Machine-Learning Methodology May Help Predict Response to Lithium, Quetiapine in Bipolar Disorder

A novel machine-learning method, elastic net regularization, could help predict clinical responses to lithium and quetiapine in patients with bipolar disorder, according to results from a study published in Bipolar Disorders.

Researchers analyzed data from the Clinical Health Outcomes Initiative in Comparative Effectiveness study that included 482 patients with bipolar I or II disorder who were treated with lithium or quetiapine in combination with another evidence-based therapy. The technique was used to create predictive models for these 2 agents, and a test set was used to evaluate the predicative abilities of the model. Clinical responses were measured using the clinical global impressions scale-bipolar version.

After analysis, the researchers found that 17.4% and 32.1% of the variance seen in clinical scores of participants who were treated with lithium and quetiapine, respectively, were explained by predictions from the model. In addition, they reported that the predictive accuracy of elastic net regularization was better than other theoretical models.

Of the baseline variables selected, those with the largest parameter estimates were severity of mania; attention-deficit/hyperactivity disorder comorbidity, nonsuicidal self-injurious behavior, employment, and comorbidity with social phobia/society anxiety and agoraphobia.

One key limitation of the study was the lack of a prospective sample to further validate the model.

“This is the first effort to systematically examine predictors of response to lithium or quetiapine in the treatment of [bipolar disorder],” the researchers wrote.

“The further development of the tools and methods exemplified in this work should provide information that, ultimately, can be used by clinicians to make better-informed treatment decisions for their patients,” they concluded.

Reference

Kim TT, Dufour S, Xu C, et al. Predictive modeling for response to lithium and quetiapine in bipolar disorder [published online February 7, 2019]. Bipolar Disord. doi: 10.1111/bdi.12752