September 24, 2025
A new machine learning model developed by researchers at the University of South Australia (UNISA) will help clinicians identify patients' successful halting their long-term antidepressant use.
Using artificial intelligence (AI) to track distribution data from the Pharmaceutical Benefit Scheme (PBS), UNISA researchers identified the most successful suppressive case of 100,000 patients prescribed over the decade.
Antidepressants are rising rapidly around the world, leading Australia, Iceland, Portugal, Canada and the UK topped the world's top consumer levels.
While these drugs can be life-changing, long-term use is associated with a variety of side effects, including weight gain, sexual dysfunction, and heart problems.
However, managing the therapeutic benefits and risks is a delicate and balanced act, as 50% of patients experience withdrawal effects when patients stop taking the medication.
“People are often reluctant to stop prescribing antidepressants due to concerns about the effectiveness of withdrawal, making it difficult for doctors to know who can safely stop treatment,” says Dr. Ranwara.
“By applying AI to the PBS database, we identified patterns associated with predicting that patients are most likely to succeed when removing antidepressants.”
Successful inhibition was defined as the lack of antidepressants for at least 1 year (>12 months) after previous long-term use (>12 months). If the intensity of the medicine increased within 6 months after the reduction trial, it was labeled as a disability.
Researchers say the findings could provide clinicians with a powerful decision support tool and help them begin depletion with more confidence.
Two machine learning algorithms were trained and tested. One assessed the final prescription record (achieved 81% accuracy), and the other monitored the patients who were followed from the initial prescription, dose reductions and results (achieved 90% accuracy).
“These results show true promises,” says Andre Andrede, co-author and associate professor at UNISA.
“The most accurate model was one that provided a more subtle picture that better reflects the patient's experience and better reflects the patient's experience,” says Andrade Association.
The findings suggest that controlled healthcare data may help predict clinical outcomes and improve medical decision-making.
“This data is collected passively and not used by healthcare professionals and good candidates for AI use.”
Based on AI tools to predict safe antidepressant withdrawal, researchers will focus on making the technology more accurate and easier to use. They aim to test effectiveness in clinics and explore how similar approaches can help patients improve their medication use.
The study, “Predicting machine learning to suppress antidepressant drugs using management data,” was published at Medinfo 2025, an international conference on digital health and informatics.
doi:0.3233/shti250959
Notes for editors
*https://www.oecd.org/en/publications/2023/11/health-at-a-glance-2023_e04f8239/full-report/pharmaceutical-consumption_4b6cb013.html)
From 2023 to 24, antidepressants were distributed to 14% of Australia's population. People aged 10-24 had a relative increase in long-term use (an increase of 110%) and a percentage of long-term users (35%). The average duration of treatment episodes increased by 25% for all ages, with the 10-24-year-old group showing the largest increase (56%).
……………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………
Researcher contact information: Dr. Rasantha Ranwara E: ranly015@mymail.unisa.edu.au
Media Contact: Candy Gibson M: +61 434 605 142 E: candy.gibson@unisa.edu.au
