The effectiveness of psychosocial interventions in war-traumatized refugee and internally displaced minors: systematic review and meta-analysis

ABSTRACT Background: The United Nations reported that in 2016 over 65 million people worldwide have forcibly left home. Over 50% are children and adolescents; a substantial number has been traumatized and displaced by war. Objective: To provide an overview of the effectiveness of psychosocial interventions in this group we conducted a narrative review and a meta-analysis of intervention studies providing data on posttraumatic stress symptoms (PTSS), depression, anxiety, grief, and general distress. Method: We searched PILOTS, MEDLINE, WoS, Embase, CENTRAL, LILACS, PsycINFO, ASSIA, CSA, and SA for studies on treatment outcomes for war-traumatized displaced children and adolescents. Between-group effect sizes (ES) and pre-post ES were reconstructed for each trial. Overall pre-post ES were calculated using a random effects model. Results: The narrative review covers 23 studies with a variety of treatments. Out of the 35 calculated between-group ES, only six were significant, all compared to untreated controls. Two of them indicated significant adverse effects on symptoms of general distress or depression. When calculating pre-post effect sizes, the positive between-group results of cognitive behavioural therapy (CBT) and interpersonal therapy (IPT) were reproduced and singular other treatments showed significant positive effects. However, the mean pre-post effects for PTSS and depression could not be interpreted due to the high heterogeneity of the included studies (PTSS: ES = 0.78; I2 = 88.6%; depression: ES = 0.35; I2 = 93.1%). Only the mean pre-post effect for seven active CBT treatment groups for depression (ES = 0.30, 95% CI [0.18, 0.43]) was interpretable (Q = 3.3, df = 6, p = .77). Conclusion: Given the large number of children and adolescents displaced by war there were regrettably few treatment studies available, and many of them were of low methodological quality. The effect sizes lagged behind the effects observed in traumatized minors in general, and often were small or non-significant. However, CBT and IPT showed promising results that need further replication.


Introduction
The last few years have seen a fourfold increase in mass displacement resulting from wars and conflicts worldwide. In 2016, over 65 million people worldwide were forced to leave their homes (UNHCR, 2017). Children below 18 years make up 51% of this population, a rate that has been rising constantly. Although forced displacement and its causes constitute sufficiently severe stressors to cause suffering in anyone, children and adolescents are particularly vulnerable. A substantial body of literature documents the accumulation of diverse traumata and psychosocial risk factors in displaced war-affected minors (Jensen & Shaw, 1993;Kletter et al., 2013), leading to large rates of psychological problems (Lustig et al., 2004) and subsequent complications (e.g. Vervliet, Lammertyn, Broekaert, & Derluyn, 2014). For example, several research groups report prevalence rates for mental health problems of up to 80% in unaccompanied asylum-seeking children, with posttraumatic stress disorder (PTSD), depression, and anxiety disorders (Bronstein, Montgomery, & Dobrowolski, 2012;Bronstein, Montgomery, & Ott, 2013;Huemer et al., 2009;Jakobsen, Demott, & Heir, 2014), as well as traumatic grief and conduct problems (Betancourt, Newnham, Layne, et al., 2012b), the latter being the most frequent diagnoses.
Overall, there is a consensus among health care professionals that these young people require particular assistance, but a number of obstacles hamper the delivery of psycho-social support: there are not enough trained psychotherapists available; access to effective treatment for refugees is limited for a number of reasons like geographical location or finances; the concept of psychotherapy and motivation for it differs across cultures; and there is no clear-cut recommendation on whether western evidence-based treatments are effective and applicable in this group.
Recent meta-analyses on PTSD treatment in children and adolescents after various kinds of trauma show converging evidence with overall effect sizes of g = 0.83 and 0.89 when compared to waitlist and of g = 0.41 and 0.45 when compared to active controls (Gutermann et al., 2016;Morina, Koerssen, & Pollet, 2016;respectively). Effect sizes for depression were 0.60 and for anxiety 0.67 when compared to waitlist, and 0.37 and 0.42 in comparison to active controls (Gutermann et al., 2016). Both meta-analyses, however, report considerable heterogeneity in their study, rendering the interpretation of the results difficult.
Contrary to the emerging evidence for children and adolescents with PTSD after various traumatic events, the evidence-based guidance for the treatment of war-affected displaced minors, though increasingly necessary, remains scarce.
Some authors provide a first overview of the field. Tyrer & Fazel (2014) reviewed health interventions for displaced children in school and community settings, identifying 21 studies. Due to the considerable variation in the delivered treatments, the authors refrained from calculating an overall effect size for controlled comparisons. Cohen's d were reported for two studies on depression (d = 0.57 and 0.93), two on anxiety (d = 0.64 and 0.93), three on PTSD-symptoms (d = 0.31-0.92), and six studies on other conditions like functional impairment and behavioural problems (d = 0.32-0.79), without specifying the respective control groups.
A recent meta-analysis on war-affected children and adolescents in low-and middle-income countries (Morina, Malek, Nickerson, & Bryant, 2017) identified 21 randomized controlled trials and reported pre-post effect sizes of g = 1.15 and a medium effect size of g = 0.53 compared to waitlist for posttraumatic stress symptoms. Effects on depressive symptoms were considerably lower, with g = 0.30 and g = 0.25, respectively. However, substantial heterogeneity impairs the interpretation of these overall effect sizes. The study sample in this publication consisted of children with mixed traumata, including a number of original studies on child soldiers who were perpetrators and victims at the same time. Their treatment might need to be specifically adapted (Betancourt, Newnham, Brennan, et al., 2012a, showed that treatment effects in abducted and non-abducted children were different) and should possibly include reconciliation.
Summarizing the above, previous reviews were based on samples of young people who were either affected by war or displaced, but not necessarily both. We aim to detect the efficacy of any putatively useful psychosocial interventions for forcedly displaced minors, as these children constitute a group with particular needs. To this end, we build on our previous work on displaced minors (Eberle-Sejari, Nocon, & Rosner, 2015), focus specifically on those who were traumatized by war, and expand on it by including a meta-analytic approach. To include all identified publications irrespective of their study design, we will also consider pre-post effect sizes.

Eligibility criteria
To detect any psychosocial intervention for the affected population, including primary prevention programmes, we chose broad search criteria: • Participants: refugees and internally displaced persons with direct war-related trauma exposure at the age of 18 years or under. Child soldiers were excluded. • Interventions: any intervention within the health care system intended to alleviate symptoms of trauma-related disorders. • Comparators: all comparators were eligible. • Outcomes: diagnosis and symptoms of PTSD, depression, anxiety, general distress, and complicated grief, as measured by some form of structured interview or questionnaire. • Study design: randomized controlled trials (RCT), controlled trials (CT), and uncontrolled pre-post studies. If they provided any information about negative effects, we aimed to include case studies and case reports in the narrative review. There were no language restrictions. We planned to identify all possible studies and search for translators in cases where no member of the research team could read the paper. We aimed to include all relevant studies, regardless of their publication status, to avoid publication bias. Unpublished and ongoing studies were sourced and included; to this end, potential authors were contacted. Conference abstracts were not included.

Literature search
We searched for published and unpublished studies in the following databases: Pilots, Medline (through the PubMed interface), Web of Science Core Collection, Embase, Central, Lilacs, PsycInfo, ASSIA, CSA, and SA (date of last search: 28 June 2017). Snowballing techniques included searching websites, journal hand searches, contacting authors and colleagues, and reference list checking. Searches were not restricted by language or publication date.
Keywords were child, adolescent, war, refugee, therapy, treatment, psychotherapy, treatment outcomes, and emotional trauma, according to the thesaurus and truncation options of the respective database.

Data collection
One author screened titles and abstracts (AN). Two other authors independently rated study characteristics (RES, JU). Each of them checked the other's work to ensure accuracy. Coded study characteristics were: (1) residence status (refugee, internally displaced); (2) group assignment (randomized, matched, convenience, no control group); (3) age range; (4) percentage of female participants; (5) exclusion of severe cases (suicidal, severe symptomatology, psychosis, language, other); (6) status of providers with regard to training; (7) type of intervention; (8) dose of intervention (in 50 minute sessions); (9) handling of drop-outs (intent to treat analysis, last observation carried forward, completer analysis); (10) number and duration of follow-up assessments; (11) control condition (waitlist/no treatment, unspecific treatment like support or counselling, psychotherapy). In cases where the raters were discordant on any characteristic, two other authors recoded the information (AN, RR). One author extracted data on statistical measures (AN). Primary outcomes were changes in symptom measures such as PTSD, depression, anxiety, general distress, or complicated grief.
As a major part of identified studies was of low study quality (no control group, no quantitative data), we indicated risk of bias only via the broad category of study design (RCT, CT, uncontrolled prepost study). We aimed to use all available studies at this very early stage of research in the field, and focused the meta-analysis on within-group treatment effects. To appraise the validity of the pre-post effects, we additionally considered unexplained drop-out rates and whether intent-to-treat analyses were used.

Statistics
For controlled trials, between-group effect sizes (standardized mean differences; SMD) were reconstructed as Cohen's d with pooled standard deviation. To detect all putatively positive effects regardless of study design, prepost effect sizes were reconstructed for all eligible treatment conditions (i.e. from control groups and uncontrolled studies also). In these cases, standardized mean changes (SMCs) were computed as raw score standardized mean changes (Morris & DeShon, 2002). In the paper by Onyut et al. (2005) the depression effect size was calculated using the marginal odds ratio (Zou, 2007), missing values were imputed using follow-up data on n = 6, and 0.5 was added to every cell of the contingency table to address the problem of empty cells (Cox, 1970). All effect sizes were corrected for sample size.
We used the random effects model for data synthesis as we did not expect all included studies to share one true effect size (samples may, for example, differ with regard to types of traumatization, age ranges, and treatments). I 2 was used as the measure for consistency and the Q-test for heterogeneity. To test for small study effects that can indicate publication bias we used regression tests (Egger, Smith, Schneider, & Minder, 1997). The effect of residence status as moderator was assessed using a mixed-effects model for subgroup comparisons (Borenstein, Hedges, Higgins, & Rothstein, 2009) and the Q-test for moderator variables was reported (QM). We refrained from further moderator analyses (e.g. regarding control conditions) as the small number of studies only allowed for moderator analysis with dichotomized moderators.
All statistics were computed using the statistics software R 3.2.2 (R Core Team, 2015) including the metafor package (Viechtbauer, 2010).
Nine studies were excluded although they initially fulfilled eligibility criteria. The pilot study of Sadeh and colleagues (2008), reported baseline data in Israeli children who were internally displaced during the second Lebanon-Israel war, with post-assessment after the end of war and the return home. It was excluded for methodological bias, as the end of the war may account for any effects detected in this study. Five publications were excluded because displaced persons represented a minority (up to 31%) in the study sample (Diab, Peltonen, Qouta, Palosaari, & Punamäki, 2015;Kangaslampi, Punamäki, Qouta, Diab, & Peltonen, 2016;Newnham et al., 2015;Punamäki, Peltonen, Diab, & Qouta, 2014;Tol et al., 2008). Two studies were excluded because only a minority (up to 35%) reported war-related traumata (Unterhitzenberger et al., 2015;Ruf et al., 2010). One study (Ispanovic-Radojkovic, 2003) was a double publication with Ispanovic-Radojkovic et al. (2002).

Study characteristics
All 23 identified publications were published in the last 15 years. The studies included children of diverse psychopathological status: children stemming from a waraffected area irrespective of their individual distress to children with full-blown PTSD diagnosis. Sample sizes ranged from N = 4 (Ooi, 2012) to N = 399 (Tol et al., 2012).
Eight studies used a randomized controlled design, six used non-randomized control groups, and eight were uncontrolled pre-post studies (see Table 1, where the studies are grouped accordingly). Among the eight RCTs, three studies did not provide full details of the randomization method (Kalantari, Yule, Dyregrov, Neshatdoost, & Ahmadi, 2012;Lange-Nielsen et al., 2012;Tol et al., 2012) and three used small experimental and control groups (n < 30) (Catani et al., 2009;Schauer, 2008;Schottelkorb, Doumas, & Garcia, 2012), which means that the randomization may not have been successful. However, no significant group differences on putatively relevant characteristics were detected in either of these studies. Among all studies with control groups, only five studies reported blinding of raters to participant group allocation (Betancourt et al., 2012a;Catani et al., 2009;Dybdahl, 2001;Ellis et al., 2013;Schauer, 2008). The success of the blinding procedures was not tested in any of these studies.
Two studies reported treatment fidelity problems. In the study of Schottelkorb and colleagues (2012), parents did not participate in the therapies as expected. In the study of Schauer (2008), trained teachers supported families beyond the scope of the manual.
Additionally, the following non-evidence based treatments were examined either in the experimental or the control condition: creative play (Betancourt et al., 2012a), child-centred play therapy (Schottelkorb et al., 2012), writing intervention Titles and abstracts identified and screened n=2151 Full copies retrieved and assessed for eligibility n=85 Excluded n=2051 Unable to obtain / further information required to make assessment n=15 Publications meeting inclusion criteria n=32 Excluded n=60 Adult sample n=2 No intervention n=5 No data n=17 Child soldiers/mixed with child soldiers n=3 Not displaced/no information whether displaced n=19 No war-related trauma n=7 No outcomes of interest n=4 Case study w/o adverse effects n=3 Excluded n=9 Methodical bias n=1 <50% with war-related trauma or displaced status n=7

Double publication n=1
Studies identified from contact with experts n=2 Studies identified from search in reference list n=3

Between-group effect sizes in studies using control groups
To enhance the comparability of the achieved treatment effects, we report between-group effect sizes according to broad categories of the used control conditions: no treatment/waitlist, unspecific treatment, psychotherapy. Table 1 summarizes the between-group effect sizes for PTSD and depression, together with their respective 95% confidence intervals. Two studies using CBT reported medium and large effects on PTSD compared to untreated controls (SMD = 0.88, Ehntholt et al., 2005; SMD = 0.37 in the group of girls, Tol et al., 2012; see Table 1). Any other effects on PTSD were null or not significant, that is, the confidence intervals included the zero. Regarding treatment effects on depression, IPT had large positive effects compared to waitlist controls, but only in girls (SMD = 1.06, Betancourt et al., 2012a), while writing for recovery had large adverse effects (SMD = −1.25, Lange-Nielsen et al., 2012) which had disappeared by follow-up. Any other reported effects were null or small, and not significant.
Therapeutic effects for traumatic grief, anxiety, and general distress are not presented in Table 1 and thus are presented here. The study of Kalantari and colleagues was the only study on traumatic grief symptoms (Kalantari et al., 2012). The authors used writing for recovery as treatment condition and reported a medium effect compared to untreated controls (SMD = 0.67, 95% CI [0.15, 1.19]). The study of Fazel et al. (2009) (Ehntholt et al., 2005), all controls were untreated, not one effect was significant. Ispanovic and colleagues reported only qualitative positive effects of the participation in psychosocial youth club activities on symptoms of anxiety (Išpanović-Radojković et al., 2002).
Altogether, 35 between-group effect sizes were calculated. Six were significant, four of them were positive and two negative. All significant effect sizes were achieved in group treatment settings. Due to the great variety of treatments and small numbers of studies in every group of control conditions, we did not further integrate the effect sizes.

Pre-post effect sizes in all active treatment groups
To detect any sign of an effective treatment, pre-post effect sizes were calculated for any experimental and control condition with an active treatment condition, irrespective of whether the treatment was psychotherapeutic or unspecific (see Table 2). Fourteen of 20 prepost effect sizes for PTSD were significant, ranging from small (SMC = 0.29, Lange-Nielsen et al., 2012) to large (SMC = 1.94, Onyut et al., 2005). Eight of the significant effects were achieved using CBT techniques, two in a meditation-relaxation condition (Catani et al., 2009;Schauer, 2008), one with a general education programme (Gupta & Zimmer, 2008), one with EMDR combined with psychodynamic therapy (Oras et al., 2004), one with a writing intervention (Lange-Nielsen et al., 2012), and a multilevel treatment particularly oriented to the needs of young refugees (Ellis et al., 2013). Seven of 19 calculated effect sizes for depression were significant, six of them showed positive effects between medium (SMC = 0.31, Tol et al., 2012, in the group of girls) and large (SMC = 1.98, Betancourt et al., 2012a, in the group of girls treated with IPT), with the       In the case of non-assessment cells are left empty. nr = not reported. $ post-assessment three months after the end of the intervention. § calculated from follow-up data. & group sizes unclear due to not indicated missings. # ES also computed for control group, as it was treated and assessed after the waiting period. ¥ ES are not based on post-assessment, but on group-differences in pre-to post-assessment changes. £ For the review and meta-analysis, n = 2 missing data at post-assessment were imputed using the follow-up data. ¶ In the study of Tol, the missing group pre-treatment variance was replaced using the variance for the total sample. clinically relevant results being achieved in three CBT conditions, two conditions that used IPT (Betancourt et al., 2012a), one using EMDR (Oras et al., 2004), and one using creative play (Betancourt et al., 2012a (Kalantari et al., 2012) (not reported in Table 2). Thirteen withingroup comparisons on general distress were calculated, with eight significant effect sizes, ranging from medium (SMC = 0.40, Dybdahl, 2001) to large (SMC = 1.11, Catani et al., 2009). Five conditions with significant effect sizes used CBT, one a combination psychosocial support and medical care of the mother (Dybdahl, 2001), one meditation-relaxation (Catani et al., 2009), and one was eclectic (O'Shea et al., 2000). Regarding effects on anxiety, two of five effect sizes were significant and medium to large, both in the study of Tol and colleagues (SMC = 0.79 in boys and SMC = 1.06 in girls, Tol et al., 2012, with CBT).

Discussion
In our review on treatments for displaced minors who were traumatized by war, we considered data from published and unpublished studies and did not exclude nonrandomized trials. Our goal was to include any possibly effective interventions at this early stage of treatment research. In spite of our very broad search criteria, we were able to find only 23 studies. The results demonstrate that treatment studies in displaced minors constitute a relatively young field of research. All identified studies were published in the last 15 years, with a substantial number of pilot studies and studies of poor methodological quality. Very diverse treatments were applied, most of them not evidence-based, and a substantial number of them did not yield significant effects. Before discussing this further we should consider the limitations of this paper.

Limitations
First, some of the original studies had methodological flaws which affected the reported statistics in diverse directions. Only a minority of studies used randomized controlled designs. Some authors reported serious problems in motivating the parents to participate, or changes in the residence status that prevented the children from attending all therapy sessions. Almost all studies relied on completer analyses, with sometimes high attrition rates and no reported reasons for dropout. These shortcomings, together with others like no blinding of raters, might have led to overestimation or underestimation of the true effect sizes.
Second, there are limitations on the level of the narrative review. Between-group effect sizes were not comparable, as even randomized controlled trials control for diverse variables which have a substantial impact on the treatment effect (e.g. gender).
Third, there are limitations on the level of the metaanalysis of the pre-post effect sizes. We included uncontrolled studies in order to detect any possibly effective treatment, and the computed SMCs certainly include time effects. So, the one non-heterogeneous overall effect for the treatment of depressive symptoms with CBT is probably overestimating the true effect.

Treatment effects
Most between-group effect sizes in war-affected displaced minors were not significant, some were negative, although several trials applied evidence-based psychotherapies. Concerning PTSD symptoms, only two studies using CBT detected a clinically relevant difference between the treatment group and untreated controls. The analysis of pre-post effects confirmed the positive effects for CBT, but there were some other treatments with positive pre-post effects that might be promising candidates for future investigation: EMDR, meditation-relaxation, an educational programme in schools as developed by the UNESCO, a stepped systemic treatment designed to the needs of young refugee youth, and writing for recovery. The positive CBT and EMDR effects are in line with the recent literature on youth from the general population (Gillies, Taylor, Gray, O'Brien, & D'Abrew, 2013;Gutermann et al., 2016). The mean pre-post effect size in our study cannot be meaningfully interpreted due to the high heterogeneity, which has also been reported in a number of other meta-analyses (e.g. Morina et al., 2017). In our view this shows how understudied this field of research is, as there are too few studies with obviously too different interventions available. Effect sizes in our sample were somewhat lower than effect sizes from the general population (Gillies et al., 2013: SMD = 1.34 compared to any control condition; Gutermann et al., 2016: SMC = 0.89 and SMD = 0.89 compared to untreated controls; Morina et al., 2016: SMD = 0.83 compared to waitlist controls), although they probably overestimated true treatment effects. They were also lower than those reported in children affected by war (Morina et al., 2017: SMC = 1.15). Part of the difference might be accounted by the small overlap between Morina et al. and our study. First, Morina and colleagues only included RCTs and the overlap with the eight RCTs in our study involves only three publications (Catani et al., 2009;Schauer, 2008;Tol et al., 2012). This is explained by differing inclusion criteria: Morina et al. included child soldiers, restricted their analysis on those who live in low-and middle-income countries, and did not include those who had to flee to a high-income country. However, some part of the difference in effect sizes might be no artefact: displaced children in need of psychosocial treatment might benefit substantially less from available treatments due to a number of factors, like loss of social support, and their specific living conditions might account in some part for the lower effects in our study. As we did not find any indication that residence status (refugees or internally displaced) had some impact on treatment benefit, this could hold true for both groups.
Our results for depression as outcome reveal a similar lack of evidence. Compared to untreated controls, only IPT was proven to be beneficiary out of all investigated treatments. Significant pre-post effect sizes were achieved using established interventions like CBT, IPT, and EMDR. This, again, is in line with the meta-analysis of Gillies et al. (2013), who showed that exposure-based interventions in children and adolescents from the general population had larger effects on depression symptoms than other psychological approaches. CBT treatments resulted in an overall small to medium prepost effect size for depressive symptoms (SMC = 0.30) in this group, which is lower than psychotherapy effects in general population minors with mixed traumata (Gillies et al., 2013: SMD = 0.80 compared to any control condition; Gutermann et al., 2016: SMC = 0.62;Morina et al., 2016: SMD = 0.30 compared to waitlist controls).
It is worth noting that in our review several significant effects resulted from treatments that were provided by trained lay persons (Betancourt et al., 2012a;Catani et al., 2009;Dybdahl, 2001;Gupta & Zimmer, 2008;Pfeiffer & Goldbeck, 2017;Schauer, 2008;Tol et al., 2012), demonstrating that substantial symptom reduction can be achieved by this means. On the other hand, some treatments delivered by clinicians were ineffective (Schottelkorb et al., 2012;Thabet et al., 2005) or even harmful (Lange-Nielsen et al., 2012). Some treatments were effectively administered in group settings (Betancourt et al., 2012a;Pfeiffer & Goldbeck, 2017) or at school (Gupta & Zimmer, 2008;Ooi et al., 2016;Schauer, 2008;Tol et al., 2012). Other reviews (Newman et al., 2014;Rolfsnes & Idsoe, 2011) have reported moderate (Rolfsnes & Idsoe, 2011) to large (Newman et al., 2014) effect sizes for trauma-related psychotherapies in the group/school setting, even if they were provided by trained lay persons like teachers or social workers. A substantial number of affected schoolaged refugee children could be reached this way, if effective interventions were available. The potential of dissemination of evidence-based treatments by training lay practitioners should be further investigated.
To summarize, we saw that therapeutic effects for war-affected displaced minors stay behind the expected range, which is especially discouraging given the fact that they often decrease on the long run, even one month post-treatment (Gillies et al., 2016(Gillies et al., , 2013Schauer, 2008). Additionally, an overall zero effect indicates that, while some subjects might benefit from a treatment, it probably has negative effects on others; a fact that has been discussed in the past (Ertl & Neuner, 2014;Tol et al., 2014). We think that displaced children constitute a particularly vulnerable group with specific challenges for therapists. Perhaps available interventions need to be adapted to the specific needs of this population and the specific context factors in this group, as even CBT-based interventions showed only moderate effect sizes. The population is certainly understudied, which is deplorable in light of their number. Hence, more large quality studies are urgently needed for concrete treatment recommendations.

Disclosure statement
No potential conflict of interest was reported by the authors.