Key Takeaways

• PETRUSHKA, an AI-powered prescribing tool developed by the University of Oxford, helps clinicians match patients with depression to the antidepressant most likely to work for them.
• In a randomised trial of more than 500 adults with major depressive disorder across the UK, Brazil, and Canada, patients using PETRUSHKA were around 40% less likely to stop their medication within the first eight weeks.
• By 24 weeks, patients in the PETRUSHKA group reported greater improvements in depression and anxiety symptoms than those receiving usual care.
• The tool takes about three minutes to run and is designed for use during routine consultations, including in primary care.
• Findings were published in the Journal of the American Medical Association (JAMA), marking the first time a mental health clinical prediction tool has been shown to be effective in a large trial.
• The study was funded by the National Institute for Health and Care Research (NIHR) and ran across 47 sites internationally.

---

What the AI Tool Does and Why It Matters

A new AI prescribing tool called PETRUSHKA, developed at the University of Oxford, has been shown to significantly improve antidepressant adherence and clinical outcomes for people with depression. In an international randomised trial, patients whose medication was selected with PETRUSHKA were roughly 40% less likely to quit their treatment within the first two months and reported better depression and anxiety outcomes after six months compared with those prescribed antidepressants the usual way.

The result matters because antidepressant prescribing has long been a guessing game. Current estimates suggest around 80% of patients stop taking antidepressants within a few weeks, often because of side effects or because the drug simply isn’t working. PETRUSHKA, whose full name stands for Personalising antidEpressant Treatment foR Unipolar depreSsion combining individual cHoices, risKs and big datA, attempts to fix that problem at the point of prescription rather than after months of unsuccessful attempts.

How PETRUSHKA Works

The tool combines three streams of information: clinical evidence from antidepressant trials, real-world data on how medicines perform in practice, and the individual patient’s own profile and preferences. That patient input is central. Rather than asking a clinician to guess which side effects a person might tolerate, the tool asks the patient directly — weight gain, sexual dysfunction, sleep disturbance, emotional blunting, and other common trade-offs — and feeds those answers into its recommendation.

Demographic details, medical history, and symptom severity are also factored in. The algorithm then ranks the antidepressants most likely to suit the specific person sitting in the consultation room.

The whole process takes about three minutes and is intended for use during a normal appointment, either in person or over video. Clinician and patient go through it together, which turns the prescription decision into a genuinely shared one.

Inside the International Trial

The PETRUSHKA trial, launched in 2024, enrolled more than 500 adults with major depressive disorder (MDD) across 47 sites in the United Kingdom, Brazil, and Canada. Half received usual care; the other half had their antidepressant selected using the tool.

The findings were published in the Journal of the American Medical Association and represent, according to the research team, the first time a clinical prediction tool in mental health has been validated as effective in a trial of this scale.

The Lead Researcher’s View

Professor Andrea Cipriani, NIHR Research Professor at the University of Oxford and honorary consultant psychiatrist at Oxford Health NHS Foundation Trust, led the investigation. Speaking about the significance of the results, he said:

“Mental health is lagging behind other fields of medicine and for too long, antidepressant treatment has relied on trial and error. PETRUSHKA shows that by combining the best available evidence with patients’ own preferences, we can personalise antidepressant treatment from the outset and help more people in the NHS stay on the medication that is right for them.”

On the 40% adherence gain specifically, he added:

“The results of the tool are outstanding because if we use the PETRUSHKA tool, we increase by 40% the probability of continuing the medication. And this is important because continuing the medication means that the medication is effective and tolerable. And if you continue taking the medication the symptoms improve and this is what we demonstrated in the PETRUSHKA trial, reducing both depressive and anxiety symptoms.”

Cipriani described standard prescribing as “a lengthy process and also it is harmful for patients,” arguing that the tool replaces guesswork with evidence.

A Patient’s Experience

Henry Winchester, 45, a freelance writer from Bristol, had lived with depression since university and spent five years cycling through medications that caused severe side effects. He was one of the trial’s 500 participants.

“I was quite sceptical of antidepressants, I’d never really got on with them and had had severe side effects,” he said. “PETRUSHKA found one that was much milder than what I had taken before. It’s been life changing in a way, having one tailored, and finding the right one for me. It could change the way antidepressants are prescribed and make it a lot easier for patients and better for GPs.”

He described the trial process as straightforward — a health survey, questions about which side effects he wanted to avoid, then a recommendation — and said the tool gave him “an inner confidence that I’ve never really had before.”

Why Adherence Is the Right Target

Antidepressants only work if people take them, and the early weeks of treatment are when most patients give up. Side effects tend to peak before therapeutic benefit kicks in, creating a window where discouragement is common. By front-loading the decision with data about what the patient can tolerate, PETRUSHKA reduces the chance that the first prescription becomes the last one.

The economic case follows from the clinical one. Failed prescriptions mean repeat appointments, escalating symptoms, lost productivity, and, in severe cases, hospital admissions. Getting the first choice right saves all of that.

Primary Care Is the Real Prize

Most people with depression in the UK are treated by GPs, not psychiatrists. That’s where PETRUSHKA’s three-minute runtime and shared-decision format become especially useful. A general practitioner with ten minutes per patient does not have time for lengthy psychiatric workups, and specialist referral queues are long in many regions. A tool that produces an evidence-based, patient-informed recommendation inside a normal consultation fits the rhythm of primary care.

The research team sees particular value in settings where specialist psychiatric support is limited — whether that means rural areas of the UK or health systems in countries with fewer mental health resources.

The NIHR Perspective

The trial was funded by the National Institute for Health and Care Research (NIHR), with additional support from the NIHR Oxford Health Biomedical Research Centre and the NIHR Applied Research Collaboration Oxford and Thames Valley.

Mike Lewis, NIHR’s Scientific Director for Innovation, said:

“This study shows the real power of combining digital technology and personalised treatment. By harnessing data and embracing cutting-edge digital tools, we can tailor care more precisely to each patient — improving outcomes for individuals. Forward-thinking research such as this shows the impact of using innovative technologies to ensure patients across the UK receive smarter, more effective care.”

What Comes Next

The Oxford team wants to extend the approach to other mental health conditions and is working toward national rollout across UK GP surgeries. Further research is planned on long-term outcomes and cost-effectiveness — questions the six-month trial window did not fully answer.

Co-production with people who have lived experience of depression was part of PETRUSHKA’s design from the start, and the research team intends to keep patient involvement central as the tool expands. For a field where precision medicine has made slower progress than in oncology or cardiology, the trial offers a concrete example of what patient-centred AI can deliver when it is built around real clinical problems rather than retrofitted onto them.

If you are interested in this topic, we suggest you check our articles:

• How Can AI Help With Healthcare: 10 New Problem-Solving Solutions
• AI in Predictive Diagnostics: How Early Detection Is Saving Lives and Reducing Healthcare Strain
• AI in Healthcare: 8 Use Cases Making a Positive Impact

Sources: BBC, NIHR

Written by Alius Noreika