The Dimensionality of Proposed DSM-5 PTSD Symptoms in Trauma-Exposed Young Children

The introduction of posttraumatic stress disorder for children 6 years and younger (PTSD-6Y) in the Diagnostic and Statistical Manual, Fifth Edition (DSM-5 2013) is an important acknowledgement that stress responses of young children show developmental differences compared to adults. Studies leading up to the DSM-5 revealed that PTSD was underdiagnosed in young children (Scheeringa et al. 2001, 2005). Under DSM-IV, to receive a PTSD diagnosis, a young child must have experienced a Criterion A trauma eliciting high levels of affect, presented with at least one re-experiencing symptom, three avoidance symptoms, two arousal symptoms, and shown impaired functioning. This algorithm was problematic as several symptoms from the DSM-IV avoidance cluster (e.g., a sense of hopelessness) were rarely detectable in young children, leading to under-diagnosis of the disorder (Scheeringa et al. 2001, 2005) with consequent effects for funding, and care provision. A proposal for more developmentally sensitive criteria modified the diagnostic algorithm to four symptoms only, removed the requirement for peri-event emotions, and suggested important developmental adaptations. Prior to DSM-5, several studies showed that this alternative algorithm (PTSD-AA) was superior to the DSM-IV algorithm in its alignment with the presence of PTSD-related clinical impairment (De Young et al. 2011; Meiser-Stedman et al. 2008; Scheeringa et al. 2003a, b).

Based on these findings, PTSD-6Y was established as a distinct diagnostic subtype in the DSM-5, mirroring the structure of PTSD-AA. Consequently, the key differences between PTSD with DSM-IV and PTSD-6Y within DSM-5 are the adaptation of symptoms for young children, the addition of a mood item (DSM-5 PTSD-6Y C3), and the requirement of only four symptoms instead of five for diagnosis. However, PTSD-6Y retains the distinctive symptom clustering of the adult diagnosis reflecting an as yet untested assumption that the latent structure of preschool PTSD symptoms mirrors that in adults. Accordingly, symptoms of PTSD-6Y are arranged into re-experiencing, arousal, and avoidance and negative mood and cognition clusters (Table 1: Model 1). The avoidance and negative alterations in cognitions cluster is further subdivided conceptually into two avoidance symptoms (DSM-5 PTSD-6Y C1-C2) and four negative alterations in cognitions symptoms (DSM-5 PTSD-6Y C3–C7), but these are not separate clusters in the diagnostic algorithm. To receive a diagnosis, a child must present with at least four symptoms from the three clusters, alongside functional impairment.

We are aware of no empirical examination to date of the factor structure of PTSD symptoms in young children. Establishing the factor structure of PTSD in young children may be helpful for determining whether PTSD has a distinct symptom profile in this age group. One plausible reason the profile could be different is the marked developmental shifts in cognitive and emotional functioning seen in young children, leading to a differential expression of PTSD symptoms relative to older children or adults (Feldman and Vengrober 2011; Scheeringa 2008; Salmon and Bryant 2002). Establishing the PTSD symptom structure in younger children would therefore assist with establishing a developmentally sensitive conceptualization of traumatic stress responses across the lifespan (Scheeringa et al. 2011). If the DSM-5 clustering algorithm for young children lacks validity then this goal would be compromised, and understanding of the etiology and maintenance of PTSD may be impeded. There may be underlying cognitive or biological mechanisms that relate to clusters not currently conceptualised in the DSM, preventing identification of these relationships, and compromising subsequent treatment models targeting these underlying processes. Clinically, alternative factor models will potentially give rise to different PTSD prevalence rates and it may be that clusters not currently outlined in PTSD-6Y have a stronger relationship to clinical impairment.

Confirmatory analytic investigations of the structure of PTSD symptoms have been carried out in sixteen samples of older children and adolescents (See Table 1). A summary of model structures tested in this study is provided in Table 2. We chose to conceptualize PTSD-6Y as a four-factor model as the symptoms are organized across 4 clusters within the DSM-5, even though a PTSD-6Y diagnosis requires 4+ symptoms from only 3 of those clusters. Overall, this four-factor model, consistent with the DSM-5 PTSD-6Y¹ structure (Table 2; Model 1), was the best fitting model in four samples of older youth (Sack et al. 1997; Saul et al. 2008) and an adequate model in other studies (Ayer et al. 2011; Ford et al. 2009; Liu et al. 2016; Wang et al. 2010, 2015). A unifactorial model (Table 2; Model 2) was adequate in four studies (Anthony et al. 1999; Ayer et al. 2011; Chen et al. 2012; Kassam-Adams et al. 2010), but unsupported in two others (Anthony et al. 1999). A two-factor model (Table 2; Model 3) was the best fitting model in one study (Ford et al. 2009) and an adequate model in two studies (Ayer et al. 2011; Saul et al. 2008). The DSM-IV 3-factor model (Table 2; Model 4) was the best fitting model in one study (Bal and Jensen 2007) and a good fitting model in 8 others (Ayer et al. 2011; Chen et al. 2012; Ford et al. 2009; Kassam-Adams et al. 2010; Saul et al. 2008; Wang et al. 2013). Finally, an alternative 4-factor solution – the so-called Dysphoria Model (Table 2; Model 5), a more complex model grouping symptoms that have a high degree of co-morbidity with anxiety and depression – has been proposed. This model retains the re-experiencing cluster and contains active avoidance, dysphoric arousal and anxious arousal clusters under the rationale that symptoms overlapping with depression and anxiety will cluster more strongly with one another. This model was the best fitting model in four samples (Boyes et al. 2012; Kassam-Adams et al. 2010; Wang et al. 2010) and adequate in eight studies (Ayer et al. 2011; Elhai et al. 2009; Liu et al. 2016; Sumner et al. 2014; Wang et al. 2010, 2013, 2015). In summary, in older youth there is some indication that 1-, 2-, 3-, and 4-factor models all variously meet thresholds for good fit on key fit indices, although for the most part, more complex models fit the data better.

Given the questions this raises for projecting adult structural models onto youth samples, the primary aim of the present study was to explore the latent structure of DSM-5 post-traumatic stress symptoms in young children aged 3–6 years using confirmatory factor analysis. The DSM-5 PTSD-6Y 4-factor model (Table 2; Model 1) was compared to the aforementioned competing models tested in the adult and older child literatures: a cluster free 1-factor account (Table 2; Model 2 – the basis for DSM-5 acute stress disorder), a 2-factor solution [arousal/avoidance and re-experiencing] (Table 2; Model 3), the DSM-IV 3-factor model (Table 2; Model 4) and the alternative 4-factor Dysphoria model (Table 2; Model 5). Based on the literature in older age children and adolescents, which showed that in the majority of studies more complex models fit the data better than more parsimonious models, we tentatively predicted that a similar pattern might be observed in our young child data. We did not make a prediction as to whether PTSD-6Y or the Dysphoria model would be superior as both have performed similarly well in studies of older age children and adolescents.

The main aim of any DSM diagnosis is to identify a child that is in clinical need, and therefore a second goal was to establish the clinical validity of each model. To do this, we investigated the convergent validity of the different diagnostic models with the exception of DSM-IV, which has shown to be poor in numerous previous studies, to internalizing and externalizing subscales of the Child Behavior Checklist (CBCL). We chose to look at the relation of a general ‘internalizing’ dimension as it has been shown to differentiate youths with and without anxiety (Seligman et al. 2004) as well as PTSD (Haller and Chassin 2012) symptoms in adolescents. PTSD is thought to be associated with both issues of externalization (e.g., anger and irritability) and internalization (e.g., thought avoidance). Based on previous research carried out in older age children and adolescents (Haller and Chassin 2012), it was predicted that PTSD would be associated with both internalizing and externalizing dimensions, but would show the stronger association with the internalizing dimension. We also investigated the criterion validity of the different models in relation to the presence of clinical impairment, conceptualized as being impaired in at least one area of functioning (e.g., relationships with parents, teachers etc.) (cf., Kassam-Adams et al. 2012).

This study aimed to evaluate for the first time the latent structure of PTSD symptoms in young children, following the introduction of PTSD for children 6 years and younger (PTSD-6Y) in the DSM-5. PTSD-6Y was compared to four alternative models evaluated in previous research. The 4-factor ‘Dysphoria’ and PTSD-6Y models provided the better accounts of symptom structure, although neither model met the requirements of a good fitting model without removal of the poorly loading flashback and intrusion symptom items. The partial support for these two models on fit indices did not remain after MIMIC modelling to control for differences in symptom expression associated with trauma type, gender, and age.

In terms of criterion validity, out of these two models the Dysphoria model offered the better balance of sensitivity and specificity. The more parsimonious 1-factor solution offered the best conceptualization in this respect, but none of the models was particularly strong. In terms of convergent validity, the two 4-factor models also displayed only small levels of convergence with CBCL dimensions. The findings indicate that current models of PTSD symptom clustering derived from the adult field do not do a compelling job of capturing the symptom profiles of trauma-exposed preschoolers.

This failure to find an adequate model fit with all symptoms included is at odds with CFAs in older children where it is common to find multiple excellent fitting models involving the full list of symptoms (Ayer et al. 2011; Elhai et al. 2009; Ford et al. 2009; Kassam-Adams et al. 2010; Saul et al. 2008). Low rates of endorsement do not explain the finding as flashbacks were endorsed by 18.3% of the sample and intrusions were endorsed by nearly half the sample (44.7%). One interpretation is that flashbacks and intrusions are not a central component of the construct in young children, although this is inconsistent with clinical observation. Another possibility is the potentially poor validity of parent report for these symptoms in young children (Meiser-Stedman et al. 2007). A final possibility is that the range of PTSD symptoms enshrined in the DSM for young children is too narrow and detailed qualitative investigations of young children’s clinical presentations are indicated, in line with similar endeavors in adults (e.g., Kendler 2016).

Some aspects of the study merit comment. Despite being a diverse sample, almost half of the children experienced only one trauma type (Hurricane Katrina; 48%) and only 30% experienced an interpersonal stressor. Furthermore, only 23% of the fathers in the sample lived in the family home, which is lower than American prevalence estimates found in the general population and may have had some impact on rates of clinical prevalence. How representative the current sample is of the wider trauma population is therefore an important empirical question for future research. It will also be important to replicate these findings in a clinic-attending sample. Finally, to test criterion validity we used the same impairment criterion required for the diagnosis. It is important to replicate our findings around criterion validity using an impairment criterion that is more distant from the diagnosis; for example, a measure of quality of life.

There are other important avenues for future research. Additional tests are needed to evaluate the predictive validity of competing models and their impact on recovery trajectories and relapse rates. In line with the goals of the Research Domain Criteria (Insel et al. 2010) and other trans-diagnostic developmental work (McLaughlin et al. 2011), a logical progression of the present study involves exploratory factor analytic work across disorders in preschool children including core pathology common to depression, anxiety and adjustment disorders, as well as other core psychological processes associated with early stress responses. Further research is also needed in larger samples to determine the differences in psychopathology expressed by simple and complex trauma groups.

To summarize, the present study suggests that in young children, when considering the latent structure of PTSD symptoms there is inconclusive support for four separable symptom clusters as proposed in the DSM-5. Overall, the results suggest that PTSD factor models largely established in the adult literature do not provide an adequate fit of symptom clustering in pre-school children.