A web-based clinical decision-support tool that personalizes antidepressant selection based on patient characteristics and preferences reduced treatment discontinuation by approximately 40% at 8 weeks compared with usual care in a randomized clinical trial of adults with major depressive disorder.
Background
Major depressive disorder affects an estimated 280 million patients worldwide and is forecast to remain a leading contributor to global health burden through 2050. Antidepressants are recommended as first-line therapy in patients with moderate to severe symptoms, yet many patients discontinue treatment prematurely. Early adverse effects contribute to discontinuation, and physicians lack reliable tools to predict which medication will be best tolerated or most effective for an individual patient.
Previous attempts to personalize antidepressant treatment — including clinical subtyping, pharmacogenomics, imaging, and other biomarker approaches — have not yielded reproducible or clinically meaningful predictors of medication response.
The PETRUSHKA Tool
The Personalising Antidepressant Treatment for Unipolar Depression Combining Individual Choices, Risks and Big Data (PETRUSHKA) tool is an online platform developed with patient input. It uses statistical and machine learning prediction models trained on randomized trial and electronic health record data to generate a ranked list of three antidepressants for each patient.
Physicians enter demographic and clinical variables, including age, sex, ethnicity, socioeconomic status, body mass index, smoking status, depression severity, prior antidepressant use, childhood maltreatment, anxiety, comorbid conditions, and concomitant medications. Patients then select and rank five adverse effects they most want to avoid.
The platform generates individualized recommendations to support shared decision-making between physicians and patients.
Trial Design
The multicenter, superiority randomized clinical trial enrolled adults aged 18 to 74 years with major depressive disorder who had an indication for pharmacologic treatment and were willing to take antidepressant monotherapy.
The study was conducted at 47 sites in Brazil, Canada, and the United Kingdom, primarily in primary care settings. Patients with treatment-resistant depression, arrhythmias or QT prolongation, pregnancy, or need for urgent hospitalization were excluded.
A total of 540 patients were randomized 1:1 to the PETRUSHKA tool (n = 271) or usual care (n = 269). After excluding 20 patients found ineligible following randomization, 520 were eligible, and 493 were included in the primary analysis.
Follow-up occurred at 4, 8, and 24 weeks. The primary outcome was all-cause treatment discontinuation at 8 weeks.
Results
At 8 weeks:
-
17% of patients in the PETRUSHKA group discontinued treatment
-
27% in the usual care group discontinued treatment
This represented an approximately 40% relative reduction in discontinuation.
Discontinuation due to adverse events at 8 weeks was:
-
9% in the PETRUSHKA group
-
16% in the usual care group
The effect was more pronounced in primary care (16% vs 28%). No statistically significant difference was observed among patients treated in specialty psychiatric settings.
There were no statistically significant differences in self-rated depression or anxiety scores at 8 weeks. At 24 weeks, however:
-
Mean 9-Item Patient Health Questionnaire score was 7.1 in the PETRUSHKA group vs 9.2 in usual care
-
Mean 7-Item Generalized Anxiety Disorder questionnaire score was 4.6 vs 5.8
Observer-rated scales showed directional trends favoring the PETRUSHKA tool, but with wide uncertainty. Health-related quality of life also favored the PETRUSHKA group at 24 weeks.
Safety
Two serious adverse events occurred in the PETRUSHKA group:
-
One death attributed to undiagnosed metastatic breast cancer
-
One hospitalization for medication overdose without suicidal intent
Neither event was judged related to the tool or the prescribed medication.
Editorial Perspective
In an accompanying editorial, Gregory E. Simon, MD, of Kaiser Permanente Washington Health Research Institute, noted that prior efforts to personalize antidepressant treatment using biomarkers have largely failed to produce clinically useful predictors.
He emphasized that patient preference and baseline clinical characteristics can be assessed during routine visits without added cost or delay, unlike biomarker-based approaches.
Dr. Simon also raised the possibility that the observed benefit may reflect improved medication matching, greater patient engagement through shared decision-making, or a shift in prescribing patterns — or a combination of these factors.
“While clinicians wait for genomics, proteomics, or another field to reveal underlying biology and transform care for depression, patient preference appears to offer important clinical benefit with little risk, cost, or treatment delay,” he wrote.
Limitations
The trial was open-label, meaning patients and physicians were aware of treatment allocation, raising the possibility of disappointment effects in the usual care group. Substantial missing data at 24 weeks limited interpretation of longer-term outcomes. The patient population was predominantly White, which may limit generalizability.
The tool did not include all available antidepressants because patient-level data were unavailable for some agents.
Disclosures can be found in the research and editorial.
