Clinical patterns in a neuroimaging-based predictive model of self-reported dissociation

Assessment of trauma-related dissociation has been historically challenging given its subjective nature and the lack of provider education around this topic. Recent work identified a promising neural biomarker of trauma-related dissociation, representing a significant step toward improved assessment and identification of dissociation. However, it is necessary to better understand clinical factors that may be associated with this biomarker. Participants were 65 women with histories of childhood maltreatment, posttraumatic stress disorder (PTSD), and varying levels of dissociation (e.g., co-occurring dissociative identity disorder, DID). Data were drawn from previously published work that identified a model predicting Multidimensional Inventory of Dissociation severe pathological dissociation scores on the basis of neural functional connectivity. Here, we conducted a k-means cluster analysis to explore model performance patterns in the original prediction model. We then investigated differences among the clusters in a range of clinically-relevant variables. Our clustering analysis identified four distinct groups. The original model best predicted those at the low (cluster 1, 82 % PTSD) and high (cluster 3, 86 % DID) ends of self-reported dissociation. Cluster 2 also largely included participants with DID (67 %), but the predictive model was less accurate for these individuals. Follow up analyses revealed that DID participants in cluster 2 reported lower levels of self-state intrusions, a type of DID-specific dissociation, compared to those in cluster 3. Thus, the identified model performance patterns suggest that the original prediction model may be linked to DID-specific dissociation. Our findings indicate that patterns of functional connectivity may be valuable to improve accurate assessment of DID.