PETRUSHKA Tool Improves Antidepressant Persistence in MDD, With Andrea Cipriani, MD, PhD

A recent study found that the PETRUSHKA decision-support tool improved antidepressant treatment persistence in major depressive disorder (MDD). In an accompanying interview, Andrea Cipriani, MD, PhD, professor of psychiatry at the University of Oxford, emphasized its potential to extend personalized treatment into primary care and other nonpsychiatric settings through structured, shared decision-making.

“If you don't have a psychiatric or specific knowledge in the field, like family doctors or other specialties—neurology, cardiology, oncology, whatever—using the tool is much better than the clinical judgment of the known specialist.” Cipriani told HCPLive.

Antidepressant selection in routine care often reflects clinician familiarity rather than individualized prediction, particularly in primary care where most prescriptions originate. This pattern has limited implementation of guideline-recommended personalization strategies, despite longstanding recognition of heterogeneity in treatment response and tolerability.

The PETRUSHKA platform generates a ranked list of antidepressants using routinely collected clinical and demographic variables, including symptom severity, comorbidities, and prior treatment history, combined with patient preferences regarding adverse effects. This approach supports a 2-fold strategy of broadening pharmacologic consideration while actively incorporating patient priorities into treatment selection.

The multicenter PETRUSHKA randomized clinical trial enrolled 540 adults with MDD across 47 sites in Brazil, Canada, and the UK.¹ Participants were randomized 1:1 to a web-based decision-support system or usual care. The primary endpoint, all-cause treatment discontinuation at 8 weeks, occurred in 17% (41/241) of the PETRUSHKA group versus 27% (69/252) under usual care (adjusted relative risk, 0.62; 95% CI, 0.44-0.88; P =.007). This absolute difference of approximately 10 percentage points reflects a clinically relevant improvement in early treatment persistence.

Secondary outcomes further supported clinical benefit at 24 weeks. Mean PHQ-9 scores were 7.1 (SD, 5.4) in the PETRUSHKA group versus 9.2 (SD, 6.5) in usual care (adjusted mean difference, −1.92; 95% CI, −3.06 to −0.78; P <.001). Anxiety symptoms improved similarly, with GAD-7 scores of 4.6 versus 5.8 (adjusted mean difference, −1.39; 95% CI, −2.26 to −0.52; P =.002). These findings suggest downstream clinical benefit associated with improved early treatment matching and persistence.

Treatment discontinuation due to adverse events at 8 weeks occurred in 9% (n = 22) of the PETRUSHKA group compared with 16% (n = 39 in usual care (adjusted relative risk, 0.59; 95% CI, 0.36-0.97; P =.04). No new safety signals were identified, and serious adverse events were not attributed to the intervention or selected treatments.

Cipriani emphasized the broader significance of these findings in a field constrained by limited biomarkers. He noted the tool demonstrates meaningful gains using routinely available clinical and demographic inputs, improving prediction without reliance on advanced biological markers. He added that future development efforts aim to integrate additional data layers, including genetic and imaging information, to further refine treatment selection.

The study reinforces a shift toward shared decision-making in mental health care. The PETRUSHKA tool facilitates real-time collaboration between clinician and patient, incorporating individual preferences related to tolerability into final treatment selection. Cipriani described this as a move away from clinician-centered prescribing toward a more collaborative model, with particular relevance in primary care, neurology, cardiology, and oncology settings.

Limitations include the open-label design and missing data for some secondary outcomes, which may affect interpretability. In addition, findings may not fully generalize to specialty psychiatric settings, although subgroup analyses suggested greater benefit in primary care environments.

“The big challenge is what we call external validity, so to use the prediction for people that are not included in the building up of the prediction model. That's one word of caution for people not to trust automatically any kind of tool. We need to raise the bar in the level because we need to prove that these tools are effective in delivering what we want,” Cipriani said. “The second thing [that] is crucial is a shared decision-making process.”

Cipriani has no relevant reported disclosures.

References

Cipriani A, Fernandes KBP, Mulsant BH, et al. A Decision-Support System to Personalize Antidepressant Treatment in Major Depressive Disorder: A Randomized Clinical Trial. JAMA. 2026;335(14):1219-1231. doi:10.1001/jama.2026.1327