Photo: Pixabay via Pexels
Algorithms used for consumer predictions are being retooled to help identify which medicines are optimal and may carry hidden risks for patients.
Researchers from Monash University used a different approach in uncovering medicine safety patterns. In their study, they utilised association discovery, a data mining technique first applied to market basket analysis, which identifies relationships between items frequently purchased together in transactions.
FINDINGS
The study, published in Clinical Pharmacology and Therapeutics, claims to be the first to apply the data mining technique in longitudinal healthcare data to identify both drug safety and drug repurposing signals.
A 10% data sample from the Pharmaceutical Benefits Scheme (PBS) was utilised, covering more than 300 million prescription records from 2014 to 2024, to identify links between medicines and three chronic conditions: coronary heart disease, type 2 diabetes, and epilepsy.
Some of the patterns found from their study include expected links between cholesterol-lowering and blood-thinning drugs and heart disease. In type 2 diabetes, antipsychotics, diuretics, and statins were tied to a higher risk, while medicines for Parkinson’s and osteoporosis showed a surprising protective effect. For epilepsy, antidepressants and antipsychotics were linked to a higher risk, though one blood pressure drug appeared to reduce it.
According to the research team, these findings showed how the association discovery technique may be used for early-stage, hypothesis-generating screening of healthcare data to identify drug safety and drug repurposing signals.
Further studies, however, are needed to refine and validate the AI signals by testing the timing, biological plausibility, and consistency across various datasets, they said.
WHY IT MATTERS
The study showed how prescription data can be utilised to reveal "hidden connections" between medicines and health outcomes, said research co-lead Dr George Tan, a fellow at Monash Institute of Pharmaceutical Sciences Centre for Medicine Use and Safety.
"Medicine use patterns tell a story. By studying prescriptions, we can not only see how conditions are being treated, but also discover surprising links that may point to new risks, new protections, or even new uses for existing medicines," Dr Tan explained.
Meaningful relationships uncovered across millions of prescriptions may also guide the direction of future clinical studies, added study co-lead Dr Lynn Miller.
Moreover, the study demonstrated how big data can generate "early clues" about medicine safety and effectiveness, noted the study's senior author and Monash IT professor Geoff Webb.
"By harnessing the same algorithms that predict what people might buy next, we can begin to anticipate which medicines may work best for which patients - and which might pose hidden risks."
THE LARGER TREND
Recently, another study in South Australia also used 10-year PBS data to train an AI model to predict safe antidepressant withdrawal. It has demonstrated 81% accuracy in assessing final prescription records and 90% accuracy in monitoring patients' dose reductions and outcomes from their first prescription.
Meanwhile, in South Korea, Daewoong Pharmaceutical unveiled last year its AI-driven drug development system based on a database of 800 million compounds collected over 40 years. The web-based Daewoong AI System, or Daisy, allows researchers to discover new compounds and quickly predict drug properties.
