Prevalence of mental disorders in refugees and asylum seekers: a systematic review and meta-analysis

Background Studies have identified high rates of mental disorders in refugees, but most used self-report measures of psychiatric symptoms. In this study, we examined the percentages of adult refugees and asylum seekers meeting diagnostic criteria for major depressive disorder (MDD), post-traumatic stress disorder, bipolar disorder (BPD), and psychosis. Methods A systematic literature search in three databases was conducted. We included studies examining the prevalence of MDD, post-traumatic stress disorder, BPD, and psychosis in adult refugees according to a clinical diagnosis. To estimate the pooled prevalence rates, we performed a meta-analysis using the Meta-prop package in Stata (PROSPERO: CRD42018111778). Results We identified 7048 records and 40 studies (11 053 participants) were included. The estimated pooled prevalence rates were 32% (95% CI 26–39%; I2 = 99%) for MDD, 31% (95% CI 25–38%; I2 = 99.5%) for post-traumatic stress disorder, 5% (95% CI 2–9%; I2 = 97.7%) for BPD, and 1% (95% CI 1–2%; I2 = 0.00%) for psychosis. Subgroup analyses showed significantly higher prevalence rates of MDD in studies conducted in low-middle income countries (47%; 95% CI 38–57%, p = 0.001) than high-income countries studies (28%; 95% CI 22–33%), and in studies which used the Mini-International Neuropsychiatric Interview (37%; 95% CI 28–46% p = 0.05) compared to other diagnostic interviews (26%; 95% CI 20–33%). Studies among convenience samples reported significant (p = 0.001) higher prevalence rates of MDD (35%; 95% CI 23–46%) and PTSD (34%; 95% CI 22–47%) than studies among probability-based samples (MDD: 30%; 95% CI 21–39%; PTSD: 28%; 95% 19–37%). Conclusions This meta-analysis has shown a markedly high prevalence of mental disorders among refugees. Our results underline the devastating effects of war and violence, and the necessity to provide mental health intervention to address mental disorders among refugees. The results should be cautiously interpreted due to the high heterogeneity.


Introduction
During the past decade, the trend of global displacement has been growing (UNHCR, 2020). In 2020, the United Nations High Commissioner for Refugees (UNCHR) estimated that more than 11 million people have been displaced throughout the year, and the proportion of this population has been continued to rise, despite the COVID-19 pandemic and closure of borders (UNHCR, 2020). Although refugees (people who, owing to a well-founded fear of persecution, are forced to escape from his or her country) and asylum seekers (people seeking protection from persecution or serious harm in a country other than their own) are defined in different ways (UNHCR, 2020), both groups may have been forced to face various stressors, such as persecution, violence, torture, detention, and the loss of homes and livelihoods. Such traumatic events may result in persistent mental health problems and overall decreased functioning (Steel et al., 2009;Marquez, 2016).
During the last years, several studies have examined the prevalence of mental disorders in refugees and asylum seekers. However, the resulting prevalence rates in this population present great variability (Fazel et al., 2005;Porter and Haslam, 2005;Bogic et al., 2015;Foo et al., 2018;Morina et al., 2018;Charlson et al., 2019;Blackmore et al., 2020), ranging from 5% (Fazel et al., 2005) to 80% (Bogic et al., 2015) for depression, from 4% (Charlson et al., 2019) to 88% (Morina et al., 2018) for PTSD, and from 1.5% (Blackmore et al., 2020) to 2% (Fazel et al., 2005) for psychosis. A systematic review on the prevalence of serious mental disorders in refugees resettled in high-income western countries reported weighted average rates of 5% for major depressive disorder (MDD), 9% for posttraumatic stress disorder (PTSD) and 2% for psychotic disorders (Fazel et al., 2005). A meta-analysis among refugees and asylum seekers residing outside their country of origin reported a prevalence of 31.5% for PTSD, of 31.5% for depression, and 1.5% for psychosis (Blackmore et al., 2020). Another meta-analysis that examined the prevalence of mental disorders in conflict settings showed estimated rates of 22.1% for any mental disorder (13% for depression and 4% for PTSD) (Charlson et al., 2019). The variability across studies could be explained by differences in the origins and background of the analysed populations, the sample size, the sampling methods, and the diagnostic tools used (e.g. self-report, semi-structured or structured clinical interview) (Westermeyer and Janca, 1997;de Jong et al., 2003;Fazel et al., 2005;Bogic et al., 2015;Giacco et al., 2018;Giacco and Priebe, 2018;Charlson et al., 2019). Furthermore, the nature of the displacement, the context of the emergency, as well as the aspects of the host country/environment are additional aspects to consider in the variability of these studies (Giacco and Priebe, 2018).
Many epidemiological studies use self-report instruments to estimate the prevalence of mental disorders in refugees because they are easily administered and cost effective. However, such instruments overestimate prevalence rates considerably (Domken et al., 1994;Fazel et al., 2005;Steel et al., 2009). In fact, self-report instruments have been shown to overestimate true PTSD-rates by a factor of about 3.5 (Engelhard et al., 2007). Similar results have been also shown in depression studies (Domken et al., 1994;Steel et al., 2009;Krebber et al., 2014). One of the reasons is that many self-report instruments are designed for quickly picking-up mental disorders while minimising false negatives, therefore cut-offs are set with high sensitivity. Therefore, they identify more patients who may have a mental disorder than those diagnosed with clinician-administered interviews . Another reason could be that selfreport tools usually do not take functional impairment due to symptoms into account (McKnight and Kashdan, 2009). Besides, translations of existing self-report instruments may not adequately measure psychiatric symptoms across cultures (Hunt and Bhopal, 2004). Finally, overestimation may occur as a consequence of the so-called over-endorsement bias, the tendency of respondents to generously claim different types of symptoms on a checklist (Kroenke, 2001). Thus, more rigorous instruments, such as clinical diagnostic interviews, are more appropriate to estimate the prevalence of mental health disorders in refugees and asylum seekers (Fazel et al., 2005;Steel et al., 2009). In addition, clinicians' experience in using structured interviews can increase the reliability of symptoms measurement and psychiatric diagnoses even among samples with different cultural backgrounds (Alarcon et al., 1999;Aboraya et al., 2006). Nevertheless, among the structured and semi-structured interviews there may also be some differences in accuracy. For example, studies have shown that the Mini-International Neuropsychiatric Interview -MINI, which has been designed as a briefer and quicker diagnostic screen than other interviews (Sheehan et al., 1997) identified more people as depressed than the Composite International Diagnostic Interview -CIDI, and the Structured Clinical Interview -SCID Levis et al., 2019;Wu et al., 2021).
In the present study, we aimed to estimate the prevalence of diagnoses of serious mental disorders [MDD, PTSD, bipolar disorder (BPD), and psychosis] in refugees and asylum seekers from conflict-affected areas. Serious mental disorders are conditions resulting in serious functional impairment, which interferes with major life activities (NIH, 2019). Other definitions (Slade et al., 1997) take into account the duration and the disability they produce, for example in terms of the disability they are associated with. Within the Global Burden of Disease (GBD) study, the top-three mental disorders associated with the highest level of disability are MDD, BPD and psychosis (Kronenberg et al., 2017;James et al., 2018). Although not included in the GBD, PTSD is particularly relevant for refugees and asylum seekers since refugees may be exposed to multiple traumatic events, including interpersonal violence that may cause severe disability (Palic et al., 2016).
Similar to previous studies (Fazel et al., 2005;Blackmore et al., 2020), we included only studies that employed diagnostic interviews. Further, our study differs from previous studies in that we were able to include a larger number of studies than before since the number of studies increased during the past years. In addition, we were able to examine the prevalence of less common disorders such as BPD. Finally, due to the increased number of included studies, we could perform additional subgroup analyses, such as comparing the prevalence of serious mental disorders between refugees resettled in low-middle-income v. high-income countries, and among different kind of population sample (convenience samples v. probability-based samples).

Search strategy and selection criteria
We report our meta-analysis according to Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement (www.prisma-statement.org) (Moher et al., 2009) (see online Supplementary materials).
The present study was registered in PROSPERO on 5 th December 2018 under the number: CRD42018111778.
A comprehensive search was performed in the bibliographic databases PubMed, Embase.com and APA PsycInfo (via Ebsco) from inception to June 4, 2020, by a medical librarian. Search terms included controlled terms (MeSH in PubMed, Emtree in Embase and PsycINFO thesaurus terms) as well as free text terms. The following terms were used (including synonyms and closely related words) as index terms or free-text words: 'refugees' and 'mental illnesses'. A search filter was used to limit the results to adults. The search was performed without date or language restrictions. Duplicate articles were excluded. The full search strategies for all databases can be found in the online Supplementary materials.
Studies were considered eligible for this meta-analysis if they examined (a) the prevalence of serious mental disorders (MDD, PTSD, BPD, and psychosis), (b) in an adult (⩾18 years) population of refugees and asylum seekers who had to cross their country borders, (c) according to the criteria of Diagnostic and Statistical Manual of Mental Disorders (DSM III, IV, or 5) or International Classification of Diseases (ICD 9 or10) (d) assessed with a structured or semi-structured clinical interview, such as the Structured Clinical Interview -SCID, Mini-International Neuropsychiatric Interview -MINI, or Composite International Diagnostic Interview -CIDI. No cut-off has been considered to evaluate the diagnosis' severity. Disorders have been evaluated serious according with the National Institute of Mental Health (NIH) and GBD considerations (Kronenberg et al., 2017;James et al., 2018;NIH, 2019).

Global Mental Health
The definitions of refugees (a person who, owing to a wellfounded fear of persecution, is forced to escape from his or her country) and asylum seekers (a person who seeks protection from persecution or serious harm in a country other than their own) were in agreement with the UNHCR Master Glossary of Terms (UNHCR, 2006) and Asylum and Migration Glossary 6.0 (Network, 2018). No restrictions were applied to the origin of the studies. Cross-sectional studies, follow-up studies, cohort data studies and register-based studies were eligible. To reduce possible selection bias, we excluded studies conducted on samples selected in psychiatric services and concerning internally displaced people (IDPs) (UNHCR, 2006).
All title and abstracts were screened by two researchers independently (MP/MS or SG). We retrieved the full texts of all abstracts that seemed eligible for inclusion. Subsequently, we performed a full-text selection according to our pre-specified eligibility criteria. Disagreements between the reviewers were resolved by discussion.

Data extraction
For every eligible study, two reviewers independently (MP/MS or SG) extracted data related to the year of publication, the country of data collection, the country of the refugees' origins, the sample size of the population, the age, the gender, the time elapsed as refugees or asylum seekers, the diagnostic tool used. Finally, we extracted the prevalence rates of MDD (current episode and recurrent), PTSD, BPD and psychotic disorder. In the absence of absolute numbers and if allowed by the data reported, the percentages of prevalence rates were converted into numbers. The time since displacement was reported in three different categories: (1) people who have spent more than five years as refugees or asylum seekers (⩾5), (2) people who have spent between 5 years and 1 year as refugees or asylum seekers (>1 < 5) and (3) people who have spent less then 1 year as refugees or asylum seekers. Countries were categorised as low-, middle-(LMIC) and in highincome (HIC) based on the World Bank list of economies 2020 (WBC, 2020).

Quality assessment
Two reviewers (MP/MS or SG) evaluated the assessment of the methodological quality of individual studies using a modified version of the Joanna Briggs tool -JBI Critical Appraisal Checklist for Studies Reporting Prevalence Data (Munn et al., 2015) based on five questions, which permitted a critical assessment of prevalence rates: (1) Were study participants recruited in an appropriate way? (2) Was the sample size adequate? (3) Were the study subjects and the setting described in detail? (4) Was the condition measured in a standard, reliable way for all participants? (5) Was the response rate adequate, and if not, was the low response rate managed appropriately?. The response categories per each question were yes, unclear, and no. Three or more unclear or negative answers were considered to define a study with a high risk of bias (see online Supplementary materials). Any disagreements were resolved through discussion or by involving a third reviewer (PC).

Data synthesis and statistical analysis
The prevalence rates of serious mental disorders were calculated by pooling the study-specific estimates. To stabilise the variance of binomial data, the Freeman-Tukey double arcsine transformation was used (Miller, 1978). Because we expected considerable heterogeneity between studies, we computed the pooled estimates under the random effects (DerSimonian and Laird) model based on the transformed values and their variance (Nyaga et al., 2014). The I 2 was calculated as an indicator of the inter-study heterogeneity in percentages. Values of I 2 are considered low, moderate and high when I 2 equals 25, 50, and 75% respectively (Melsen et al., 2014). Based on previous meta-analyses on prevalence rates, we expected considerable heterogeneity in the prevalence estimates (Higgins, 2008;Charlson et al., 2019;Blackmore et al., 2020). To examine possible sources of heterogeneity between the included studies, we ran subgroups under the mixed-effects model with inverse-variance weights under the random-effects model (time since displacement ⩾ 5 years v. time since displacement >1 < 5years v. time since displacement <1year, MINI v. other diagnostic interviews, LMICs v. HICs, high risk of bias v. low risk of bias, convenience samples v. probability-based samples). Prevalence rates were calculated per each disorder separately. Pooled rates for subgroups were indicated, when at least three studies were present for each disorder. All statistical analyses were conducted using Meta-prop package in STATA/SE 16.1 for Mac (Nyaga et al., 2014).

Results
We initially identified 11 749 records. After exclusion of duplicates, 7048 records remained. We excluded 6705 studies based on titles and abstracts selection. We examined 343 full texts against our eligibility criteria and included 40 studies in the present meta-analysis (Fig. 1).
To reduce the high heterogeneity and confirm our prevalence rates, we also ran analyses without the outliers when it was possible. The analysis showed a prevalence of 33% of MDD and 32% of PTSD (Table 2 and online Supplementary materials).

Discussion
This systematic review and meta-analysis provide the most recent and extensive overview of the prevalence rates of serious mental disorders across refugees and asylum-seeking populations. A comprehensive search was performed from inception to June 4, 2020 and was not intentionally updated to avoid interference caused by the effect of COVID-19 pandemic in this population. We included 40 studies in 11 053 participants, and all diagnoses were established with diagnostic interviews. Our study revealed that the most prevalent serious mental disorder in refugees and asylum seekers was MDD (32%), followed by PTSD (31%), recurrent episode of MDD (16%), and BPD (5%). The prevalence of psychotic disorders was 1%. Subgroup analyses showed that MDD appeared to be more prevalent (47%) among studies conducted in LMICs than in HICs (28%), and when the MINI (37%) has been used, compared to other diagnostic instruments (26%). PTSD and MDD showed higher prevalence rates (34% and 35% respectively) in studies where participants were from convenience samples in comparison to studies that used probability-based samples. PTSD and MDD were the most frequently evaluated disorders with 36 studies and 32 studies respectively, as opposed to BPD and psychosis of which we had only a few studies. According to previous studies, the world-wide prevalence rate reported for MDD is 4.4% (WHO, 2017), the lifetime prevalence of PTSD in World Mental Health Surveys has been calculated between 3.9% and 5.6% (Koenen et al., 2017) and the prevalence rates of psychosis and BPD in the general population are 0.4% (Moreno-Kustner et al., 2018) and 2.5%, respectively (Carta and Angst, 2016). Comparing our results with the prevalence of the same disorders in the general population, suggests that MDD is seven times more likely in refugees and PTSD is 4 to 5 times more prevalent than in the general population (Koenen et al., 2017;WHO, 2017). Although BPD and psychosis are much rarer than MDD or PTSD, our study also indicated that externally displaced refugees and asylum seekers are two times more likely to be diagnosed with BPD and with psychosis than the general population (Carta and Angst, 2016;Moreno-Kustner et al., 2018). Studies have highlighted the role of trauma and stress experienced by minorities, social defeat, and discrimination as important risk factors for psychosis in refugees (Brandt et al., 2019;Duggal et al., 2020;Selten et al., 2020).
Our results are in line with the prevalence rates of the Blackmore and colleagues (Blackmore et al., 2020) study, who found a prevalence rates of 31.4% (95% CI 24.43-38.5) for PTSD, of 31.5% (95% CI 22.64-40.38) for depression, and 1.5% (95% CI 0.63-2.40) for psychosis. In addition, our study reports the random-effects pooled prevalence for BPD which has never examined before. Compared to the 2005 systematic review findings (Fazel et al., 2005), our rates prevalence showed an increase of serious mental disorders prevalence among refugees and asylum seekers from 5% (Fazel et al., 2005) to 32% for MDD, and from 9% (Fazel et al., 2005) to 31% for PTSD. One of the reasons for the discrepancy between these results could be the LMICs exclusion in Fazel and colleagues' systematic review. Furthermore, it might be explained by increased exposure to adverse events, increased financial hardship, social isolation, decreased access to adequate health care, as well as the lack of appropriate policies and investments in the last 15 years.
Our study methodology has several strengths, among which strict inclusion and exclusion criteria, a large number of included studies, and updated statistical methods. However, due to the low number of studies that examine refugees and asylum seekers separately, a limitation of this study was to consider these two groups together. Further, we found a large variability of prevalence rates and a high level of heterogeneity between studies. While high heterogeneity is very common in meta-analyses on prevalence rates (Higgins, 2008), we examined possible sources of heterogeneity between the included studies. Running the analyses without outliers did not alter the main results. Upon inspection of the pattern of outliers, no systematic differences between outlier studies from the rest of the studies were identified that might explain the high heterogeneity. However, due to the limit number of studies greater caution is required in interpreting the prevalence rates of BP and psychosis. In addition, we analysed subgroups under the mixed-effects model with inverse-variance weights under the random-effects model. The results of these subgroup analyses showed that significantly higher prevalence rates of MDD and PTSD in studies conducted among convenience samples than in studies that used probability-based samples. This result highlights the importance of having an adequate statistically representative sample of participants to avoid overestimating prevalence rates in a population as various and difficult to study as that of refugees and asylum seekers. Furthermore, higher prevalence rates of MDD were found in LMICs than in HICs. This difference was not found for PTSD, where the prevalence rate was comparable between refugees and asylum seekers resettled in LMICs and HICs. The higher prevalence of MDD in refugees resettled in LMICs may be driven by higher exposure to post-migration living difficulties. Feelings of hopelessness, the failure of the migration project, and difficulties in integration may promote higher levels of depression (Steel et al., 2009;Charlson et al., 2019). Further, refugees resettled in LMICs can have a higher risk of developing MDD because of the lack of integration programs and mental health care due to the low investments in mental health care that unfortunately characterises the majority of these countries (Patel, 2007). On the other hand, compared to MDD, PTSD may have a stronger association with trauma exposure in the country of origin, which may not differ significantly between refugees in LMICs and HICs.
Another interesting finding was that we found a higher rate for MDD for studies conducted with the MINI than in studies that used different diagnostic interviews. This may be explained by the fact that the MINI may be administered by non-clinicians (similar to the CIDI, but dissimilar to the SCID I/P), and that its administration is shorter and its outcomes therefore potentially less precise.
Despite the high heterogeneity and methodological limitations of included studies, a high prevalence of serious mental disorders was found in refugees and asylum seekers diagnosed through structured clinical interviews. However, specific instruments culturally adapted for use across different local cultures and contexts for refugees would be advisable, since no one of the diagnostic instruments currently in use has been developed for the nonwestern populations, who represent the highest sample of refugees. For this reason, to reduce the high heterogeneity, more rigorous studies, using representative samples and culturally adapted instruments, to measure cultural concepts of distress would be recommended. Moreover, to strengthen the evidence base concerning the more serious mental disorders, further research on the prevalence of psychosis and BPD is imperative. To follow this goal, international and government investments in mental health research, population screening, and specific interventions on asylum seekers and refugees are warranted. However, until we have more rigorous studies and adequate diagnostic tools, due to the high heterogeneity between these studies, our results should be considered with caution.

Conclusion
In sum, our systematic review and meta-analysis show that it is imperative that governments and actors across the world acknowledge the devastating effect of war and prosecution on individuals' mental health. In order to prevent cycles of violence and victimisation, effective public mental health responses may be put in place to address worldwide suffering.

Data
All the data involved have been included in Tables and Figures of