AI trained to tailor antidepressants for patients

An international group of researchers developed a machine-learning algorithm capable of predicting a patient’s response to treatment with the antidepressant “Sertraline” from electroencephalography (EEG) data, with an accuracy of 83.7%.

Depression is a common mental disorder that affects a person’s well-being. Despite a large number of existing medications, many people do not respond to the first or even the second prescribed drug. As a result, doctors are often forced to search for an effective medicine by trial and error, which can take months or years.

According to Mariam Ravan, a researcher at the New York Institute of Technology (NYIT), the current approach to prescribing antidepressants to people diagnosed with depression is highly inefficient.

“The absence of biomarkers makes this field of medicine dependent on personal conversations and patient reports. We decided to test whether AI could provide more accurate clinical recommendations,” she said.

They analysed EEG data from 228 patients with major depressive disorder before they started taking any medication. They randomly split the participants into two groups: one received a placebo, and the other — Sertraline.

Then researchers used machine-learning algorithms to determine which patients responded to either treatment.

“[The placebo effect] may be based on the patient’s belief, confidence in the scientists’ professionalism, the passage of time, or have a biological basis reflected in measurable patterns of brain activity,” explained Ravan.

She added that there is a need to improve understanding of how people respond to taking a placebo. Such information could help treat those for whom it would be beneficial.

As a result, the machine-learning algorithm was trained to correctly predict a patient’s response to placebo in 83% of cases.

Ravan noted that AI systems require large datasets to transfer their results from the lab to the real world, and the project used a small dataset.

“If our algorithms are indeed as accurate as we think, their deployment will lead to a substantial improvement in the effectiveness and efficacy of psychiatric treatment,” she said.

According to Gary Heisi, a researcher at McMaster University, the team is already working on commercialising and broader deployment of the system through the startup Digital Medical Experts.

Researchers are also studying the possibility of using machine learning to identify people with suicidal tendencies.

“We conducted a study with 68 people diagnosed with major depressive disorder, in which we were able to identify thoughts of suicide using AI and EEG data with 70% accuracy,” said Ravan.

The team continued testing and training the algorithms on a larger dataset.

In September, American researchers were involved in the development of software for diagnosing diseases by voice.

In July, the startup Deep Longevity, in collaboration with Harvard Medical School, created an AI-enabled system to improve mental health.

In September 2021, journalists revealed that Apple was working on an AI tool for diagnosing depression and autism using the iPhone.

Subscribe to ForkLog news on Telegram: ForkLog AI — all the news from the AI world!