HMN 2026: How Machine learning could help predict how people with depression respond to treatment

Researchers at Trinity College Dublin have found that a machine learning model could help clinicians predict which people with depression are more likely to improve with digital cognitive behavioral therapy compared to antidepressant medication. The study, led by researchers in the School of Psychology, also describes how digital cognitive behavioral therapy (CBT) can be personalized sooner than in other settings, such as face-to-face therapy. This is because it’s already digital and measurements can be built in from the start.

The study, published in the journal JAMA Network Open, analyzed data from 883 adults receiving either digital CBT or antidepressant medication. The study was designed to predict early symptom change in depression. The participants who received digital CBT worked through an online CBT course over a four-week period.

The model could account for 19% of the variance in how much patients’ depression improved after four weeks of digital CBT and importantly, was specific to that treatment. That is, it didn’t similarly predict people’s response to antidepressant medications.

“While 19% may seem modest,” said Professor Claire Gillan, whose lab led the study, “given the scale of the global depression treatment gap, even small improvements in our ability to allocate treatments more effectively could have a substantial impact on health and well-being, quality of life, and the economic burden of disease.”

“Depression affects millions of people worldwide and treatment responses vary quite significantly between people,” explains lead author Dr. Sharon Chi Tak Lee, who conducted the research at the Gillan Lab, in Trinity’s School of Psychology.

“Currently, clinicians rely on a trial-and-error approach to find out what treatment will work best for each patient. This study shows we can use information people provide at the start of treatment, especially questionnaires, to predict who is more likely to improve quickly with digital CBT.”

“It is important to note that a machine learning model would not replace clinicians. Our research found that the model identifies some, but not all, patients who will benefit from digital CBT, so it’s better viewed as a decision-support tool for clinicians to help match people to the right kind of care faster. That said, this has enormous potential to reduce suffering and ease burdens on our health systems.”

Research using machine learning to predict how people with depression respond to treatments has been increasing, but many earlier studies had small datasets or had poor validation of models. This study addresses this gap with a larger number of participants and stronger testing of treatment specificity.

The content is provided for information purposes only.