A network analysis of postmigration living difficulties in refugees and asylum seekers.

Background: The prevalence of mental disorders among asylum seekers and refugees is elevated compared to the general population. The importance of post-migration living difficulties (PMLDs), stressors faced after displacement, has recently been recognized due to research demonstrating their moderating role of on mental health outcomes. Traditionally, PMLDs were investigated as count variables or latent variables, disregarding plausible interrelationships among them.
Objectives: To use network analysis to investigate the associations among PMLDs.
Methods: Based on a cross-sectional measurement of seventeen PMLDs in a clinical sample of traumatized asylum seekers and refugees (N = 151), a partial correlation network was estimated, and its characteristics assessed.
Results: The network consisted of 71 of the 120 possible edges. The strongest edge was found between 'Communication difficulties' and 'Discrimination'. 'Loneliness, boredom, or isolation' had highest predictability.
Conclusion: Our finding of an association between communication difficulties and discrimination has been documented before and is of importance given the known negative impact of discrimination on mental and physical health outcomes. The high predictability of isolation is indicative of multiple associations with other PMLDs and highlights its importance among the investigated population. Our results are limited by the cross-sectional nature of our study and the relatively modest sample size.

Background

Prevalence rates of mental disorders among refugees and asylum seekers are well-known to be elevated compared to the general population (Blackmore et al., 2020). While associations between mental disorders and trauma exposure have been extensively demonstrated (Steel et al., 2009), the impact of stressors occurring after displacement on mental health has been studied less intensively. Such stressors, also termed postmigration living difficulties (PMLD), include a variety of problems, for example, socioeconomic, social, and interpersonal factors, as well as factors relating to the asylum process and immigration policy (Li, Liddell, & Nickerson, 2016). Importantly, a meta-analysis demonstrated that multiple PMLDs moderate mental health outcomes of refugees (Porter & Haslam, 2005). For example, a lack of economic opportunities, non-permanent accommodation and isolation were all associated with posttraumatic stress disorder (PTSD) and depression (Porter & Haslam, 2005; Steel et al., 2006).

Traditionally, research has treated PMLDs as one or several latent variables, or as a count variable (Stuart & Nowosad, 2020). The count model focuses on cumulative effects and does not investigate how PMLDs are related to each other (Nickerson et al., 2021). In the latent variable model, individual PMLDs are assumed to be independent from each other. It seems rather plausible, however, that PMLDs do influence each other. For example, an insecure visa status will likely negatively affect the possibility to obtain a permanent accommodation. Such interactions between variables are accounted by the network approach. This approach conceptualizes disorders and related phenomena as the result of a network of causal interactions (edges in a network) among individual symptoms or factors (nodes in the network). Thus, the network approach provides a more plausible framework for understanding PMLDs than the latent variable approach (Borsboom, 2017). In addition, the network models allow investigating how PMLDs relate to each other, which cannot be accomplished by assessing PMLDs as a cumulative count.

Regarding war affected populations, the network approach has been used to study the association between psychopathology and exposure to war (De Schryver, Vindevogel, Rasmussen, & Cramer, 2015), stressful life problems (Jayawickreme et al., 2017), and displacement stressor (Mootoo, Fountain, & Rasmussen, 2019). However, no study used a network approach to study postmigration living difficulties in refugees resettled in a high-income country yet. Notably, refugees and asylum seekers in high-income countries face at least partially different postmigration challenges compared to the populations studied in the studies outlined above (e.g. populations in a conflict-affected setting). Therefore, this study aimed to investigate the interactions of PMLDs in a population of asylum seekers and refugees in a high-income, western-European country.

Methods

The participants were recruited from a patient population seeking or undergoing treatment at the time of the assessment at two outpatients’ clinics for victims of torture and war in Zurich and Bern, Switzerland. Individuals speaking one of the study languages (Turkish, Arabic, Farsi, Tamil, German, or English), being at least 18 years old and neither experiencing acute suicidality, nor current psychotic or severe dissociative symptoms were eligible to participate in this study. All participants were assessed by a psychiatrist, clinical psychologist, or a masters-level student of clinical psychology. Participants were reimbursed with CHF 40 (approx. $USD 40). The study was approved by the Ethics Committees of the Cantons of Zurich (KEK ZH-Nr. 2011-0495) and Bern (EK BE 152/12) Switzerland and informed consent was obtained from all participants.

Participants’ exposure to trauma was assessed using 23 items derived from the Harvard Trauma Questionnaire (Mollica et al., 1992), and symptoms of PTSD were measured with the Posttraumatic Diagnostic Scale (Foa, Cashman, Jaycox, & Perry, 1997). Post migration living difficulties in the last 12-months were assessed using the adapted 17-item Post-Migration Living Difficulties Checklist (Silove, Sinnerbrink, Field, Manicavasagar, & Steel, 1997) (PMLDC). Each of the 17 PMLDC items addresses a specific difficulty which is rated on a five-point scale ranging from 0 = not a problem to 4 = very serious problem. The items are summed to a total score with a range of 0–68 points. The PMLD scale has consistently been identified as a predictor of mental health among displaced populations (Schick et al., 2016) and showed good internal consistency in this study (Cronbach’s alpha = 0.87).

The network analysis followed current recommendations (Epskamp & Fried, 2018). First, items sharing a lot of variance can bias the results. Thus, highly correlated items are usually combined for the network analysis (Burger et al., 2020). However, no formal cut-off for when items should be combined exists. We obtained zero-order correlations between all items of the PMLDC and found that two items, namely ‘Difficulties obtaining financial assistance’ and ‘Not enough money to buy food, pay rent or buy necessary clothes’ were highly correlated with each other (Spearman’s ρ = 0.68). The strength of the correlation of all other items pairs was below ρ = 0.6. Therefore, we combined both items by calculating their mean into a new variable ‘Financial problems’. Second, we estimated a Gaussian Graphical Model, in which each PMLD is represented by a node and edges correspond to partial correlations between two nodes. The parameters of the network were estimated using a regularization procedure based on the glasso. Due to the exploratory nature of our analysis, the tuning parameter (lambda) was set to 0.25. Furthermore, we assessed predictability of each node. Predictability represents the upper bound of shared variance, measured in R2 of a node, assuming that all connections from other nodes are directed towards the node in question (Haslbeck & Waldorp, 2018). Third, we carried out robustness and accuracy assessments using a bootstrapping procedure implemented in the bootnet package (Epskamp, Borsboom, & Fried, 2018). A total of 0.9% of the data was missing and was handled using pairwise deletion for the network estimation and listwise deletion for the calculation of predictability. The network analysis was carried out in the R statistical environment.

Results

A total of 151 participants were included in the final analysis. A summary of their demographic data is presented in Table 1. The type of the experienced trauma is reported in Table S1. Of all PMLDs, ‘Being unable to return to your home country in an emergency’ was the most endorsed PMLD with a mean endorsement of 3.82 (SD = 1.49) on a one to four scale, closely followed by ‘Loneliness, boredom, or isolation’ (3.79, SD = 1.16; see Table 2). Of the 120 possible edges, 71 were estimated to be non-zero. The resulting network is depicted in Figure 1. The bootstrapped 95% confidence intervals of the edge weights were mostly overlapping and are shown in Figure S1. Correspondingly, only the edge weight between ‘Communication difficulties’ and ‘Discrimination’ was significantly stronger than at least half of the other edges (also see Figure S2). Average predictability was 0.326, and with 0.609 ‘Isolation’ had the highest predictability (also see Table 2).

Table 1.

Sample characteristics

Variable
Male gender: n (%)	106 (70.2)
Age, years: mean (SD)	41.9 (9.8)
Nationality: n (%)Turkey (with n = 48 being Kurdish)IranSri LankaBosniaAfghanistanOthers	81 (53.6)13 (8.6)13 (8.6)6 (3.9)7 (4.6)31 (20.5)
Level of Education: n (%)Not completed primary schoolCompleted primary schoolAttended high schoolCompleted high schoolWent to technical schoolCompleted bachelor’s degree or equivalentCompleted postgraduate degree	21 (13.9)21 (13.9)27 (17.9)29 (19.2)17 (11.3)15 (9.9)10 (6.6)
Employment status: n (%)Full-timePart-timeUnemployedRetired/homemakerMarital status: n (%)SingleIn a relationship/marriedDivorced/widowedVisa status: n (%)Awaiting asylum decisionOther insecure visa statusSecure visa status or naturalized Swiss citizens	16 (10.6)21 (13.9)88 (58.3)23 (15.2)47 (31.1)88 (58.3)16 (10.6)32 (21.2)26 (17.2)91 (60.2)
Duration of stay in Switzerland: years (SD)	9.0 (6.6)
Average time in therapy in months	28.8 (27.9)
Number of experienced trauma types: mean (SD)	14.7 (4.1)

SD = standard deviation.

Table 2.

Mean endorsements and predictability of postmigration living difficulties

Variable	Mean	SD	Predictability
WORRI	3.77	1.27	0.248
SEPER	3.44	1.46	0.405
SENTH	3.24	1.70	0.217
RETUR	3.82	1.49	0.358
NREFU	2.36	1.72	0.325
ISOL	3.79	1.16	0.609
HEALT	2.49	1.44	0.000
GERMA	3.26	1.31	0.356
FASSI	2.83	1.41	NA
MONEY	2.80	1.41	NA
FINAN*	2.82	1.29	0.370
OFFIC	2.47	1.51	0.374
EMPLO	3.26	1.44	0.253
ECONF	1.93	1.09	0.091
DISCR	2.37	1.42	0.351
CONFL	2.38	1.41	0.301
COMM	2.95	1.44	0.495
ACCOM	2.97	1.49	0.457

SD = standard deviation. Total range for each postmigration living difficulties: 0–4.

Figure 1.

The estimated network of 16 postmigration living difficulties

Discussion

Among all PMLDs, ‘Loneliness, boredom, or isolation’ had the highest predictability and the second highest mean endorsement, making it a significant problem in the daily life of the participants in our sample. This finding stands in contrast to prior studies investigating PTSD and daily stressors in post conflict population in low-income countries, which found problems concerning basic needs to be most important to the network structure (De Schryver et al., 2015; Jayawickreme et al., 2017). This could reflect the different settings of the studies’ populations with our participants currently living in a high-income country where basic needs and safety are generally provided. Accordingly, mean endorsement of ‘not having enough money for necessities’ was low in our analysis (see Table 2). However, our finding is in accordance with several studies reporting high level of loneliness and isolation in refugees and asylum seekers, e.g. (Strijk, van Meijel, & Gamel, 2011). The high predictability of isolation and loneliness indicates that it might be a consequence of other PMLDs. Indeed, it seems likely that having communication difficulties and being separated from one’s family, with both being related to isolation in our analysis, can lead to perceived loneliness.

The edge between ‘Communication difficulties’ and ‘Discrimination’ was the strongest indicating that these two PMLDs were strongly related, even after controlling for the influence of all other PMLDs. This is in line with a systematic review demonstrating that language barriers are associated with perceived discrimination in refugee women during perinatal care (Small et al., 2014). Importantly, discrimination is associated with worse mental and physical health outcomes (Pascoe & Smart Richman, 2009). Thus, language and cultural interpreters might help to reduce discrimination perceived by refugees and could also lead to better health care outcomes.

Our study is subject to several limitations. First and most importantly, the sample size was relatively small for the number of nodes included in the network, leading to limited generalizability and validity of our results. Second, PMLDs are known to be related to mental disorders and additional factors. Thus, by only including PMLDs in the network analysis we were not able to investigate how these contextual factors impact the associations between different PMLDs. Due to these limitations and the exploratory purpose of this study, we see multiple avenues for future research. First, our preliminary findings should be replicated in studies with bigger sample sizes and in populations stemming from different settings. Second, future research should include factors known to be related with PMLDs (e.g. psychopathology) in the network analysis to broaden the understanding of the context in which PMLDs arise. Third, based on our findings suggesting that communication difficulties and perceived discrimination are uniquely associated with each other, future research could investigate derived hypotheses, for example, whether this relationship is mediated by religion or ethnicity.

Click here for additional data file.