AI Algorithm Identifies if SSRI Will Benefit Patients with Depression

February 15, 2020
Samara Rosenfeld

Using the AI, providers can skip the trial and error process that often comes with prescribing antidepressants.

Trivedi Madhukar, MD

Providers might have an alternative way to treat depression, according to recent findings which highlighted the successes of an artificial intelligence (AI) algorithm for predicting which patients would benefit from selective serotonin reuptake inhibitors (SSRIs).

Madhukar Trivedi, MD, and a team of investigators developed a machine-learning algorithm based on electroencephalogram (EEG) data to predict which patients would benefit from medication.

The algorithm helped identify the right treatment and reduced the trial and error process often associated with choosing antidepressants, Madhukar Trivedi, MD, a principal investigator, said in an interview with HCPLive®.

“Currently, antidepressant selection is based on clinical judgment and is not tied to a specific person’s brain circuits,” he said. “This allows you to make more objective decisions and to base them on an individual’s pathology.”

The results could begin to help arm people with better knowledge and certainty so they will stick to their treatment.

Trivedi, a psychiatrist at University of Texas Southwestern, and colleagues included more than 300 participants with depression in the study. Participants were randomly chosen to receive either a placebo or an SSRI. The team used an EEG to measure the electrical activity in the participants’ cortex prior to treatment. After the EEG, the investigators developed a machine-learning algorithm to analyze and use the EEG data to predict the patients who would benefit from the medication within 2 months.

The algorithm, called Sparse EEG Latent SpacE Regression (SELSER), was designed to reduce noise and increase brain signal, Trivedi said.

The investigator trained SELSER on data from the largest neuroimaging-coupled placebo-controlled randomized clinical study of antidepressant efficacy—Establishing Moderators and Biosignatures of Antidepressant Response in Clinic Care (EMBARC) study. EMBARC involved randomization of 309 medication-free depressed outpatients to receive the SSRI sertraline or placebo for 8 weeks.

For patients randomized to sertraline, if they had a high signature response to the medication, it meant they would benefit. When the algorithm was applied to the placebo group, there was no response, which signified the response was specific to the sertraline.

The investigators tested the generalizability of the antidepressant-predictive signature on 2 other samples of depressed patients.

One independent sample of 24 depressed patients assessed convergent validity and neurobiological significance. In a final sample of 152 depressed patients, the investigators tested whether the strength of the EMBARC-trained resting-state EEG signature predicted outcome with an antidepressant treatment with a putatively different mechanism of action.

If a provider can identify if a person has abnormalities from their sertraline signature response, then it is significantly important, Trivedi said.

The AI accurately predicted outcomes and additional research showed that patients who were doubtful to respond to an antidepressant were likely to improve from another intervention like psychotherapy or brain stimulation.

Similar to getting a lab test, a provider could be able to order an EEG for a patient, Trivedi said. Once the results are in, they can figure out what the best treatment plan is based on the signature.

The study, “An electroencephalographic signature predicts antidepressant response in major depression,” was published online in the journal Nature Biotechnology.