 |
Scottish National Users' Group (SNUG) PodcastAuthor: snughealth
Supporting Scotland's Primary Care GP software users: welcome to the Scottish National Users Group (SNUG). Language: en Genres: Health & Fitness, Medicine, Technology Contact email: Get it Feed URL: Get it iTunes ID: Get it |
Listen Now...
Can AI help reduce polypharmacy? (Part 2)
Episode 71
Wednesday, 18 February, 2026
In this second part of our polypharmacy discussion with Steve Williams, we continue to explore whether large language models like Microsoft Copilot could play a practical role in supporting medication reviews. We consider a real case from a duty day in general practice where Copilot was used to assess prescribing safety and generate a summary of deprescribing opportunities. This prompt was used: “Review the following medication list using the latest British Geriatrics Society (BGS) guidance and the Scottish Polypharmacy Guidance (7-step approach). For each medicine, identify: Indication and whether it is still appropriate Clinical risks (frailty, falls, anticholinergic burden, renal function, interactions) Deprescribing opportunities Safer alternatives if applicable Monitoring requirements Then provide a concise summary of priority actions and any safety red flags that need urgent review. Here is the medication list: [PASTE MEDICATIONS HERE] Include references to the relevant guideline steps where appropriate.” The conversation also covers a new study led by Professor Tony Avery, which tested an LLM against an expert clinician to assess medication safety in nearly 300 anonymised GP patient records. While the model achieved 100% sensitivity in detecting clinical issues, it matched the expert's full assessment in under half of cases, with failures arising from overconfidence, lack of contextual reasoning, and occasional hallucinations such as misidentifying medications. Steve is self-described as a “curious pragmatist” and feels that LLMs are of great interest, and their ability to flag problems with high sensitivity - when guided by good prompts and established clinical frameworks - makes them a genuinely useful preparation tool, provided the clinician still does the thinking. As Steve puts it, the technology looks promising, but "human intelligence is very underrated..." A Real-World Evaluation of LLM Medication Safety Reviews in NHS Primary Care. Evidence Based Polypharmacy Reviews and the 7 Step Process: TURAS training Thinking Critically About AI (Video lecture by Dr Jessica Morley) How we make decisions – dual process theory and unconscious biases (MeReC Bulletin 2011) You can subscribe to the SNUG podcast on the following platforms: SNUG podcast on Apple podcasts SNUG podcast on Spotify
