Antidepressants are some of the most commonly prescribed medicines in the world.
For many people, they provide vital relief from depression and anxiety, helping them manage their daily lives.
But while these drugs can be life-changing, staying on them for too long can bring problems of its own. Side effects such as weight gain, sexual issues, and even heart problems are common with prolonged use.
At the same time, suddenly stopping antidepressants can cause difficult withdrawal symptoms, which means both patients and doctors face a complicated balancing act when it comes to deciding whether to stop treatment.
Now, researchers from the University of South Australia (UniSA) have developed new artificial intelligence (AI) models that could help doctors predict who can safely stop taking antidepressants.
Their study analyzed real-world prescription data from the Australian Pharmaceutical Benefits Scheme (PBS), following 100,000 patients over a period of 10 years. Using machine learning, the team was able to spot clear patterns that separated successful withdrawal attempts from unsuccessful ones.
This research was recently presented at MedInfo 2025, a major international conference on digital health, and published in Studies in Health Technology and Informatics.
The lead researcher, Dr. Lasantha Ranwala, who is both a practicing doctor and an AI scientist, explained the challenge: many health care providers are reluctant to reduce or stop antidepressants because they fear patients will experience withdrawal symptoms. This uncertainty often leaves patients on the medication for much longer than necessary.
By applying AI tools to the PBS database, the team was able to create a way of forecasting which patients are most likely to stop antidepressants without relapse or withdrawal problems.
For the study, a “successful withdrawal” meant that the person had stopped antidepressants completely for at least one year after previously taking them long-term (defined as more than 12 months). If a patient tried to reduce their dose but ended up needing a stronger prescription within six months, the attempt was considered unsuccessful.
The researchers trained two different machine learning models. One focused on final prescription records and achieved 81% accuracy in predicting outcomes. The other model took a broader approach, monitoring patient records from the very first prescription through every dose change and withdrawal attempt.
This second model was even more effective, with 90% accuracy. According to UniSA’s Associate Professor Andre Andrade, this more detailed method gave a fuller picture of patient experiences and provided more reliable predictions.
Antidepressant use has been rising worldwide, with countries like Australia, Iceland, Portugal, Canada, and the UK showing some of the highest consumption rates. Because so many people rely on these medications, finding safe ways to eventually stop them when they are no longer needed is becoming a pressing issue.
The new AI tools could become valuable support systems for general practitioners, giving them more confidence to recommend deprescription when it is safe to do so.
The findings also highlight a broader idea: that administrative health care data, such as prescription records, holds untapped potential for improving clinical decision making.
This information is already routinely collected but rarely used in research or medical practice. By applying AI to such large-scale data, doctors may soon have access to tools that guide them in making safer, more effective decisions for their patients.
In reviewing this study, the results are promising but still in the early stages. The accuracy of 81–90% is encouraging, yet real-world testing in clinics will be essential before these tools can be trusted in everyday medical practice.
Another key question is whether the same approach could be used for other medications, especially those with complicated withdrawal processes. If so, AI might help patients across many areas of medicine reduce unnecessary long-term drug use while avoiding harm.
Overall, the UniSA study shows how advanced technology could soon help solve a problem that doctors and patients have struggled with for years: knowing when it is safe to stop antidepressants. If refined and put into practice, this kind of AI tool has the potential to make mental health care safer, more personalized, and more effective.
The study is published in Studies in Health Technology and Informatics.
Copyright © 2025 Knowridge Science Report. All rights reserved.
