EEG-Based Model Predicts Antidepressant Responses in Depression Patients

Electroencephalography (EEG) can predict an antidepressant’s response, according to a new prognostic study led by Benjamin Schwartzmann, MSc, of the school of mechatronic systems engineering at Simon Fraser University in British Columbia.¹

Schwartzmann and colleagues conducted the cohort study to prevent people from receiving ill side effects from antidepressants, namely those under the serotonin reuptake inhibitor (SSRI) drug class.

Previous studies examined EEG and how it could be used to determine an antidepressant’s response, but such studies were not robust enough to be implemented in a clinical setting, or contained conflicting data. While one study claimed there was lower theta activity recorded by the EEG when responding to several SSRI medications, another study claimed there was higher theta activity when responding to the SSRI, fluoxetine.

“Failure to replicate across studies can be attributed to small sample size and the complexity and heterogeneity of depression,” investigators wrote. “Testing individual EEG features in small, underpowered studies has hindered the development of EEG-based predictive models for clinical use.”

Thus, because of the small sample size, the data were inconclusive.

A study on this topic poses challenges—one being the type of EEG technology. Other study models used high-density devices with ≥64 channels, which require a lot of training and expertise to provide quality data; but for clinical viability, the EEG Signal should be easy to record. To conquer this challenge, portable EEG devices with fewer channels serve as an important tool.

To get conclusive data, the team collected 2 independent cohorts: the Canadian Biomarker Integration Network in Depression (CAN-BIND) and the Biosignatures of Antidepressant Response for Clinical Care (EMBARC).

The CAN-BIND group was an open-label trial used for internal validation with 125 participants with a median age of 36.4 years. They received an 8-week treatment regimen of 10 – 20 mg escitalopram.

Meanwhile, the EMBARC group was a randomized double-blind trial used for external validation with 105 participants who had the median age of 38.4 years. Here, 118 participants were randomized for placebo treatment.

Other than using the two cohorts, the team’s study differentiated from previous studies because they reduced the number of EEG features.

Zhandnov’s previous study used features that came from 58 electrodes at the baseline,² but the new study only used baseline data for clinical relevance and instead stuck with 32 electrodes.

Also, to lower the number of EEG features, the 32 electrodes were “grouped into 14 brain regions.”

“Compared with research-grade EEG devices with 64 or more channels, these devices are more suitable for clinical settings with limited resources and time, prioritizing cost-effectiveness, ease of use, patient comfort, and signal quality over spatial resolution,” investigators wrote.

The study also looked for balanced accuracy, sensitivity, and specificity, which was evaluated with either the normal approximation method or the Clopper-Pearson exact method.

A machine learning approach replaced a traditional statistical analysis in this study. Eligible participants included people aged 18-64 years old who had a major depressive disorder (MDD) diagnosis. The data was analyzed January - December 2022.

The results found that age or sex made no difference in the results in both CAN-BIND and EMBARC. The data showed EMBARC had significantly lower severity at baseline than participants.

The team fed the model trained from CAN-BIND with data from EMBARC to see how well the model could generalize new data and found that the model had a balanced accuracy of 63.7%, a sensitivity of 58.8%, and a specificity of 68.5%.

The model was then again tested with data from the EMBARC placebo group and the balanced accuracy was 48.7%

As with CAN-BLIND data, the model had a 64.2% balanced accuracy.

“Using objective measures such as EEG data to predict treatment response is preferable to relying on subjective variables in clinical setting,” the team wrote. “Objective biomarkers have been suggested to reduce the self-stigma associated with seeking help and reporting symptoms on clinical questionnaires.”

The team concluded by writing it would be better, in the future, to create a model focusing on a specific class of antidepressants, since this study featured a mix of the antidepressant’s escitalopram and sertraline.

Resources

1. ^{Schwartzmann B, Dhami P, Uher R, et al. Developing an Electroencephalography-Based Model for Predicting Response to Antidepressant Medication. JAMA Netw Open. 2023;6(9):e2336094. Published 2023 Sep 5. doi:10.1001/jamanetworkopen.2023.36094}
2. ^{Zhdanov A, Atluri S,WongW, et al. Use of machine learning for predicting escitalopram treatment outcome from electroencephalography recordings in adult patients with depression. JAMA Netw Open. 2020;3(1):e1918377. doi:10.1001/jamanetworkopen.2019.18377}