The latent structure of post-traumatic stress disorder among Arabic-speaking refugees receiving psychiatric treatment in Denmark

The present study is, to our knowledge, the first to assess the factorial structure of PTSD in an Arabic-speaking population of refugees. Results replicate previous findings that a 4-factor model with a separate avoidance factor provides a better fit than the DSM-IV model. In this regard, the study adds cross-cultural support for the decision to place symptoms of effortful avoidance in a separate cluster in the DSM-5.

Two models could not be properly estimated, namely those separating items D2 and D5. Looking at the original, English, wording in the HTQ, symptom D5 is formulated: “Feeling jumpy, easily startled.” Asking two separate Arabic translators to back translate this item, they noted a connotation of “flare up” (سرعة الهيجان). This constitutes an unintended content overlap with item D2, Feeling irritable or having outbursts of anger. As noted, removing item D2 was the only solution to this problem, which could be applied across all of the models. With this modification, CFI, TLI and RMSEA were virtually identical across the Numbing-, Dysphoric Arousal-, and Aroused Intrusion model. That the 5-factor model did not offer a significantly better fit than the 4-factor Numbing model supports the latter as a more parsimonious model. This was also reflected in the BIC, which had a steeper penalty of model complexity and thus provided strong evidence against the 5-factor model. We cannot know for certain how much the unintended overlap of items D2 and D5, as well as our removal of item D2, influences these results. Removing item D5, rather than D2, would perhaps do more justice to the Dysphoria and Dysphoric Arousal models as item D5 introduces a component of anger in the anxious arousal cluster. But with only two items on the anxious arousal factor, item D5 was indispensable.

Another notable error correlation concerned symptoms B1, “Recurrent thoughts or memories of the most hurtful or terrifying event”, and B3, “Feeling as though the event is happening again”. There was no indication that the content overlap was any larger in the Arabic translation than in the original, English version. Rather than being an artefact, we believe this result reflects a particularly close relation between these two symptoms. At least in these authors’ clinical experience, trauma patients can easily progress from cued recall, e.g. when asked about circumstances surrounding the trauma, to gradually dissociate in their sensory experience. A higher average endorsement of item B1 (M = 3.45) over item B3 (M = 3.30) provides some support for the idea that recall offers a prerequisite for re-experiencing. Future studies may wish to report if such a substantive error correlation is replicated.

Unspecified residual correlations may have a crucial impact on the evaluation of alternative models. When MI and EPC suggest freeing up the correlation between B1 and B3, it means that item B1 and B3 had a higher correlation than the shared construct of intrusion could properly account for. This is reflected in the correlations among the residuals, which for items B1 and B3 is estimated to be 0.27 for optimal fit. Restricting this correlation to zero introduces strain on the intrusion factor, which is forced to account for all of the shared variance of B1 and B3. The loadings of the involved item may become inflated in this situation, and/or remaining loadings on the factor may become deflated [8]. The strain may influence models differently; For the items D2 and D5, the MIs indicated that the best fitting baseline model was least influenced, while the poorest fitting 4-factor model was influenced the most. A comparison of models based exclusively on global fit may thus favour a model because it is less influenced by unintended local sources of strain. Researchers in the field of transcultural psychiatry should be particularly alert to such potential methods effects as there is more to go wrong upstream in cross-cultural assessment, including translation and cultural adaptation of instruments.

Another methodological issue that may easily influence the validity of cross-cultural CFA-studies is the minimal size of some of the theorised factors. Fewer items in a factor will generally challenge replicability of a given factor structure [24, 50]. In this regard, the original DSM model provided a relatively even distribution of items across clusters. Although the separation of avoidance symptoms in the DSM-5 is informed by a vast number of studies reporting superior fit of this configuration, we believe it deserves further psychometric scrutiny on two accounts. First, to our knowledge, no prior study has tested the Numbing model against a DSM-IV model with error terms between the two avoidance items. If such a test does not favour the 4-factor Numbing model, the psychometric support for a separate avoidance factor becomes less evident. Second, if avoidance is found to constitute an independent latent trait, then that trait should arguably receive full content coverage. If other factors are covered by five to seven symptoms, then, from a psychometric perspective, it is not clear why avoidance is only characterized by two symptoms.

Similar points can be raised with regard to the two-item anxious arousal factor. But given that it is not unequivocally supported, and given the introduction of new symptoms in the DSM-5, future directions for testing this factor are less clear. The arousal cluster sees one new symptom in the DSM-5, reckless behaviour, which could potentially help stabilise either the dysphoric arousal or anxious arousal factor. According to [14], it was included in the DSM-5 because it is seen as an important symptom in traumatised adolescents. Initial factor studies of the DSM-5, however, indicate poor loadings of this item [25, 29]. If small factors persist, we would urge future cross-cultural CFA studies to exert caution in the evaluation of these.

The single-factor model with three error correlations showed good fit indices. One should always be highly cautious when interpreting fit indices based on modelled fit residuals, as they are likely to represent overfitting to the particular sample [18, 23]. However, the result indicates that in this particular sample, the HTQ can be treated as a unidimensional scale. This implies that an analysis based on item response theory could provide more information about how individual items contribute to the scale as a whole, as well as how they contribute differently across gender and age. Future studies may test whether a general factor PTSD model, which allows residual correlations between item B1 & B3, and between the avoidance items, shows acceptable fit in other samples of refugees.

A number of observations regarding individual items are worth noting. Symptom C3, trauma related amnesia, consistently loaded below 0.2 and thus make a poor contribution to the construct of PTSD in this sample. A number of previous cross-cultural studies also reported this as the weakest loading item [22, 26, 38, 44, 53]. It is perhaps the most disputed symptom of PTSD and critics question whether dissociative amnesia is a likely, or even possible, result from traumatic experiences [27, 39]. Symptom C6, inability to feel emotions, also displayed a relatively low endorsement and loading. According to a number of clinicians and interpreters working with the present sample, it is the item most frequently inquired about. We believe that an ‘inability to feel positive emotions’, in accordance with the DSM-5, will be a much easier concept to convey across cultures, particularly in a questionnaire form.

Regarding avoidance symptoms, patients will often report verbally that they try intensely to avoid thoughts and feelings of the traumatic events but repeatedly fail in these efforts. Asking patients to rate their distress from any attempted avoidance of thoughts and feelings, rather than only successful avoidance, may afford a more valid assessment in this population. Similarly, regarding the avoidance of activities, some patients express a perceived comfort in complete social isolation, while being distressed from social demands. Although they may have particularly fearful reactions to domain specific situations, such as seeing uniformed men, they will often report being uncomfortable around strangers in general. In this context, it may be beneficial to assess social isolation with specific and separate reference to depression and anxiety, e.g. “Avoid leaving my home because I expect other people to look down on me” and “Avoid leaving my home because I expect to witness or become a victim of violence”.

As noted, the DSM-5 introduces a number of new symptoms, which are not included in the HTQ, and consequently not covered in this study. The DSM-5 was officially introduced in 2013, and other PTSD scales have been updated to meet the new content, e.g. the Posttraumatic Stress Disorder Checklist [6] and the Clinician-Administered PTSD Scale (PCL-5; [54]). However, we found no indication that revisions of the HTQ are planned. To promote transparency and standardization, future studies may wish to adapt DSM-5 and ICD-11 items from existing questionnaires, such as the PCL-5. The new DSM-5 symptoms mainly concern the numbing factor, which now contains seven symptoms and is named ‘negative alterations in cognitions and mood’ (Table 1). We note that the Numbing cluster is the only large cluster, which is not divided in any of the models. Thus, in terms of factor structure, it could be considered the cluster least likely to be affected by additional items. The impact of the new arousal item, ‘reckless or self-destructive behavior’, is more difficult to estimate. As already noted, it has received low endorsement in initial studies, and, based on clinical experience, we would expect this to be the case in the population tested in the current study. We would encourage future studies to report whether the contribution of this particular item is clear, and when this is not the case, to explore alternative solutions.

From a clinical perspective, the current study supports the construct of PTSD in Arabic-speaking refugees, and as such supports the use of interventions targeting PTSD. Still, it is important to consider the influence of comorbid depression in a sample as chronic as this. Not only is the comorbidity of depression almost absolute, also symptom C5, “Feeling detached or withdrawn from people”, and C7, “Feeling as though you don’t have a future”, displayed the highest loadings on the single factor PTSD model. These symptoms will likely need addressing from the beginning of therapy, in order to provide a motivational platform for deliberate cognitive and behavioural exercises. One possible way to pursue this is through working with personal values, both rediscovering old values and adapting them to the new life circumstances, e.g. as described in Acceptance and Commitment Therapy [52]. Facilitating social contact in a welcoming environment, with no stigmatizing associations within the culture of the patient, may also be important for patients who have become excessively isolated.