Finding the right antidepressant can take months of failed attempts and difficult side effects. Now, Dutch researchers have developed an AI model that predicts whether a specific antidepressant will work for an individual patient — and it does so within just one week of starting treatment.
At a glance
- Antidepressant prediction: The AI model analyzes MRI brain scans and clinical data to predict whether sertraline — the most commonly prescribed antidepressant in the U.S. — will be effective for a given patient, cutting the usual wait from six to eight weeks down to one.
- Major depressive disorder: Around 60% of people with depression don’t find a suitable medication on their first try. This model could spare two-thirds of patients an ineffective course of treatment and its accompanying side effects.
- Anterior cingulate cortex: The algorithm focuses on blood flow in this brain region, which is involved in emotion regulation, along with symptom severity measured one week after treatment begins, to make its prediction.
The study, published in the American Journal of Psychiatry, came from a collaboration between Amsterdam University Medical Center (UMC) and Radboud UMC. It represents one of the more concrete clinical applications of AI in psychiatry so far — not a theoretical model, but one tested against real patient data.
“Normally, it takes six to eight weeks before it is known whether an antidepressant will work,” said Liesbeth Reneman, Professor of Neuroradiology at Amsterdam UMC. That wait, multiplied across multiple failed prescriptions, can push people to give up on treatment entirely.
How the algorithm works
The team drew on data from a prior U.S. study involving 229 patients with depression. Before treatment began, each patient received an MRI brain scan and had clinical data recorded. They were then given either sertraline or a placebo. That full dataset was fed into the AI.
The algorithm zeroed in on two factors: blood flow in the anterior cingulate cortex before treatment, and the severity of symptoms at the one-week mark. Patients with high blood flow in that region responded well to the drug. Those without it largely did not.
“The algorithm suggested that those who had a lot of blood flow in the anterior cingulate cortex, the area of brain involved in emotion regulation, would be helped by the drug,” said Eric Ruhé, a psychiatrist at Radboud UMC. “And at the second measurement, a week after the start, this turned out to be the severity of their symptoms.”
The result: the model predicted that sertraline would work for only about one-third of participants. For the other two-thirds, it could flag early that a different approach would likely be needed — before weeks of ineffective treatment and side effects played out.
Why this matters for patients
Depression treatment is genuinely complex. The range of available drug classes — SSRIs, SNRIs, atypical antidepressants, tricyclics, and MAOIs — means clinicians are often working through options without a reliable biological guide. This is part of a broader set of public health wins emerging from AI-assisted diagnostics, where pattern recognition in biological data is beginning to outpace what’s visible to the human eye.
“With this method, we can already prevent two-thirds of the number of ‘erroneous’ prescriptions of sertraline and thus offer better quality of care for the patient,” Reneman said. “Because the drug also has side effects.”
Those side effects are not trivial. Antidepressants can take up to six months to reach their full effect, and adverse effects — ranging from sleep disruption to sexual dysfunction to emotional blunting — can persist throughout. For people already struggling, that burden is often enough to abandon treatment.
An estimated 11% of the U.S. population holds a prescription for antidepressants. Translating even a modest improvement in prescribing accuracy across that population would mean millions fewer people enduring treatments unlikely to help them.
What comes next
The current model is specific to sertraline. That’s a meaningful limitation — it doesn’t yet help clinicians choose between the full range of available medications. The researchers are explicit about this: their next goal is to extend the same modeling approach to a wider range of antidepressants and to make the algorithm more personalized over time.
The study also relied on data from a prior U.S. trial rather than a new prospective trial, which means the model still needs validation in prospective clinical settings before it could be used routinely in practice. Those trials take time and resources.
Still, the underlying principle — that biological markers visible before and shortly after treatment begins can predict long-term outcomes — opens a real pathway toward precision psychiatry. If the model extends to other drug classes, the implications for how depression is treated globally would be substantial. This kind of work echoes what’s happening elsewhere in personalized medicine: researchers are finding that what looks like guesswork from the outside is actually detectable signal, if you know where to look. The National Institute of Mental Health has long called for more biomarker-based approaches to psychiatric diagnosis, and studies like this one show what that can look like in practice.
For the roughly 280 million people the World Health Organization estimates live with depression worldwide, faster and more accurate prescribing isn’t a minor convenience. It could be the difference between getting better and giving up.
Read more
For more on this story, see: New Atlas — Novel AI model rapidly determines if an antidepressant will work
For more from Good News for Humankind, see:
- Personalized mRNA cancer vaccine shows six-year results for pancreatic cancer
- Australian researchers find a SIDS biomarker in a newborn blood test
- The Good News for Humankind archive on global health
About this article
- 🤖 This article is AI-generated, based on a framework created by Peter Schulte.
- 🌍 It aims to be inspirational but clear-eyed, accurate, and evidence-based, and grounded in care for the Earth, peace and belonging for all, and human evolution.
- 💬 Leave your notes and suggestions in the comments below — I will do my best to review and implement where appropriate.
- ✉️ One verified piece of good news, one insight from Antihero Project, every weekday morning. Subscribe free.
