Hi!
Today is LLM Friday. A day dedicated to Large Language Models.
Today, I have an important question for you: if I could personally help you solve one problem related to AI, what would that be?
Please, hit reply with your answer. I promise I will read and answer each one of them directly to your email.
Now, grab a drink and let’s dive into today’s issue.
✨LLMs
This study analyzed how 75 healthcare professionals used generative AI to draft replies to patients in their Electronic Health Record (EHR).
Researchers used EHR audit logs from a NYC Langone Health system over 11 months to measure use rates, efficiency gains, and workflow impacts.
🔬 Methods
Participants: 75 HCPs (72% physicians, 19% clinical support, 9% administrative support).
GenAI-drafted responses to patient messages in EHR.
Assessment: Use rates, response time, actions, linguistic analysis of drafts.
📊 Results
Overall use: 19.4%. This improved from 12% to 20% after prompt refinements.
Administrative staff showed highest adoption (32%), followed by clinical support staff (14%), and physicians (12%).
Time savings with AI: 6.76% faster median response time.
7.27% increase in message open time (to review the AI draft added time)
80% of the messages did not receive a response. Reviewing unnecessary drafts increased the burden by 135% compared to reading messages alone.
Successful AI drafts were:
Concise, fewer words, but more informative.
Highly relevant to patient questions.
Easy to read (higher readability scores).
More subjective/empathetic.
🔑 Key Takeaways
Use differs significantly with role. Messages need to be tailored based on professional needs.
Concise and relevant responses are more likely to be used.
GenAI can accelerate message response time.
Drafts need to be generated only for messages requiring responses.
Prompts need to be refined based on use patterns.
Aligning AI tools with existing clinical processes needs to be considered.
🔗Mandal S, Wiesenfeld BM, Szerencsy AC, et al. Utilization of Generative AI-drafted Responses for Managing Patient-Provider Communication. npj Digital Medicine. 2025;8:591. https://doi.org/10.1038/s41746-025-01972-w
🦾TechTools
Last week, we talked about different prompting styles.
Now, when you give prompts to AI this week, try adding these quick details:
Who you want the LLM to be: If you want a physician, PT, or OT’s opinion, be specific.
S: Age, signs and symptoms, clinical history of the patient.
O: Patient’s age, vitals, physical exam findings, test results.
A: A possible diagnosis, differential diagnosis (ask for orientation if you need it).
P: Treatments (ask for official guidelines).
You’ll see the jump from generic advice → clinically relevant information.
Let me know your results.
That’s all for today.
You’re already ahead of the curve in medical LLMs — don’t keep it to yourself. Forward AIMedily to a colleague who’d appreciate the insights.
Thank you!
Until next Wednesday.
Itzel Fer, MD PM&R
Join my Newsletter 👉 AIMedily.com
Forwarded this email? Sign up here
P.S. I’ll be waiting for your response (any language is welcome)🙂.