Our voxel based morphometry analysis of

Our voxel based morphometry analysis of the S-NR group provided a more specific cluster of reduced volume within the left middle frontal gyrus — which improved the specificity of this step of the decision tree from 77% to 89%. At a whole p53 tumor suppressor level no other GM regions were associated with this S-NR group. The significance of this specific structural feature in the selected- non-remitter group was highlighted by a failure to identify any significant GM regions when a comparison was performed comparing remitters with non-remitters in the whole cohort. The significant cluster corresponds to the dorsolateral prefrontal cortex, a region implicated in a number depression studies, which is widely held to be involved in the control of executive and emotion functions in the brain (Hamilton et al., 2012; Fitzgerald et al., 2008). Our previous analysis of GM volume reduction at baseline in MDD demonstrated profound GM reductions in this region compared to normal subjects (Grieve et al., 2013a). Our findings also extend validation of reduced GM volume in this region found to be associated with reduced likelihood of antidepressant response in a recent meta-analysis (Fu et al., 2013). The right angular gyrus was the second component of our optimal decision tree, the exact role of this region in depression is poorly understood, but may relate to its putative involvement in the default mode network, cognition (e.g. memory, sematic processing, attention) or in its role in the synthesis of complex social and environmental information (Seghier, 2013). While the GM volumetric decision tree was superior to the DTI tree, the latter model was still significant and produced comparable overall results (84% accuracy). More importantly, the DTI decision tree had considerable overlap with the GM volumetric tree with 100% accuracy of non-remission in individuals (n=10) who were identified by both tests as non-remitters. Although the fraction of subjects both these tests applied to is small, this result does illustrate the potential of additive (in series or in parallel) usage of such data in a clinical algorithm. A great deal of literature highlights functional brain changes as predictors of treatment outcome in MDD (Fu et al., 2013; Mayberg et al., 1997). The inclusion of fMRI measures may add further value to the predictive algorithms we have identified. This study had several limitations. The iSPOT-D trial is a naturalistic study, and as such, lacks a placebo arm, this places important limitations on how we can use this data to understand the brain circuits that govern remission — the most critical of these being differentiating the specific effects of ADM therapy from spontaneous remission. On the other hand, the design of the iSPOT-D trial to mimic routine clinical practice means that our results reflect a real world setting. We included both participants who were ADM naïve or had previous history of ADM use. Although participants with previous ADM use underwent a washout period prior to participating in the study, the long-term impacts of ADM use on brain structural features that may impact on remission are not well understood. It is also possible that participants with previous ADM use may represent a treatment resistant group. As a result, the sensitivity of our findings with respect to medication history should be inferred with caution. Our findings are limited to the three commonly prescribed ADMs used in the study, the generalizability of these findings to other classes of ADMs currently available needs further work. Although our results provide encouraging evidence that neuroimaging measures can play a role in ruling out ADM use, markers that may reliably predict treatment success for a particular patient still remain to be identified. An inherent limitation of the employed approach is that the test could be applied only in MDD participants who met cutoff criteria for the identified measures. This may lower the overall sensitivity and specificity of prediction when considering the whole cohort. However, a test, albeit applicable to a smaller group but which has a greater reliability in identifying whether a patient will not remit to ADMs, would still be clinically meaningful and an improvement to current practice. Further work is required to understand the role that neuroimaging may play in guiding the use of other types of depression treatments (e.g. cognitive behavioral therapy, repetitive transcranial magnetic stimulation, deep brain stimulation, etc.) versus ADMs or combination treatment regimes.