Researchers at Trinity College in Dublin have found that a machine learning model could help clinicians predict which patients with depression are more likely to improve with digital cognitive behavioral therapy than with antidepressants.
The study, led by researchers in the Department of Psychology, also describes how digital cognitive behavioral therapy (CBT) can be individualized more quickly than other settings, such as face-to-face therapy. Because it is already digital and measurements can be included from the beginning.
Digital CBT
Research published in journals JAMA network open, Data from 883 adults receiving digital CBT or antidepressants was analyzed. This study was designed to predict changes in early symptoms of depression. Participants who received digital CBT took an online CBT course over a 4-week period.
This model could explain 19% of the variance in how much patients’ depression improved after 4 weeks of digital CBT, and importantly was unique to the treatment. In other words, they could not predict people’s reactions to antidepressants in the same way.
Professor Claire Gillan, from the lab who led the study, said: “While 19% may seem modest, given the scale of global depression treatment disparities, even small improvements in the ability to allocate treatments more effectively could have a huge impact on health and wellbeing, quality of life and the economic burden of the disease.”
“Millions of people around the world suffer from depression, and treatment responses vary widely from person to person,” said lead author Dr. Sharon Chi Tak Lee, who conducted the study in Gillan’s lab at Trinity University’s School of Psychology.
“Currently, clinicians rely on a trial-and-error approach to find out which treatments are most effective for each patient. This study shows that information provided by people at the start of treatment, particularly questionnaires, can be used to predict who is likely to improve faster with digital CBT.”
“It’s important to note that machine learning models are not meant to replace clinicians. Our study found that the models identify some, but not all, patients who would benefit from digital CBT, so they are better viewed as decision support tools to help clinicians match the right type of treatment more quickly. That said, it has great potential to alleviate suffering and ease the burden on healthcare systems.”
A growing number of studies are using machine learning to predict how people with depression will respond to treatment, but many of the early studies had small datasets or poorly validated models. This study addresses this gap with a larger number of participants and a more powerful test of treatment specificity.
Learn more about the research
The 4-week prognostic study used a fully digital protocol designed to predict early symptom changes in depression. A total of 883 participants completed the baseline and final assessments.
The digital CBT group included a total of 776 participants and was recruited through Irish mental health charity Aware and NHS Talking Therapies clinics in the UK. A total of 107 people were in the antidepressant group and were recruited from around the world through online and print advertisements.
The open source paper can be read in full on the journal’s website.
