Discriminating Bipolar Depression From First-episode Depressive Patients

Background:The therapeutic strategy and clinical outcomes of bipolar depression and unipolar depression are completely different,although both have similar clinical symptoms.It is urgent to find biomarkers which could discriminate bipolar and unipolar depression because of the insufficient effect of clinical feature on distinguishing both.Some researchers argue that hyper-and hypo-sensitivity to reward is the core feature of bipolar and unipolar depression respectively,and many published paper reported different network metrics of reward circuit in both.But these features may not reflect early signs of bipolar depression and have an ambiguous effect on discrimination because of lack of validation and ignoring the role of clinical features.Therefore,the present study would explore the network metrics of reward circuit of first-episode bipolar and unipolar depression,and identified and compared the discrimination effect of classifiers based on network metrics feature,clinical feature and both.Method : Follow-up the first-episode,drug-na?ve depressive patients collected between May 2010 and September 2016(7.51 ± 1.73 years),and selected 32 first-episode bipolar depression patients(fBD),48 first-episode unipolar depression patients(fUD)based on follow-up outcomes.We also recruited 31 healthy controls(HC).Functional connectivity(FC),dynamic functional connectivity and graph theory method were used to explore the difference of network metrics of reward metrics between groups.Support vector machine and leave-one-out cross-validation was used to examine and compare discrimination effect of network metrics and clinical features.Results:Compared to HC and fUD,fBD showed decreased degree centrality and nodal efficiency(NE)of right mediodorsal thalamus(MD)and left ventral pallidum,respectively.Compared to HC and fBD,right MD of fUD showed increased FC with bilateral dorsolateral prefrontal cortex and left ventromedial prefrontal cortex.fUD also have decreased NE of right MD.Classifier trained on network metrics of reward circuit centered on MD connectivity metrics,which performance was superior to classifier trained on clinical feature(Accuracy: 70.00% vs 68.75%,Sensitivity: 59.38% vs 59.38%,Specificity:77.08% vs 75.00%),had moderate classification performance.Classifier converged network metrics and clinical features achieved high classification performance(Accuracy:87.50%,Sensitivity: 78.13%,Specificity: 93.75%).Conclusion:Altered network metrics centered on MD connectivity metrics are important endophenotype of mood disorders,and different abnormal pattern could discriminate bipolar depression from unipolar depression.Network metrics and clinical features are complementary,and classifier converged both is high-effect for early identification of bipolar depression.The study provide additional evidence for reward sensitivity hypothesis of mood disorders.Further studies should verify the results and elaborate the questions we raised.