Search

Objectives: Antidepressants are first-line treatments for major depressive disorder (MDD), but 40-60% of patients will not respond, hence, predicting response would be a major clinical advance. Machine learning algorithms hold promise to predict treatment outcomes based on clinical symptoms and episode features. We sought to independently replicate recent machine learning methodology predicting antidepressant outcomes using the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) dataset, and then externally validate these methods to train models using data from the Canadian Biomarker Integration Network in Depression (CAN-BIND-1) dataset., Methods: We replicated methodology from Nie et al (2018) using common algorithms based on linear regressions and decision trees to predict treatment-resistant depression (TRD, defined as failing to respond to 2 or more antidepressants) in the STAR*D dataset. We then trained and externally validated models using the clinical features found in both datasets to predict response (≥50% reduction on the Quick Inventory for Depressive Symptomatology, Self-Rated [QIDS-SR]) and remission (endpoint QIDS-SR score ≤5) in the CAN-BIND-1 dataset. We evaluated additional models to investigate how different outcomes and features may affect prediction performance., Results: Our replicated models predicted TRD in the STAR*D dataset with slightly better balanced accuracy than Nie et al (70%-73% versus 64%-71%, respectively). Prediction performance on our external methodology validation on the CAN-BIND-1 dataset varied depending on outcome; performance was worse for response (best balanced accuracy 65%) compared to remission (77%). Using the smaller set of features found in both datasets generally improved prediction performance when evaluated on the STAR*D dataset., Conclusion: We successfully replicated prior work predicting antidepressant treatment outcomes using machine learning methods and clinical data. We found similar prediction performance using these methods on an external database, although prediction of remission was better than prediction of response. Future work is needed to improve prediction performance to be clinically useful., Competing Interests: JJN, TN, YZ, RTN, JC, and RU have no disclosures. BNF has a research grant from Pfizer. RM has received consulting and speaking honoraria from AbbVie, Allergan, Janssen, KYE, Lundbeck, Otsuka, and Sunovion, and research grants from CAN-BIND, CIHR, Janssen, Lallemand, Lundbeck, Nubiyota, OBI and OMHF. SHK has received honoraria or research funds from Abbott, Alkermes, Allergan, BMS, Brain Canada, CIHR, Janssen, Lundbeck, Lundbeck Institute, Ontario Brain Institute, Ontario Research Fund, Otsuka, Pfizer, Servier, Sunovion, Xian-Janssen, and holds stock in Field Trip Health. RWL has received ad hoc speaking/consulting fees or research grants from: Allergan, Asia-Pacific Economic Cooperation, BC Leading Edge Foundation, Canadian Institutes of Health Research (CIHR), Canadian Network for Mood and Anxiety Treatments (CANMAT), Canadian Psychiatric Association, Healthy Minds Canada, Janssen, Lundbeck, Lundbeck Institute, MITACS, Myriad Neuroscience, Ontario Brain Institute, Otsuka, Pfizer, University Health Network Foundation, and VGH-UBCH Foundation. CAN-BIND is sponsored by non-for-profit and government funding agencies and by commercial sponsors including Lundbeck, Bristol-Myers Squibb, Pfizer, and Servier. None of the sponsors were involved in study design, conduct of the study, data analysis, data sharing (except for the government funded Ontario Brain Institute, as stated in the Data Availability Statement), interpretation of results, drafting of the manuscript, or final approval and publication of the manuscript. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

Treatment-resistant depression (TRD) occurs in many patients and causes high morbidity and mortality. Because TRD subjects are particularly difficult to study especially longitudinally, biological data remain very limited. In a preliminary study to judge feasibility and power, 25 TRD patients were referred from specialty psychiatric practices. All were severely and chronically depressed and mostly had comorbid psychiatric disorders as is typical in TRD. Nine patients were able to complete all required components of the protocol that included diagnostic interview; rating scales; clinical magnetic resonance imaging; medication washout; treatment with maximally tolerated olanzapine-fluoxetine combination for 8 weeks; and pre- and post-treatment fluorodeoxyglucose positron emission tomography. This drug combination is an accepted standard of treatment for TRD. Dropouts arose from worsening depression, insomnia, and anxiety. One patient remitted; three responded. A priori regions of interest included the amygdala and subgenual cingulate cortex (sgACC; Brodmann area BA25). Responders showed decreased metabolism with treatment in the right amygdala that correlated with clinical response; no significant changes in BA25; better response to treatment the higher the baseline BA25 metabolism; and decreased right ventromedial prefrontal metabolism (VMPFC; broader than BA25) with treatment which did not correlate with depression scores. The baseline metabolism of all individuals showed heterogeneous patterns when compared to a normative metabolic database. Although preliminary given the sample size, this study highlights several issues important for future work: marked dropout rate in this study design; need for large sample size for adequate power; baseline metabolic heterogeneity of TRD requiring careful subject characterization for future studies of interventions; relationship of amygdala activity decreases with response; and the relationship between baseline sgACC and VMPFC activity with response. Successful treatment of TRD with olanzapine-fluoxetine combination shows changes in cerebral metabolism like those seen in treatment-responsive major depression., Competing Interests: Eli Lilly & Co, a commercial source, provided partial funding for this work. The contents herein do not express the views of the Department of Veterans Affairs or the US government. Parties interested in data sharing should contact the Minneapolis Veterans Health Care System IRB (One Veterans Drive, Minneapolis, MN 55417) to establish a data use agreement. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

The application of machine learning techniques to psychiatric neuroimaging offers the possibility to identify robust, reliable and objective disease biomarkers both within and between contemporary syndromal diagnoses that could guide routine clinical practice. The use of quantitative methods to identify psychiatric biomarkers is consequently important, particularly with a view to making predictions relevant to individual patients, rather than at a group-level. Here, we describe predictions of treatment-refractory depression (TRD) diagnosis using structural T1-weighted brain scans obtained from twenty adult participants with TRD and 21 never depressed controls. We report 85% accuracy of individual subject diagnostic prediction. Using an automated feature selection method, the major brain regions supporting this significant classification were in the caudate, insula, habenula and periventricular grey matter. It was not, however, possible to predict the degree of 'treatment resistance' in individual patients, at least as quantified by the Massachusetts General Hospital (MGH-S) clinical staging method; but the insula was again identified as a region of interest. Structural brain imaging data alone can be used to predict diagnostic status, but not MGH-S staging, with a high degree of accuracy in patients with TRD.

Several studies demonstrated improvement of depressive symptoms in treatment resistant depression (TRD) after administering dopamine agonists which suggest abnormal dopaminergic neurotransmission in TRD. However, the role of dopaminergic signaling through measurement of striatal dopamine D(2/3) receptor (D2/3R) binding has not been investigated in TRD subjects. We used [(123)I]IBZM single photon emission computed tomography (SPECT) to investigate striatal D2/3R binding in TRD. We included 6 severe TRD patients, 11 severe TRD patients on antipsychotics (TRD AP group) and 15 matched healthy controls. Results showed no significant difference (p = 0.75) in striatal D2/3R availability was found between TRD patients and healthy controls. In the TRD AP group D2/3R availability was significantly decreased (reflecting occupancy of D2/3Rs by antipsychotics) relative to TRD patients and healthy controls (p<0.001) but there were no differences in clinical symptoms between TRD AP and TRD patients. This preliminary study therefore does not provide evidence for large differences in D2/3 availability in severe TRD patients and suggests this TRD subgroup is not characterized by altered dopaminergic transmission. Atypical antipsychotics appear to have no clinical benefit in severe TRD patients who remain depressed, despite their strong occupancy of D2/3Rs.