A web-based tool that personalizes antidepressant selection based on clinical characteristics and patient preferences was associated with a significant decrease in early treatment discontinuation in adults with major depressive disorder (MDD).

Patients whose antidepressants were selected using the decision-support system were 38% less likely to stop taking their medication within the first 8 weeks of treatment compared with those who received usual care.

Discontinuation due to adverse events was also markedly lower, and at 24 weeks, patients in the intervention group showed greater improvements in depression and anxiety symptoms.

“The principle behind the tool is to support collaborative decision-making between a patient and a physician to pick an antidepressant,” investigator Benoit Mulsant, MD, professor of psychiatry at the University of Toronto, Toronto, Ontario, Canada, told Medscape Medical News.

Results were significant only in patients treated in primary care settings, although the study may have been underpowered to measure the tool’s efficacy in a psychiatric clinic, investigators noted.

The study was published online on March 4 in JAMA.

A Different Approach

Only 20%-25% of patients with MDD achieve remission with their initial antidepressant. Early discontinuation — often driven by adverse events — undermines clinical effectiveness, and efforts to personalize treatment using pharmacogenomics have produced mixed results.

The PETRUSHKA tool (Personalizing Antidepressant Treatment for Unipolar Depression Combining Individual Choices, Risks and Big Data) takes a different approach.

Built on prediction models derived from individual-patient data across hundreds of randomized trials and electronic health records, the tool incorporates clinical and demographic characteristics — including BMI, symptom severity, prior antidepressant use, and comorbidities — to predict how each of 16 available antidepressants would perform for a given patient.

Patients then select and rank five side effects they most want to avoid, adjusting their relative importance using sliding bars. The tool combines these predictions with patient preferences to generate a personalized list of three recommended antidepressants.

To study treatment discontinuation rates with PETRUSHKA, investigators led a randomized controlled trial in 47 clinical sites in Brazil, Canada, and the United Kingdom. Primary care settings predominated, with only one psychiatric care site in Canada.

Investigators randomized 540 adults aged 18-74 years with moderate-to-severe MDD to an intervention group or usual care. The primary analysis included 493 participants (median age 35 years; 58% female).

In the intervention group, the PETRUSHKA tool was used to select the antidepressant. Patients in the usual care group were prescribed an antidepressant chosen by a clinician before randomization.

The primary outcome was all-cause treatment discontinuation. Secondary outcomes were adverse event-related discontinuation and patient-reported changes in symptoms of depression, measured with the Nine-item Patient Health Questionnaire (PHQ-9), and anxiety, measured with the Seven-item Generalized Anxiety Disorder (GAD-7) questionnaire.

Shift in Prescribing Patterns

At 8 weeks, participants in the PETRUSHKA group were 38% less likely to have discontinued their antidepressant than those receiving usual care (discontinuation rates, 17% vs 27%; adjusted relative risk [aRR], 0.62; P = .007). Discontinuation specifically due to adverse events was also lower at 9% vs 16% (P = .04).

By 24 weeks, the discontinuation gap had narrowed and was no longer statistically significant (34% intervention vs 40% usual care). The trial included no interim assessment between 8 and 24 weeks, making it difficult to determine when the adherence advantage began to erode.

However, symptom improvements favoring the PETRUSHKA group emerged at 24 weeks. Patients in the intervention group had a mean PHQ-9 depression score of 7.1 vs 9.2 for those in the usual care group (adjusted mean difference, -1.92; P < .001), and mean GAD-7 anxiety scores of 4.6 vs 5.8 (P = .002). Quality of life also favored the intervention group.

The tool shifted prescribing patterns considerably. In the PETRUSHKA group, the most commonly prescribed antidepressants were mirtazapine (29%), escitalopram (28%), and vortioxetine (24%). In usual care, sertraline dominated at 52%, followed by citalopram (15%) and fluoxetine (9%). Vortioxetine was prescribed to less than 1% of usual care patients.

Notably, a 2018 meta-analysis in The Lancet found higher population-level discontinuation rates for mirtazapine and vortioxetine than for sertraline — yet the PETRUSHKA group, which heavily prescribed those drugs, had better adherence.

Limitations included the trial’s open-label design, and observer-rated depression scales did not show significant differences at 24 weeks, though patient-reported measures did.

Benefits Limited to Primary Care

Subgroup analysis revealed that benefits were largely confined to primary care, where 16% in the PETRUSHKA group discontinued treatment vs 28% in the usual care group (aRR, 0.57; P = .002). Among patients treated by psychiatrists, the difference in discontinuation rates between groups was not statistically significant.

Mulsant cautioned against over-interpreting the finding. With more than 400 patients in primary care but only about 40 in psychiatric settings — just 6% of the total sample — the study was likely underpowered to detect a benefit among specialist-treated patients, he noted.

Still, he acknowledged that psychiatrists, having more expertise with antidepressants, may already incorporate elements of the tool’s prediction measures in their clinical practice.

The researchers are now launching a follow-up trial that will test whether adding pharmacogenetic data to the tool further improves outcomes. That study will enroll patients in the UK, Pakistan, and Ethiopia.

Decision-support tools are already standard in other specialties, Mulsant noted, yet psychiatry has lagged. “It already exists in the rest of medicine. Why don’t we do it in psychiatry yet?”

Power of Shared Decision-Making

In an accompanying editorial, Gregory E. Simon, MD, MPH, of Kaiser Permanente Washington Health Research Institute in Seattle, praised the tool’s real-time usability and the study’s approach for personalized antidepressant.

“Given the current disappointing state of knowledge regarding subtypes of depression and mechanisms of action in antidepressant medications, personalization based on differences in expected effectiveness and adverse events will likely benefit patients sooner than personalization based on differences in presumed therapeutic mechanism,” Simon wrote.

Incorporating patients into the treatment decision is also key, Manish Jha, MD, of UT Southwestern Medical Center in Dallas, told Medscape Medical News.

“The use of shared decision-making is a very powerful concept,” said Jha, who was not part of the study. “Enlisting people so that they are more engaged in making that treatment decision” may itself explain why patients stayed on their medications longer.

But Jha raised concerns about what happened after prescribing. The study does not describe whether clinicians in either group used measurement-based care — systematic monitoring of side effects, adherence, and dose titration — which the STAR*D trial showed is critical to optimizing outcomes regardless of which antidepressant is chosen.

Jha also flagged the tool’s heavy recommendation of vortioxetine — prescribed to nearly a quarter of patients in the intervention group vs less than 1% of those in the usual care group. Because the algorithm was trained on data shared by pharmaceutical companies, its recommendations may reflect which companies made their data available rather than a true clinical signal. That has implications for health equity, he noted, because vortioxetine is typically a brand-name medication that may not be covered by insurance in all settings.

“Even 8 weeks may not be long enough,” Jha said of the primary time point. “We need to look for longer-term treatment outcomes.”

The study was funded by the National Institute for Health and Care Research. Mulsant reported receiving funding from the University of Toronto and grants from the CAMH Foundation, Brain Canada, and the Canadian Institutes of Health Research. Simon reported receiving grants from the National Institute of Mental Health. Jha reported having no relevant disclosures.