A machine learning program that analyzes patients’ brain imaging data along with many clinical variables of major depression—such as symptom severity—predicted whether patients with major depressive disorder would respond to the antidepressant sertraline, according to a report in AJP in Advance.
Machine learning is a type of artificial intelligence that combines a very large number of patient variables—more than a single physician could collect—to try to reliably predict an outcome of interest for individual patients. With each new piece of data, the computer “learns” to refine its prediction—hence the term “machine learning.
Maarten G. Poirot, M.S., and Henricus G. Ruhé, M.D., Ph.D., of the University of Amsterdam and colleagues used data from 229 patients with major depression who had enrolled in the Establishing Moderators and Biosignatures of Antidepressant Response in Clinical Care (EMBARC) study, a randomized controlled trial designed to evaluate variables that predict antidepressant response. Brain MRI images and a wide variety of socioeconomic, behavioral, and neuropsychiatric variables were collected before and one week after treatment with sertraline.
The researchers first tested their machine learning program on 105 patients who received sertraline and found the program could predict treatment response after 8 weeks using both pretreatment data (patient baseline variables) or early treatment data (changes after one week) significantly better than random chance; accuracy ratings ranged from 62% to 68%. The machine learning program did not generally perform as well when assessing whether patients in the placebo group responded to treatment, indicating that the prediction tool was specific to sertraline.
Moreover, the analysis was able to pinpoint which variables were most important in the prediction. “The algorithm suggested that blood flow in the anterior cingulate cortex, the area of brain involved in emotion regulation, would be predictive of the efficacy of the drug. And at the second measurement, a week after the start, the severity of their symptoms turned out to be additionally predictive,” said Ruhé in a press release. In the article, the researchers noted that since their program would likely not need input from a second session of MRI scanning to be accurate, the cost and burden on patients would be lowered in clinical practice.
The researchers concluded, “With additional external validation, these findings will contribute toward the use of predictive modeling in individualizing clinical sertraline treatment of patients with MDD.”
For related information, see the Psychiatric News article “Research Using Machine Learning in Psychiatry Expands Rapidly.”
Don't miss out! To learn about newly posted articles in Psychiatric News, please sign up here.
