User-friendly Tool Predicts Antidepressant Response

Megan Brooks

January 27, 2016

Researchers have developed an algorithm to help physicians predict whether a patient with depression will respond to the selective serotonin reuptake inhibitor (SSRI) citalopram (Celexa, Forest Laboratories, Inc).

"We see this tool being used in the first line so people may be more likely to be on a drug that is going to work for them sooner," which would save time and money, lead researcher Adam Chekroud, doctoral candidate in the Human Neuroscience Laboratory at Yale University, in New Haven, Connecticut, told Medscape Medical News.

"If the model predicts that you aren't going to respond, you probably should just start with another drug and may have saved 3 months of taking the wrong drug," he added.

The research was published online January 20 in the Lancet Psychiatry.

Precision Medicine

Using patient-reported data from patients with depression in the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) trial, the Yale team identified 25 variables that were most predictive of treatment outcome. They used this information to train a machine-learning algorithm to predict clinical remission with citalopram.

The model, which was internally cross-validated, predicted outcomes in the STAR*D cohort with accuracy "significantly above chance" (64.6%, P < .0001), the researchers report in their article. The model outperformed a pilot sample of 23 psychiatrists and residents; their mean accuracy in predicting outcomes for 26 STAR*D patients was 49.3% (chance was 53.9%).

The model achieved an average area under the curve of 0.70, "suggesting sufficient predictive signal in the 25 questions," the researchers say.

The model prospectively identified 62.8% of patients who eventually reached remission (sensitivity) and 66.2% of patients who failed to achieve remission (specificity). The positive and negative predictive values were 64.0% and 65.3%, respectively.

The model was externally validated in the escitalopram (Lexapro, Forest Laboratories, Inc) treatment group of the Combining Medications to Enhance Depression Outcomes (COMED) clinical trial, with accuracy of 59.6% (P = .043).

The model performed significantly above chance in COMED patients who received the combination of escitalopram and bupropion (multiple brands), with accuracy of 59.7% (P = .023), but not in those taking the serotonin-norepinephrine reuptake inhibitor (SNRI) venlafaxine (multiple brands) plus mirtazapine (Remeron, Organon Pharmaceuticals, Inc), with accuracy of 51.4% (P = .53).

"It worked on the SSRI and the SSRI with another drug, but it didn't work on the SNRIs, so we think the model may have picked up something about SSRIs," Chekroud explained.

Trial and Error

Currently, only about 30% of depressed patients achieve remission with initial drug therapy. In most patients, there is a prolonged period of trial and error, which delays clinical improvement and increases both risk for the patient as well as cost. "Patients and doctors really have no idea whether an antidepressant is going to work when they start taking it," Chekroud commented.

Psychiatry lacks precision-medicine tools, in contrast to other areas of medicine, such as oncology, cardiology, and critical care, in which algorithms are often used to guide treatment decisions, the researchers note in their article.

They say their findings are a "step in the direction of precision medicine for psychiatry. The model uses easy-to-obtain (patient-reportable) information, and could be hosted online or in the clinic using a mobile device, laptop, or desktop computer."

"This is one of a growing number of studies that are using baseline information to predict differential depression treatment response," Ronald C. Kessler, PhD, of Harvard Medical School, in Boston, Massachusetts, who was not involved in the study, told Medscape Medical News.

"It is becoming increasingly clear from such studies that useful distinctions can be made among patients prior to beginning treatment to personalize the selection of treatments. A practical, integrated system to do this is not yet available but almost certainly will become available over time, based on the accumulation of findings such as those reported here," Dr Kessler said.

Carmine M. Pariante, MD, PhD, FRCPsych, of the Institute of Psychiatry, King's College London, United Kingdom, told Medscape Medical News that this article "addresses an important clinical need ― can we predict before starting an antidepressant whether someone will actually benefit?

"Using clinical information to do this is one approach," Dr Pariante said, "but there is also a lot of work going on with blood biomarkers to personalize antidepressant response. My feeling is the best algorithm will combine perhaps five to 10 clinical variables integrated with perhaps two or three key biomarkers, which can really bring us closer to 100% predictive value," said Dr Pariante, who was not involved in the study.

Yale University provided funding for this research. Several authors reported relationships with pharmaceutical companies, which are listed in the original article.

Lancet Psychiatry. Published online January 20, 2016. Abstract


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: