11/08/2026
📱 【AI 入社區診所?港大團隊最新研究話你知】
🏥 香港大學李嘉誠醫學院家庭醫學系周建棟教授,聯同美國麻省總醫院和MIT,喺《通訊醫學》期刊發表咗一項研究,探討大型語言模型(例如 ChatGPT, Claude 呢類 AI)喺基層醫療嘅應用同安全問題。
🔍 而家做到啲乜?—— 最穩陣嘅用途係「幫手打雜」
✅ 幫醫生寫病人回覆、整理病歷摘要、改善醫患溝通
✅ 慳時間、醫生同病人都覺得體驗唔錯
✅ 但目前主要改善嘅係「流程效率」,仲未有明確證據顯示可以直接提升治療效果或診斷準確度
⚠️ 短板與風險 —— 唔好當佢係「全科神醫」
❌ 輔助診斷、治療決策、改善病人結局嘅證據仲好有限
❌ 多數研究仲喺實驗室階段,真係放落社區診所用,仲有好多不確定因素
❌ 人口健康、成本效益呢啲大命題,而家只係「構想階段」,等臨床試驗驗證
🔒 安全第一 —— 四道「保險掣」唔少得
🛡️ 私隱保護、數據安全、演算法透明、模型可靠
🛡️ 目前建議:先用喺「量大、風險低」嘅文書同溝通工作
🛡️ 最終決策權同核實責任必須由醫生承擔,AI 只係助手
💡 結論一句講晒
AI 喺基層醫療「有前景,但唔使急」—— 先用好佢做文書雜務,等治理同公平機制建立好,先再慢慢拓展。醫生把關,病人安心,先至係正路。
📌 研究由港大、麻省總院、MIT 合作完成,即将正式發表於《Communications Medicine》。
#人工智能 #基層醫療 #大型語言模型
===
🤖 AI in Primary Care? HKU Team Breaks Down the Evidence
A new study led by Prof. Edmond Chow from HKU's Department of Family Medicine and Primary Care, in collaboration with Massachusetts General Hospital and MIT, was just published in Communications Medicine. It looks at how large language models (like ChatGPT, Claude) can be used safely in primary care settings.
🔍 What works right now? – Best use is "assisting with the paperwork"
✅ Drafting patient replies, summarising clinical notes, improving doctor-patient communication
✅ Saves time, both doctors and patients report good experiences
✅ But improvements are mainly in workflow efficiency – no solid evidence yet that they directly boost diagnostic accuracy or treatment outcomes
⚠️ Limitations & risks – Don't treat it as a "virtual GP" just yet
❌ Evidence for supporting diagnosis, treatment decisions, or patient outcomes is still very limited
❌ Most studies are still in simulated settings – real-world clinics bring many unknowns
❌ Population health and cost-effectiveness? Still at the "hypothesis" stage – waiting for prospective trials
🔒 Safety first – Four guardrails are essential
🛡️ Privacy, data security, algorithm transparency, and model reliability
🛡️ Current advice: start with high‑volume, low‑risk tasks like documentation and messaging
🛡️ Final decisions and accountability must stay with doctors – AI is only an assistant
💡 Bottom line
AI in primary care has promise, but no rush. Use it for administrative grunt work first, build solid governance and fairness mechanisms, then expand slowly. Doctor in charge, patient at ease – that's the right way forward.
📌 Study by HKU, MGH, and MIT, published in Communications Medicine.
#人工智能 #基層醫療 #大型語言模型
大型語言模型在基層醫療的應用與安全