Stanford: An AI Can Now Help Predict the Individual Response of Patients to Antidepressants

By using a simple electroencephalogram (EEG), researchers at Stanford were able to predict which depressed patients would benefit most from treatment with the antidepressant sertraline (Zoloft).

Depressed Person

With only a few electrodes and a little artificial intelligence, it would be possible to predict which antidepressive treatment would most likely help the patient. A simple electroencephalogram combined with analysis by a computer model would thus make it possible to identify patients who could benefit from sertraline (Zoloft). This new work is published in Nature Biotechnology.

30% do not respond to treatments for depression

“I would be surprised if it is not used by clinicians in the next five years,” said Amit Etkin, professor of psychiatry and behavioral science at Stanford, who led the work. In the US, one in five people have suffered or will suffer from depression, and only 30% of them respond to the prescribed treatment. “Treatment for people suffering from depression usually begins with the prescription of an anti-depressant. If this does not work, a second antidepressant is prescribed. Each of these ‘trials’ usually takes at least eight weeks to assess whether the drug has worked and whether the symptoms are relieved,” says Etkin. If this fails, psychotherapy and transcranial brain stimulation (a treatment that involves a specific type of non-invasive brain stimulation) can also be offered. “Current methods of diagnosing depression are simply too subjective and imprecise to help physicians quickly determine the appropriate treatment,” explains Amit Etkin, and leads to the discouragement of patients, which contributes to the symptoms of depression.

Sadness, lack of desire, lack of energy, sleep disorders… The diagnosis of depression is based on many of the symptoms reported by the patients, which leads to the choice of treatment. To be more effective, doctors dream of personalized medicine based on more precise signs.

Stanford University used data from the EMBARC study, the largest image-guided, placebo-controlled antidepressant study of its kind, which included 309 depressed patients with and without an antidepressant called Sertraline (Zoloft). An electroencephalography (EEG) test, in which electrodes were placed on the patients’ scalps, was used to measure the electrical activity of the brain before starting treatment. Using theories from neuroscience, clinical science, and biotechnology, the scientists also built an advanced predictive model based on a new machine-learning algorithm called SELSER, which specializes in the analysis of EEG data. The researchers then applied SELSER to the EEG data of the 309 study participants and hoped that the machine learning technique could predict the depressive symptoms of the participants after treatment.

A more active prefrontal cortex

“These studies were more successful than any member of our team could have imagined,” says psychiatrist Madhukar Trivedi, who led the study. SELSER was able to reliably predict the individual response of patients to sertraline based on a specific type of brain signal called alpha waves. From a physiological point of view, alpha waves, in conjunction with a general state of relaxation, reduce the processing capacity of a specific area of the brain. The prefrontal cortex is particularly affected, an area of activity that is often impaired in depressed people. The prefrontal cortex is involved in our emotional reactions, particularly via other regions of the brain, which are responsible for controlling the neurotransmitters dopamine, norepinephrine, and serotonin, which are important for mood regulation. According to the authors, these findings suggest that “the prefrontal cortexes of those who respond best to treatment are more active or excitable than those who respond poorly.

This EEG-based model surpassed conventional models that used EEG data or other types of individual data, such as the severity of symptoms and demographic characteristics. Finally, another independent data set showed that participants in whom SELSER found a small improvement with sertraline were more likely to respond to transcranial magnetic stimulation in combination with psychotherapy.

“It can be devastating for a patient if an antidepressant does not work,” concludes Madhukar Trivedi. “Our research shows that they soon no longer have to endure the painful process of trial and error.”

References

An electroencephalographic signature predicts antidepressant response in major depression

Brain-wave pattern can identify people likely to respond to antidepressant, study finds