Discussion
The current study examined the construct validity of the DSM-5 PTSD model, alongside four competing models identified in the literature, in capturing the trauma-related psychopathology of those exposed to persecution and displacement. To our knowledge, this is the first study to investigate with a refugee sample the validity of the Dysphoria and Dysphoric Arousal models, based on DSM-5 criteria, in addition to the newly proposed Anhedonia and Externalising Behaviours models. Findings from five CFAs revealed that all models of interest, including the DSM-5 model, demonstrated good fit across CFI and TLI but higher than desirable RMSEA. However, relative to the alternate four, five and six-factor models examined, the DSM-5 model was the poorest fitting model for our data. Moreover, our study found preliminary evidence in support of the Anhedonia model as the best fitting model for our sample of traumatised refugees.
Our findings add to a small, but growing, body of research on the symptom structure of PTSD within culturally diverse refugee and post-conflict samples. With regards to the DSM-5 model, our findings accord with previous research [
9,
36]. Schnyder et al. [
36] found that the DSM-5 model produced adequate to good model fit among treatment-seeking refugees in Switzerland. This is consistent with our study where the DSM-5 model demonstrated adequate fit for our refugee sample. Importantly, however, the present study found that the DSM-5 model had the poorest relative fit for our sample compared to all four of the competing models that we tested: the four-factor Dysphoria model, five-factor Dysphoric Arousal model, and six factor Anhedonia and Externalising Behaviours models. This additional finding is not necessarily inconsistent with Schnyder et al.’s findings as alternate PTSD models based on DSM-5 criteria were not tested, so it is entirely possible that their sample would have yielded similar patterns of relative fit if alternate DSM-5 models had been tested. Notably, our findings are broadly consistent with Michalopoulos et al.’s [
9] study, which assessed relative model fit of the DSM-IV model, DSM-IV Dysphoria model, and DSM-IV Emotional Numbing model against an approximation of the DSM-5 model among three culturally diverse samples from low or middle income countries: sexual assault survivors from the DRC, Burmese refugees in Thailand, and Iraqi torture survivors. Michalopoulos et al. found that although their approximation of the DSM-5 model evidenced adequate to good fit for their DRC and Burmese samples, it demonstrated poor fit for Iraqi survivors of torture. Moreover, the DSM-5 model was not the best fitting model for any of the samples. Instead, the best fitting model varied across samples: the Emotional Numbing model demonstrated superior fit for the DRC sample, the Dysphoria model was most appropriate for the Burmese sample, and none of the models tested adequately represented the Iraqi sample. The authors acknowledged that time since trauma exposure may have been a factor explaining the poor fit of the models to the Iraqi sample, as the height of trauma exposure for the Iraqi participants was approximately 20 years prior to data collection. However, Michalopoulos et al. did not assess more complex PSTD models, such as the newly proposed Anhedonia model, which may have been more appropriate for this sample. Taken together, our findings build on extant research to demonstrate that the DSM-5 model, although generally producing adequate fit, was not the best representation of the latent structure of PTSD when applied to this culturally-diverse refugee sample.
In the current study, the six-factor Anhedonia model fit the data better than the four-factor DSM-5 and Dysphoria models, the five-factor Dysphoric Arousal model, and the six-factor Externalising Behaviours model. This finding replicates previous research with non-refugee samples that found the Anhedonia model to be superior to all other four-, five- and six-factor models [
26,
27] and conforms to a trend in previous research where best fitting models tend to specify more factors [
7]. However, it should be noted that two factors of the Anhedonia model, the Avoidance and Anxious Arousal factors, comprise only two items. This may be problematic from a statistical standpoint as models that specify factors with less than three indicators may result in an under-identification of the model and inaccurate or unstable parametric estimates in CFA [
7,
48]. While this is a notable statistical limitation of the Anhedonia model, it is important to consider the ubiquity of this problem as many of the DSM-5-derived models for PTSD include at least one factor (Avoidance) that is only specified by two items, based on the DSM-5 symptom criteria in which only two symptoms are specified in this cluster. Moreover, other frameworks, such as the
International Statistical Classification of Diseases and Related Health Problems (ICD) are moving towards more parsimonious PTSD models that include fewer factors with fewer items in each factor. For example, the proposed PTSD diagnosis for the ICD-11 comprises three factors with two symptoms each [
50]. As such, there is a need to reconcile this statistical limitation with the value of theoretically-derived parsimonious models that consider the availability of clinical resources in the field. A possible solution for future investigations of such models, offered by Marsh and colleagues [
51], could be to use larger samples, of 400 participants or greater, in analyses to ensure fully valid solutions.
The Anhedonia model, comprising intrusion, avoidance, negative affect, anhedonia, dysphoric arousal and anxious arousal, deviates from the DSM-5 model in two key ways. First, it divides symptoms of arousal into dysphoric arousal, comprising symptoms of irritability or anger (E1), reckless or self-destructive behaviour (E2), difficulty concentrating (E5), and sleeping difficulties (E6), and anxious arousal, comprising symptoms of hypervigilance (E3) and exaggerated startle response (E4) [
21]. This separation is supported by CFA studies that demonstrated that anxious and dysphoric arousal were distinct constructs among representative samples from Australia and the United States [
52], Malaysian tsunami survivors [
53], terrorist attack first-responders [
54], and adolescent earthquake survivors [
55]. Notably, while Liu et al. [
21] found that the two factors of dysphoric and anxious arousal were strongly correlated (.97), this correlation is lower in our sample (.86), suggesting that the two symptom clusters, although related, are distinct from one another. Second, the Anhedonia model divides NAMC symptoms into negative affect and anhedonia (deficit in experiencing positive affect), which is supported by theoretical and empirical evidence that alterations to positive and negative affect represent distinct constructs in mood and anxiety disorders [
24,
25,
56]. It is notable that the fit of the Anhedonia model, which was the only model to differentiate between symptoms of negative affect and anhedonia, was superior to all the other models that were examined. While negative affect yielded high factor correlations with anhedonia and dysphoric arousal in our sample, this is consistent with previous research [
21,
57]. Moreover, the collapsing of these symptoms into a single factor produced the Dysphoria model, which demonstrated a significantly worse fit compared to the Anhedonia model. As such, negative affect and anhedonia appear to represent two distinct constructs of DSM-5-defined PTSD for our sample. Further, while the Anhedonia model evidenced some high factor intercorrelations, a single factor model was tested and demonstrated unacceptably poor fit statistics and significantly poorer fit relative to the Anhedonia model, which suggests that a single factor model does not best fit the underlying factor structure of PTSD in this sample.
In the current study, all PTSD symptoms displayed relatively high factor loadings on their corresponding factors in the Anhedonia model. In particular, we found that psychological and physiological reactivity to traumatic reminders (B4 and B5) displayed very high factor loadings on the re-experiencing factor (.97 and .92 respectively). Similarly, Schnyder et al.’s [
36] study also evidenced high factor loadings across these symptoms (.92 and .89 respectively) in a sample of treatment-seeking refugees. This finding may be reflective of the unique experiences of refugees who are exposed to a complex constellation of cumulative interpersonal trauma, uncertainty and on-going stressors. First, our sample was exposed to a very high number of interpersonal traumatic events, i.e. trauma that is perpetrated by another human, and previous research has found distress to reminders (as well as intrusive memories) are significantly elevated among survivors of interpersonal trauma compared to non-interpersonal trauma [
58]. In addition, many refugees regularly encounter highly salient reminders of traumatic events by virtue of exposure to information about ongoing persecution and conflict in the home country (i.e., via media reports and contact with family in the home country), which is likely to contribute substantially to psychological and physiological distress. As such, reactivity to traumatic reminders may be especially characteristic of symptoms of re-experiencing for refugees. Following this, refugees may attempt to avoid thinking about or talking about past traumatic experiences, which may account for the high factor loadings of avoidance of traumatic thoughts and reminders (C1 and C2) in this study. Further research is required to elucidate the phenomenology of these symptoms and their inter-relationships in trauma-affected refugees.
Our sample also displayed very high factor loadings for the symptoms of hypervigilance and an exaggerated startle response (E3 and E4) on the anxious arousal factor, which may have been due to the high rates of prolonged and repeated exposure to interpersonal traumatic events such as torture, combat, kidnapping and sexual violence. This would be consistent with longitudinal research by Forbes et al. [
58] that found significantly higher rates of both hypervigilance and an exaggerated startle response symptoms among survivors of interpersonal trauma compared to those exposed to non-interpersonal trauma.
Previous CFA studies using non-refugee samples repeatedly found relatively low factor loadings for symptoms of reckless or self-destructive behaviours [
21,
57,
59], yet, this was not the case for our sample where factor loadings were high (.81). Notably, this symptom also produced a high factor loading for Schnyder et al.’s [
36] refugee sample (.74). This suggests that symptoms of reckless or self destructive behaviours may be more relevant to the presentation of PTSD within refugee samples. Indeed, refugees, who experience multiple and prolonged traumatisation, often present with complex reactions to traumatic events, which can manifest as reckless behaviour [
30]. This finding calls for further investigation to identify what kinds of reckless behaviours trauma-affected refugees may be especially likely to engage in. Studies that have investigated this further have differed substantially in how they define these behaviours. For example, Michalopoulos et al. [
9] operationalised reckless behaviour as “drinking too much alcohol” in their Iraqi and Burmese samples and although the authors did not report factor loadings, they found that this was the least frequently endorsed item for both samples. An alternative conceptualization of reckless behaviours that may be especially relevant is self-harming behaviours, which have been found to be elevated among refugee populations [
60,
61]. Further research is required to elucidate the specific manifestations of this symptom amongst trauma-affected and displaced refugee populations.
Symptoms relating to anger (E1), negative beliefs (D2), distorted blame (D3), and persistent negative emotional state (D4) also displayed relatively high factor loadings in our sample, ranging from .77 to .87. This finding also aligns with previous research that found that refugees displayed different forms of emotion dysregulation in reaction to traumatic events, such as excessive guilt, self-blame and outbursts of anger [
30]. In light of this, the decision to broaden the scope of PTSD in the DSM-5 to include the new symptoms of reckless or destructive behaviours (E2), negative beliefs (D2), distorted blame (D3), and persistent negative emotional state (D4), may be particularly pertinent to the clinical presentation of PTSD amongst refugees.
Several limitations of the current study should be acknowledged. First, our sample comprised participants from a number of different cultural backgrounds. Although this is an ecologically valid representation of the cultural diversity inherent in global refugee populations, it is possible that important cultural differences specific to a single group may have been masked in the current study. Moreover, the cultural and linguistic variation present in our sample may have influenced the model fit for our data. However, notwithstanding this notable limitation, the aim of the present study was to investigate the phenomenology of PTSD among a culturally diverse sample of people who have experienced persecution and displacement, in order to reflect the global refugee population. As such, we chose to analyse our refugee sample collectively, rather than according to specific sub-populations, as a way to adequately represent, and understand, the universal features of traumatic stress. A second limitation was that self-report questionnaires were used to assess PTSD symptoms. Although self-rated PTSD symptom scores are strongly correlated to clinician-rated PTSD symptom scores [
62], clinician-administered structured interviews provide additional standardised information regarding symptom severity and clinical impairment, which was not measured in the current study. Third, the instructions and measures of the questionnaire were translated into three languages, as well as available in English. Although care was taken to follow strict translation procedures, such as blinded back translation [
37], it is possible that minor deviations between languages in the meaning of some words remained. Finally, our study was limited by only investigating DSM-5-defined PTSD symptoms. Despite evidence of the cross-cultural validity of DSM-defined PTSD, culturally specific responses to trauma also exist, such as somatic symptoms, which are not currently included in the DSM-5 criteria for PTSD [
63]. Future ethnographic research is necessary to identify the full breadth of symptom experience for trauma-exposed refugees.