BACKGROUND
Lack of quantifiable biomarkers is a major obstacle in making diagnosis and predicting treatment response in depression. In adolescents, increasing suicidality during antidepressant treatment further complicate the problems. Emerging healthcare systems based on digital technology are beginning to show promising results in dealing with mental health issues.
OBJECTIVE
Using Smart Healthcare System for Teens At Risk for Depression and Suicide (STAR-DS) smartphone application and machine learning, we sought to evaluate digital phenotypes which represent the diagnosis and treatment response of depression in adolescents.
METHODS
Our study included 24 adolescents (15.4±1.4 years, 17 girls) with major depressive disorder (MDD) diagnosed with K-SADS-PL and 10 healthy controls (13.8±0.6 years, 5 girls). Their depression status was evaluated using the Children’s Depression Rating Scale–Revised (CDRS-R) and CGI-S every week during the study period. After collecting the baseline data for 1 week, MDD adolescents were treated with escitalopram in an 8 week, open-label trial. Both MDD and control groups were monitored for another 4 weeks after the baseline week.
We applied deep learning approach for the analysis of data. Deep Neural Network (DNN) was employed for classification and NEural network with Weighted Fuzzy Membership functions (NEWFM) for feature selection. We extracted features from directly collected data via the mobile phone (the number and total time of calls and text messages sent or received, mobile phone usage time, movement distance, amount of activity measured by gyroscope) on a daily basis. The distance from the mean value and standard deviation of each features per week were also extracted.
RESULTS
We could predict the diagnosis of depression with training accuracy of 96.3% and 3-fold validation accuracy of 77%. Of 24 depressed adolescents, 10 responded to antidepressant treatment. Including data on medications taken by the MDD group, we predicted the treatment response of depressed adolescents with training accuracy of 94.2% and 3-fold validation accuracy of 76%.
CONCLUSIONS
The STAR-DS smartphone application demonstrated preliminary evidence of predicting diagnosis and treatment response in depressed adolescents. This is the first study to predict treatment response of MDD in adolescents, examining smartphone based objective data with machine learning approaches.
Leveraging Machine Learning Approaches for Predicting Antidepressant Treatment Response Using Electroencephalography (EEG) and Clinical Data