Abstract
Background: Deep learning has utility in predicting differential antidepressant treatment response among patients with major depressive disorder, yet there remains a paucity of research describing how to interpret deep learning models in a clinically or etiologically meaningful way. In this paper, we describe methods for analyzing deep learning models of clinical and demographic psychiatric data, using our recent work on a deep learning model of STAR*D and CO-MED remission prediction.Methods: Our deep learning analysis with STAR*D and CO-MED yielded four models that predicted response to the four treatments used across the two datasets. Here, we use classical statistics and simple data representations to improve interpretability of the features output by our deep learning model and provide finer grained understanding of their clinical and etiological significance. Specifically, we use representations derived from our model to yield features predicting both treatment non-response and differential treatment response to four standard antidepressants, and use linear regression and t-tests to address questions about the contribution of trauma, education, and somatic symptoms to our models.Results: Traditional statistics were able to probe the input features of our deep learning models, reproducing results from previous research, while providing novel insights into depression causes and treatments. We found that specific features were predictive of treatment response, and were able to break these down by treatment and non-response categories; that specific trauma indices were differentially predictive of baseline depression severity; that somatic symptoms were significantly different between males and females, and that education and low income proved important psycho-social stressors associated with depression.Conclusion: Traditional statistics can augment interpretation of deep learning models. Such interpretation can lend us new hypotheses about depression and contribute to building causal models of etiology and prognosis. We discuss dataset-specific effects and ideal clinical samples for machine learning analysis aimed at improving tools to assist in optimizing treatment.
Highlights
The heterogeneity of depression constitutes a major barrier to successful treatment (Perna et al, 2018)
Given the possibility of antidepressant-specific vs. general predictors of response, we asked: “Can we identify features predictive of response to each of the four antidepressants within our model individually, as well as to the subgroup of patients with a low probability of responding to any of the drugs?” (2) Trauma-related features Specific indices of trauma emerged from the deep learning model as predictive of treatment response for both the Sequenced Treatment Alternatives to Relieve Depression (STAR∗D) and Combining Medications to Enhance Depression Outcomes (COMED) datasets
Given the four possible medications within our model, we assessed which features were important for predicting remission [as defined by a score of 5 or less on the Quick Inventory of Depressive Symptomatology (QIDS)] for each individual drug, as well as which features were predictive of a low probability of remission with any drug
Summary
The heterogeneity of depression constitutes a major barrier to successful treatment (Perna et al, 2018). Inference creates a model of datageneration to test a hypothesis about how a particular system behaves, whereas prediction forecasts possible outcome or behavior without necessarily understanding underlying biological mechanisms (Bzdok et al, 2018) Classical statistical methods, such as regression and t-tests, focus on inference and have been a dominant method for analyzing psychiatric data and offering insight into causal associations. Logistic regression models assessing the association of demographic and clinical characteristics on treatment outcome in the Sequenced Treatment Alternatives to Relieve Depression (STAR∗D) trial, a large multicenter sequenced treatment trial for depression, have shown that race, low education, post-traumatic stress disorder (PTSD), and hypochondriasis are independently associated with worsened depression (Friedman et al, 2009), as well as depression severity, energy/fatigue, race, education, and PTSD occurrence (Perlis, 2013); in addition, having witnessed or experienced trauma has been used to estimate risk for treatment-resistance among major depressive disorder (MDD) outpatients (Perlis, 2013) These results are bolstered with receiver operating characteristic (ROC) analyses showing income and education to be predictors of response in STAR∗D (Jakubovski and Bloch, 2014). We describe methods for analyzing deep learning models of clinical and demographic psychiatric data, using our recent work on a deep learning model of STAR∗D and CO-MED remission prediction
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
