PUPC Conference Coverage

Advancing Primary Care Through AI: Principles, Use Cases, and Future Directions

Anthony Calabro, MA

Key Highlights

  • The presenters detailed three high-impact use cases for AI in primary care: patient messaging, test result commentary, and clinical note drafting.
  • They also offered a novel framework for selecting AI models grounded in fairness, usefulness, and reliability.

In their Forward-Thinking Series on health care innovation at the Practical Updates in Primary Care Virtual Meeting, Anthony Pho, PhD, MPH, ANP-C, senior clinical education lead at Stanford Health Care and Timothy Tsai, DO, MMCi, clinical assistant professor in the division of primary care and population health at Stanford University School of Medicine, provided a foundational overview of how artificial intelligence (AI) continues to shape the future of primary care.

“The idea behind this talk is to give you a sense of what AI is and how AI applies to the primary care environment,” Dr Pho explained during his presentation.

The session included four main learning objectives:

  • introducing foundational AI principles relevant to health care;
  • exploring three real-world clinical use cases in primary care;
  • outlining a framework for evaluating AI model selection; and
  • discussing future directions for responsible AI integration.

Dr Tsai began his portion of the presentation with accessible definitions and analogies for key AI concepts. Machine learning (ML) is likened to a medical resident who improves through experience, while deep learning (DL) is compared with a seasoned specialist capable of identifying subtle patterns. Natural language processing (NLP), large language models (LLMs), and generative AI are each contextualized with clinical analogies and examples, such as AI-generated patient education materials or dictation tools that convert speech to structured EHR notes.

Dr Tsai also reviewed three core clinical use cases using the 80/20 Pareto principle, where a minority of inputs produces the majority of results. The three cases involved AI-generated patient message responses, interpretation and communication of test results, and automated drafting of clinical notes. Dr Tsai framed each use case as an opportunity to remove administrative burden while maintaining clinical reliability. For example, in messaging, AI can draft responses that blend patient-specific data with evidence-based content, while result commentary tools assist in generating accurate, timely follow-ups. In the note-writing use case, NLP/LLM-powered systems can convert voice-recorded patient encounters into draft SOAP (Subjective, Objective, Assessment, Plan) notes, easing documentation load.

To ensure that AI enhances rather than complicates care, the presenters introduced the Stanford FURM model—a framework for assessing fairness, usefulness, and reliability in AI models in health care systems. Dr Pho framed FURM as a guide to the ethical and effective implementation of AI systems.

“I see AI technology as doing something to enhance what human beings are doing, to make us more efficient, to make us better, and to deliver higher quality care to our patients,” Dr Pho said. “Having said that, we can't just implement anything without a sense of ethics, and quality … FURM builds that into a system.”

Dr Pho encouraged anyone interested in implementing an AI system to serve or form AI committees at their respective institutions.

“Your voice as a clinician and your training in ethics and building it into frameworks like FURM are key to the success of AI and making sure that we actually maintain autonomy, agency, and all these wonderful concepts that we've learned for so many years in terms of ethical behaviors in the way we deliver our care,” Dr Pho concluded. “AI is not exempted from any of that.”

Reference
Tsai T, Pho A. Forward-Thinking Series: Integration of AI in Primary Care. Practical Updates in Primary Care. May 7, 2025. https://www.hmpglobalevents.com/pupc