Machine learning can play a key role in planning treatment for patients diagnosed with major depressive disorder (MDD), according to new research published in JAMA Network Open.
“Because of the heterogeneity of depression and the lack of consensus on the precise mechanism of action of antidepressants, matching patients to effective treatments has been a daunting task for practitioners,” wrote lead author Andrey Zhdanov, PhD, Simon Fraser University in Surrey, British Columbia, and colleagues. “Currently, practitioners use a prolonged trial-and-error process to identify the optimal antidepressant for each patient, with patients often spending months to years experiencing distressing symptoms. Although clinical interviews and scales are available to confirm the diagnosis and severity of symptoms, they are not sufficient for selecting an appropriate treatment for each patient.”
Predicting how patients might response to certain medications could potentially make a huge impact on how healthcare providers treat patients diagnosed with MDD, the authors wrote. And electroencephalography (EEG), which records signals captured by electrodes attached to a patient’s scalp, stands as “one promising technique for identifying biological predictors of response to antidepressant treatment.”
The team studied baseline EEG data from 122 adult patients with MDD who had completed eight weeks of treatment with 10-20 mg of escitalopram, a common antidepressant. Using a machine learning model, support vector machine, Zhdanov et al. were able to identify patients who would respond to the escitalopram with an estimated accuracy of 79.2%, sensitivity of 67.3% and specificity of 91%.
For 115 study participants, EEG data from after the first two weeks of treatment was also available—an update that increased the estimated accuracy to 82.4% and sensitivity to 79.2%. Specificity, meanwhile, dropped to 85.5%.
“As hypothesized, combining baseline neural dynamics with early changes in neural dynamics (change after 2 weeks of treatment) resulted in the most accurate prediction,” the authors wrote. “Results from leave-one-site-out cross-validation also demonstrated that the large-scale analysis of data pooled across multiple sites did not have a significant effect on classifier performance.”
The authors also noted that many of the features identified through the machine learning process had already been observed through other studies. However, the study did reveal other features—asymmetry in complexity of neural activity between hemispheres, for example—that had not been previously reported.
“This discovery of new EEG features that predict treatment outcome was a result of systematic evaluation of a large number of candidate measures using a relatively large data sample,” the authors wrote.