional studies
Nine cross-sectional studies did not collect psychiatric prevalence data, but investigated variables associated with mental health conditions, or justice or health system qualities. These studies examined associations between Fig. 3 Mood disorders. The x-axis in each plot displays prevalence (0-1). The far right column displays the prevalence within each study, pooled prevalence for each subgroup, and overall pooled prevalence with their respective weights. Uche and Princewell "Prevalence…" (2015) [103] and Uche and Princewell "Clinical factors…" (2015) [102] describe the same study and sample and report the same prevalence. Only one point prevalence from this sample was included from these articles mental health-related variables (for instance, PTSD symptom severity [37]; substance misuse [38,39]) and other factors (crimes committed [37]; prevalence of sexually transmitted infections [38]; emotional intelligence [40]). Alternatively, large and Nielssen found a positive correlation between per capita prison populations and per capita psychiatric hospital beds among LMICs and a combined pool of 158 countries, but no significant correlation among high-income countries [41].

Interventions
The intervention studies included two pre-post groupfocused cognitive behavioral interventions in prisons, one for cigarette smoking dependence and one for depression, both yielding significant improvements in the treatment group compared to the control group (p < 0.001) [42,43]. The single randomized control trial demonstrated significant decreases in schizophrenia symptoms following injections of two different neuroleptics in separate treatment arms (each neuroleptic treatment group resulted in a p < 0.01 decrease in the combined schizophrenia symptom score compared to the pre-injection score, and the Flupenthixol group showed a larger decrease than the Clopnethixol group (p < 0.01)) [44]. Additional details of the intervention studies are listed in Additional file 1: Appendix S8.

Qualitative studies
The outcomes of the four qualitative studies included a shortage of medical personnel in prison mental health services in South Africa [45], poor prison conditions linked to mental health problems in a Zambian prison [46], psychiatric findings among women in prison for homicide [47], and gaps in awareness of the legal process and other legal characteristics of participants referred for psychiatric observation [48].

Structured health systems reviews
The two structured health systems reviews, both in Zimbabwe, each interviewed around 30 participants. One used an exploratory qualitative design, including interviews with people detained in the justice system, and proposed a model for transforming the medico-judicial system that involves multiple stages of mental health screening and diversion from the justice system [49]. The other used a structured needs assessment and interviews with policy makers, administrators, providers, and researchers to examine the national mental health system and found that many stakeholders called attention to the forensic mental health system although the researchers did not specifically ask about forensic mental health [22].

Descriptions of conditions of detention, policy, law, and health systems
We collected descriptions of mental health policy, laws, and systems if systematically collected and reported in any study type. Thirteen papers described conditions of detention in institutions, laws, or policy, including eight prevalence studies, three qualitative studies, and two structured health systems reviews. While it was not always clear whether the methodology used to report  such outcomes was rigorous, we collected this data because of the scarcity of existing literature on this topic and report common findings thematically in Additional file 1: Appendix S9. The most common findings were insufficient human resources for health; lack of psychosocial services; lack of timely psychiatric assessment; and limited rehabilitation, recreational, vocational, or community re-integration services. Other common themes included insufficient physical resources, food, or psychiatric medicines; delays in trials, case-processing, or release; and lack of communication between medical and justice systems.

Discussion
To our knowledge, this is the first systematic review to investigate the mental health of PDJS in Africa exclusively. Results reflect that existing studies on this topic are predominantly prevalence studies that show a high pooled prevalence of mental illness, consistent with previous findings on mental illness in detained populations globally. Notably, many people detained within the justice system in non-prison locations, such as youth institutions or forensic hospitals, were detained with no charge [6,7,[50][51][52][53][54][55]. The neglect of these populations in the literature is especially alarming in the context of pressures to deinstitutionalize mental healthcare in LMICs [15,[56][57][58][59]. A number of key populations were missing from our results. There were very few women included in the included studies, reflective of the small proportion of women found in prisons worldwide, particularly Africa, where only 3% of the total prison population are female, much lower than elsewhere [60]. Reasons for this may be that worldwide, likely due to distinct social roles, women commit fewer crimes [61] and that women are less likely to be convicted of crimes and sent to prison by courts [62]. The studies were also concentrated in a small number of African countries (73% of studies were in South Africa or Nigeria), and we found no studies in 36 of the 47 WHO-defined African countries. While more research is needed on translations of interventions across low-resource settings, we urgently need ground-work research in local contexts. Surprisingly, the general population detained in prisons was poorly represented in our sample, as prison studies were concentrated around people with particular psychiatric or forensic variables such as type of crime or trial status.
Our meta-analysis of prevalence studies revealed high pooled prevalence of mental disorders and substance use among PDJS in Africa, which underscores the urgency of addressing the mental health of detained people in Africa within the global mental health movement. However, the studies were heterogeneous. While we attempted to explore and explain the heterogeneity using subgroup analysis, nearly all subgroups were also heterogeneous. The most valid finding of the meta-analysis is that we have statistically shown that the studies were conducted on distinct populations. This is not surprising given that the populations are from across Africa and are detained in a variety of different types of institutions and under distinct penal policies. Accordingly, it is important to interpret the meta-analysis findings cautiously as the heterogeneity may limit the validity of providing a single point estimate for distinct populations. Notably, the level of heterogeneity we found was consistent with prior systematic reviews on the prevalence of mental disorders in prison settings [2]. All of this underscores the need for more research on mental health in prison settings and better standardization of the tools used to assess mental and substance use disorders.
While recognizing the heterogeneity, prevalence estimates are consistently higher than those estimated in a prior systematic review on mental illness in prison, which found a 3.6% prevalence of psychosis among men and 3.9% in women, and a 10.2% prevalence of major depression in men and 14.1% in women [2]. This prior review, however, included only one study from Africa, where resources for psychiatric care are particularly limited, and it did not include forensic wards. When we examined the prevalence of psychotic disorders in prisons alone (forensic psychiatry units excluded), the prevalence of psychotic disorders (1%; 95% CI 0.00-0.02) was much more similar to the prior review, but the prevalence of mood disorders (33%; 95% CI 0.20-0.46) remained higher. Although the prior review estimated the prevalence of major depression alone while this review includes bipolar disorder, almost all studies included in our estimate of mood disorders in prisons measured depression exclusively, and there was low prevalence of bipolar disorder. The prevalence of substance use disorder (35%; 95% CI 0.25-0.44) was similarly high as observed in a prior international systematic review on substance use and dependence in prison, which found prevalence for alcohol use and dependence ranged from 18 to 30% among men and 10 to 24% among women, while prevalence of drug abuse and dependence varied from 10 to 48% in male prisoners and 30 to 60% in female prisoners [4]. There were too few women included in our study results to warrant meta-analysis by gender; however, the gender differences observed in prior research highlight the need for intentional data collection on gender in future work.
Subgroup analysis highlighted that there are differences in prevalence between institutions, with higher prevalence of mood disorders in prisons, higher prevalence of psychotic disorders in forensic institutions, and very low prevalence of psychotic disorders in prisons. This finding suggests that people with severe psychotic illnesses are being tracked out of prisons and into forensic units, which theoretically provide more psychiatric treatment. Interestingly, however, the prevalence of any sort of mental disorder in forensic units was less than 100%, indicating that people without psychiatric diagnoses are being detained in these units in some settings. Additionally, there was relatively high prevalence of substance use across facilities, highlighting the importance of considering substance use when developing interventions in all settings of detention. The prevalence of any mental disorder in youth institutions had particularly high variability, perhaps because of the small number of studies and wide variety in reasons for youth detention. The findings of the meta-analysis, however, must be interpreted cautiously because the samples are highly heterogeneous, likely because these samples are drawn from different countries, institutions, and cultural contexts, and the outcomes are measured in a variety of ways with tools that may not be adequately validated in their settings. The heterogeneity is expected in a field with such paucity of data and is consistent with other meta-analyses of mental disorders in prisons [2].
Systematic reviews have the potential to include, and even elevate, ethically questionable studies due to the nature of an exhaustive search [31,63]. This may be particularly problematic for research involving vulnerable populations such as PDJS. We chose not to exclude any studies based on their ethical practices because excluding studies potentially detracts from knowledge that should eventually be used to help this vulnerable population. Thirty-five percent of studies failed to document ethics committee approval or informed consent procedures. Not all journals require ethics reporting, so our results do not necessarily indicate whether ethics procedures actually occurred, but raise concerns. The use of ethics assessment protocols in systematic reviews [31] and greater standardization for ethics reporting across journals could help ensure that all relevant ethical procedures are described.
The three effective interventions studies identified in the review may provide a starting point for future development of interventions and could be adapted for other settings. For instance, two interventions-one for depression [42], one for substance dependence [43]-used cognitive behavioral therapy approaches in group-focused sessions held twice a week. This group-focused, cognitive behavioral design could be adapted in future psychosocial interventions. The other intervention, a RCT for neuroleptics used to treat schizophrenia symptoms [44], highlights the strengths of pharmacological interventions for severe mental illness.

Implications for health workers and policymakers
First, we were surprised at how many studies included people who had not been convicted. Considerable work at all levels is needed to ensure that people with mental illness have rapid access to trial and are not detained without due process. Second, the high prevalence of mental disorders demonstrated across all settings of the justice system highlights the need for all staff members to receive mental health training. Likewise, those working in mental health should receive training on managing the needs of people who have been involved with the justice system. However, interventions must be coordinated and services designed to support individuals as they move across health, justice, and social systems. For instance, Munetz and Griffin propose the sequential intercept model for diversion of people with mental illness from the justice system in which various points of contact with the justice system become opportunities for connecting individuals to services; i.e. interceptions to prevent further justice involvement [3,64]. Third, the prevalence of substance use disorders was particularly high. Systematic reviews and RCTs from HICs highlight the importance of medication-assisted interventions, which combine psychological and pharmacological treatment, for substance use disorders in PDJS [3,65,66].

Directions for future research
First, the studies included have made it clear that there is high psychiatric prevalence among PDJS, but little development or testing of interventions to address the large mental health need. Contextual research on each system of detention and subpopulation, including youth, and older participants, trial or charge status, gender, and diagnosis, is essential to inform such interventions. In addition to providing mental health services within institutions of detention, the high prevalence of mental illness compared to the general populations suggests a need for research on supportive diversions of individuals with mental illness from the justice system to best connect individuals with care. Second, we have focused on the academic literature, but there should also be systematic and comprehensive data collection and analysis of institutional and governmental documents surrounding the mental health of PDJS in Africa. Third, there is need for longitudinal studies that follow participants to community re-entry, especially investigating transitions to community health care and long-term health and recidivism outcomes [67,68]. Fourth, only five studies in this review collected qualitative data from PDJS themselves. Studies must do more to include service user voices. Fifth, prisons, youth institutions, and forensic psychiatry settings must be investigated both discretely and as facilities that feed into each other: each settings' population has specific needs, but there may be revolving door effect for forensic psychiatric units and prisons, as described in Zimbabwe [49]. Sixth, economics research was not present in our results, but is needed to measure the societal costs of lack of treatment, the cost-effectiveness of treatments, and the potential cost-savings provided by interventions and diversions.

Limitations
First, a major limitation for this review was the heterogeneity of studies and overall paucity of high quality literature. We chose broader definitions of detention and mental illness to provide a comprehensive perspective on this understudied area and better define the state of the field in Africa. However, because results were more variable than expected, we were unable to analyze specific aspects or subpopulations (interventions, forensic units, youth institutions, policy) in as much depth as we would have liked. We aim to delve deeper in the future, producing discrete publications on forensic hospitals, youth institutions, conditions of detention, but reported all results of this paper collectively in accordance with PRISMA guidelines [23], and to avoid legal, ethical, and methodological issues that arise from post hoc changes to the protocol and attempts to publish slices or versions of data that have been published previously [69]. This heterogeneity in included studies makes it challenging to point to singular conclusions from the data. However, the heterogeneity of study settings is a strength since exhaustive inclusion allows us to speak to the state of the issue in Africa at large and gives readers a more comprehensive overview of key areas for future work. This approach calls attention to the systems-wide nature of detention of people with mental illness, whereas prior reviews have excluded non-prison populations with high psychiatric prevalence that are similarly detained by the state. Second, because we added the meta-analysis post hoc, it was not included in our original protocol. We did not select studies with the aim of having a homogenous set of outcomes or study designs to facilitate meta-analysis. Additionally, the disease categories we used to group analysis were generated inductively based on the included studies. They were broad and likely introduced additional heterogeneity into the groups. However, we believe that metaanalysis provides one additional way for readers to access and understand our data and triangulates the results of the narrative review, which were consistent with the meta-analysis findings. Third, we discovered eight studies during the backward search: a signal that perhaps the original search strategy was imperfect. Many of these studies were from small journals that were not indexed in the major databases that we had originally chosen to search. While we could have iteratively changed the search strategy to broaden the list of databases we searched, we chose to search only large databases of established quality and to adhere to our study protocol. Finally, a unifying theme throughout our review and in evaluating our limitations has been a data quality issue, as many studies had high risk of bias or reported so little about their methods that it was difficult to assess risk of bias.

Conclusion
This review has identified key areas that require further research, and demonstrated need for more standardized methods and ethics reporting. It has confirmed the high prevalence of mental illness among PDJS in Africa, but revealed an absence of setting diversity or diversity of study types, and revealed key populations underresearched or missing from the literature. Though the need for bio-medically focused interventions is clear from this high psychiatric prevalence, we look forward to a future in which prevention approaches and social interventions are prioritized. Social factors of stress, poverty, and discrimination may disproportionately affect people that become detained and contribute to poor mental health. Future mental health research must take on a systems-wide perspective involving both the health and justice sectors, and investigate both clinical and contextual social variables. This approach will guide interventions for coordinated service development, and better align policy with the aim of the Sustainable Development Goals' to leave no one behind in achieving equitable, universal health coverage [70,71].