By applying artificial intelligence (AI) to dispensing records from the Pharmaceutical Benefits Scheme (PBS), researchers at the University of South Australia (UniSA) have pinpointed the most successful instances of deprescribing among 100,000 patients who were prescribed antidepressants over a 10-year period.

With antidepressant use rising sharply worldwide—Australia, Iceland, Portugal, Canada, and the UK rank among the highest consumers—the UniSA breakthrough could assist general practitioners in safely tapering medications that are no longer clinically necessary.

Although antidepressants can be transformative, long-term use is often associated with side effects such as weight gain, sexual dysfunction, and heart complications.

At the same time, around half of patients experience withdrawal symptoms when stopping treatment, making it a delicate balance between therapeutic benefits and potential harms, says Dr. Lasantha Ranwala, a medical practitioner, AI researcher, and UniSA PhD candidate.

“Because of concerns over withdrawal, many healthcare providers are hesitant to discontinue antidepressant prescriptions, which creates challenges in determining which patients can safely stop treatment,” Dr. Ranwala explains. 

“By applying AI to the PBS database, we were able to uncover patterns associated with successful withdrawal, allowing us to predict which patients are most likely to discontinue antidepressants safely.”

In this study, successful deprescription was defined as going without any antidepressant medication for at least 12 months after long-term use (more than a year). Conversely, if the medication dose increased within six months of a reduction attempt, it was considered unsuccessful.

Researchers believe these insights could provide clinicians with a valuable decision-support tool, enabling them to initiate deprescription with greater confidence.

Two machine learning models were developed and tested. One focused on patients’ final prescription records, achieving 81% accuracy. The second model tracked patients from their very first prescription, capturing dose reductions and outcomes, and achieved 90% accuracy.

“These findings are very encouraging,” says UniSA co-author Associate Professor Andre Andrade.

“The more accurate model provided a richer picture of deprescription attempts, which more closely mirrors the real experiences of patients,” he explains.

The results highlight how administrative healthcare data can be harnessed to predict clinical outcomes and enhance medical decision-making.

“This type of data is routinely collected but often underutilized by healthcare professionals, making it an excellent candidate for AI applications,” Assoc Prof Andrade adds.

Building on this work, the team now plans to refine their AI tool to improve accuracy and usability. They also aim to test it in clinical settings and explore how similar AI-driven methods could support patients in optimizing their use of medicines.

Source: https://www.unisa.edu.au/media-centre/Releases/2025/using-ai-to-predict-when-patients-can-safely-stop-using-antidepressants/