Predictors of psychological risk and resilience among Syrian refugee children

Study aims

Our aim was to first identify the proportion of children with low risk for mental health problems despite adversity (i.e. displaying manifested resilience) in a sample of Syrian refugees in informal tented settlements (ITS) in Lebanon, and investigate what differentiated those meeting our resilience criteria from others exposed to the same war events but doing poorly at the time of assessment. A better understanding of these factors is important in order to promote resilience and prevent mental health problems through intervention. Based on the literature (ICED, 2019), we expected that approximately 50% of children would show evidence of manifested resilience and that both individual and social factors would differentiate children with low and high risk for mental health problems.

Study design

We used cross-sectional data from a large sample of Syrian refugee child–caregiver dyads from the BIOPATH cohort study (McEwen et al., 2022). We applied clinically validated cut-offs to measures of PTSD, depression, and externalising behaviour problems and used them to categorise children into low (LS) and high (HS) symptom groups. LS and HS children were paired according to their specific war exposure, age, gender, and time since leaving Syria to create a balanced subsample of cases with high and low symptoms of psychopathology but equal reported war exposure. We then investigated the relationships between potential risk and resilience factors and mental health outcomes in the matched subsample.

Ethical approval was granted by the Institutional Review Board of the University of Balamand, Lebanon (ref: IRB/O/024-16/1815). The study was also reviewed by the Lebanese National Consultative Committee on Ethics and approved by the Ministry of Public Health.

Setting and participants

Data collection was conducted in ITSs in the Beqaa region of Lebanon in 2017/2018. We used purposive cluster sampling, approaching settlements representing a range of vulnerabilities according to the UNHCR vulnerability index (UNICEF, UNHCR, & WFP, 2017). After receiving agreement from the community leader of each settlement, we invited one child per family from all eligible families to participate. Families were eligible if they had left Syria no more than 4 years ago, the child was aged 8–16 years, and a primary caregiver was available to participate. Where multiple children in one family were eligible, the child with the birthday closest to the recruitment date was selected, to avoid biased selection. Informed consent and assent were given by each caregiver and child, respectively, and families were compensated for their time. Questionnaire data were collected via interviews in the settlements by a team of 16 interviewers to reduce barriers to participation. Visual aids were used to help understanding (Figure S1). Interviews lasted approximately 50–60 min. In order to focus on responses to war, we excluded children with no reported war exposure (n = 49). For a more detailed explanation of recruitment, see McEwen et al. (2022).

Variables

All participants were interviewed in their homes by trained Arabic-speaking interviewers. Different interviewers conducted the child and caregiver interviews simultaneously. Some measures were exclusively child or caregiver reported, while others were reported by both (Table S1).

Exposure to war

War exposure was measured with the War Events Questionnaire (WEQ), a 25-item checklist measuring the number of different types of war event experienced (Karam, Al-Atrash, Saliba, Melhem, & Howard, 1999). Because self-report may be less reliable in younger children (Oh et al., 2018), child and caregiver responses were combined such that if either one reported that the child experienced an event, the event was considered to have occurred. For the specific items in the war questionnaire see Figure 1.

Statistical methods

Analyses were conducted in RStudio Version 1.1.383. Multiple imputation using Fully Conditional Specification (FCS) in the mice package (van Buuren & Groothuis-Oudshoorn, 2011) was applied to impute the small number of missing data (<5%). We imputed all missing measures for the analysis, except demographic variables, war exposure, and child mental health. We ran all analyses in both the imputed and original (complete case) datasets and report the pooled imputation estimates in the main text of this paper (complete case analyses are provided in Table S5). The Benjamini–Hochberg procedure was applied to all analyses in order to correct for multiple testing (Benjamini & Hochberg, 1995).

Descriptive data

We collected data from 1,600 families. Excluding seven repeat participants, two families later identified as ineligible, 14 families missing key demographic or symptom data, and 49 children with no reported war exposure, the final sample included 1,528 children (52.6% female; mean age = 11.48, SD = 2.43) and their primary caregivers (89.5% mother). According to the combined child and caregiver report, children experienced an average of 9.90 (SD = 5.34) war events. The majority of children had experienced a bombardment-related event, such as witnessing explosions (84%), but many had also experienced more direct interpersonal violence. For example, 36.6% reported witnessing torture and 44.4% reported that someone close to them was killed.

Main results

When LS and HS participants were matched on war exposure, several factors predicted group membership (Figure 2; Table S4). In individual models, self-esteem, optimism, and perceived social support increased the odds of classification into the LS group. Poor child general health, environmental sensitivity, support-seeking, problem-focused coping, avoidance-based coping, child maltreatment, bullying, loneliness, maternal psychological control, and caregiver general and mental health all decreased the odds a child belonged to the LS group.

The logistic regression model combining all individually significant predictors explained 20.6% of the variance. Self-esteem (OR = 1.38, 95% CI [1.17, 1.62]), optimism (OR = 1.07, 95% CI [1.01, 1.13]), the temperament trait of environmental sensitivity (OR = 0.83, 95% CI [0.72, 0.95]), loneliness and social isolation (OR = 0.90, 95% CI [0.86, 0.95]), child maltreatment (OR = 0.97, 95% CI [0.96, 0.99]), caregiver PTSD (OR = 0.99, 95% CI [0.98, 1.00]), and caregiver depression (OR = 0.96, 95% CI [0.93, 0.99]) survived correction for multiple testing in the combined model. Bivariate correlations between the predictors included in the combined model ranged from small to moderate (r = −.28–.65; Table S3).

The same analysis in the complete case data (n = 542) largely supported the main results, with some minor differences (Table S5). Support seeking was not predictive of group membership in the complete case data, whereas maternal acceptance was predictive. Fewer predictors remained significant in the combined model in the complete case data, likely due to lower power, but maternal psychological control did remain significant, unlike the imputed results. However, the direction of associations remained the same and effect sizes were similar.

The continuous sensitivity analysis looking at predictors of the continuous mental health composite score (n = 1,528) supported the main results, indicating that our matching technique did not introduce bias (Figure S6; Table S6). Every significant predictor in the main logistic regression remained so in the linear regression.

Resilience

Due to the high levels of adversity faced by our sample, children who scored below the cut-offs of all three outcome measures can be considered as demonstrating resilience in the context of war exposure. The relatively small number of resilient children in our sample compared to studies focusing on single outcomes (Khamis, 2019b) emphasises the importance of applying a multi-dimensional approach. The absence of symptoms on one dimension does not mean that a child is doing well overall. However, our sample also shows higher prevalence of mental health problems than some studies that did account for multiple dimensions (Çeri, Nasıroğlu, Ceri, & Çetin, 2018; ICED, 2019). This may be due to challenges specific to our sample, who continue to experience on-going adversity as refugees in Lebanon, with difficulty accessing basic necessities, services and work (McEwen et al., 2022). Worse mental health might therefore be expected compared to refugee populations resettled in more stable and less adverse conditions (Fazel, Reed, Panter-Brick, & Stein, 2012). However, it is important to note that children scoring above cut-offs, that we considered as high risk for mental health problems, may still recover over time (Masten & Narayan, 2012).

Predictors of risk and resilience

Several factors differentiated children with high and low risk for mental health problems. At the individual level, optimism and self-esteem predicted LS group membership, whereas poor general child health and the temperament trait of environmental sensitivity predicted HS group membership. More surprisingly, coping strategies (support-seeking, problem-focused, and avoidant coping) predicted HS group membership. While avoidant coping is generally maladaptive, support-seeking and problem-focused coping are considered positive coping strategies (Scharpf et al., 2021). However, the specific context and nature of the adversity the current sample were exposed to likely play an important role here; problem-focused strategies may be maladaptive when the problem is too complex to fix (Elklit, Østergård Kjær, Lasgaard, & Palic, 2012). Heightened reporting of coping strategies could also indicate increased need for those strategies, due to worse mental health.

At the social level, negative aspects of social relationships (bullying, loneliness/social isolation, child maltreatment, maternal psychological control) and caregiver mental and general health problems were predictive of HS group membership, in keeping with previous research (Karam et al., 2019; Khamis, 2019a; Scharpf et al., 2021). The key implication is that risk factors in the immediate social environment seem particularly important for children's mental health, compared to aspects of the living environment and socio-economic factors. The nonsignificant findings for some aspects, such as access to education, are unexpected, and may also be due to the specific characteristics of the sample (e.g. few children go to school, so there may not be enough variance in the sample to see significant associations).

Although we have identified a series of predictors that are important in determining which children may be more able to adapt to war exposure in the particular context faced by the current sample, we must consider that these predictors likely interact with one another. The combined model results and bivariate correlations suggest that some predictors share variance, but most correlated predictors retain distinct predictive power even when combined. For example, caregiver depression is correlated with maltreatment, which could indicate a mediating pathway whereby caregiver mental health affects children via parenting (Bryant et al., 2018). However, caregiver depression and child maltreatment both remain significant in the combined model, indicating that they have separate effects on child mental health. On the other hand, individual predictors that are not significant in the combined model indicate potential shared effects on child mental health. For example, the correlations between the social relationship measures could explain why the effect of bullying does not survive in the combined model. The specific experience of bullying may not have additional effects on children when controlling for maltreatment in the home, and social isolation, one possible effect of bullying.

Our findings were supported by the logistic regression in the complete case data, and the linear regression in the total sample, but there were some small differences to note. Support seeking was non-significant and maternal acceptance was significant in the complete case analysis, but effect sizes were almost identical to the imputed estimates. Additionally, fewer predictors remained significant in the complete case combined model, likely due to decreased power from the smaller sample size. However, maternal psychological control remained significant in the combined model in the complete case but not the imputed data. This could be explained by associations between predictors; maternal psychological control is correlated, and probably co-occurs with, other social risk factors, such as child maltreatment, as well as with child self-esteem, which is likely negatively impacted by controlling caregiver actions. The additional children in the imputed data may have increased the power to detect distinct effects beyond shared variance, but may also have tipped the balance as to exactly which predictors were significant in that specific model given these relationships. The individual predictor models indicate a number of measures likely key to our sample's adaptation to war exposure, but the complex interrelationships between predictors require further exploration.

Generalisability

One key strength of this study is the unique sample of vulnerable Syrian refugee children living in Lebanon, a middle income country bordering Syria, with an extremely high proportion of refugees (UNHCR, 2020). As the majority of refugees worldwide reside in low and middle income countries (UNHCR, 2020) our findings are generalisable to a large proportion of the global refugee population.

Limitations

Our data have certain limitations. First, cross-sectional data precludes interpretations of the directionality of associations and the more complex dynamic processes of adaptation. Consequently, outcomes and predictors are difficult to separate cross-sectionally. For example, while self-esteem can be a stable trait (Brent Donnellan, Kenny, Trzesniewski, Lucas, & Conger, 2012), it could also reflect current mental well-being. Moreover, those classified as HS may be able to successfully adapt in future, while the resilient sample will not necessarily be impervious to future challenges. Second, our measures were reported rather than observed, and some, such as war exposure, were retrospective, introducing potential recall bias confounded by participants' current state, in addition to challenges regarding reliability when collecting data from children (Panter-Brick, Grimon, Kalin, & Eggerman, 2015). However, research on refugee war experiences is necessarily retrospective, and we included caregiver report of several key scales including war exposure to improve reliability (Oh et al., 2018). Third, we were limited by evaluating mental health with established symptom scales, some of which had to be translated, and reported by different informants. However, scales were extensively piloted and the most reliable selected, and, where possible, modified to be appropriate for the dialect, literacy and context of the families. Furthermore, we derived cut-offs through clinical assessment in a subsample, choosing cut-offs with the best balance of sensitivity and specificity for our particular sample (McEwen et al., 2020). It is important to note that, despite this, specificity fell below 80% and the risk group will contain false positives, meaning that the true proportion of resilient children is likely to be higher. Fourth, individual effects of some predictors were small, which should be considered in interpretation. Fifth, the complete case combined model results differed slightly from those in the imputed data. However, according to further checks, imputation did not significantly change patterns of responding, and the effect sizes remained very similar. Finally, a potential selection bias in recruitment cannot be excluded due to restricted access to certain settlements and reliance on presence of families during recruitment. Notwithstanding these limitations, our results provide an important step to understanding resilience in a highly vulnerable child population with considerable barriers to research access.