AACE 2026: Using AI in clinical practice, best practices

Physicians should consider integrating available AI tools into daily clinical practice, David Toro-Tobon, MD, said at the 2026 AACE Annual Meeting in Las Vegas.

Dr. Toro-Tobon, an investigator in the Care and AI Laboratory within the Division of Endocrinology at Mayo Clinic in Rochester, MN, outlined both the opportunities and limitations of these technologies, offering practical strategies for their safe and effective use.

Large language models (LLMs) are mathematical systems that generate text by predicting word sequences based on patterns learned from large datasets. For endocrinologists, a key consideration is the nature and quality of those underlying data. "This is not medical data. This is not endocrine data. This is world data," said Dr. Toro-Tobon. 

He estimated that about 80% of most LLM training data comes from general Internet sources, while only about 2% represents high-quality curated data—and medical information is only a fraction of that subset. An additional 4% consists of synthetic data generated to train models.  

"You have to keep this in mind when understanding the predictions that these models are doing and recognizing the risks when interacting with them," said Dr. Toro-Tobon.  

He presented data from The New England Journal of Medicine showing that AI scribes can reduce documentation time by approximately 10% in some systems. However, their effectiveness depends heavily on how clinicians use them.

He suggested the following strategies to improve accuracy. 

• Speak naturally, but explicitly. Clearly articulate key findings, diagnoses, and plans.
  

• Close each problem verbally. "For X, the plan is …"
  

• Signal transitions. "Next, let’s discuss ..."
  

• Be explicit in reasoning. This is especially when ruling conditions in or out.

He also emphasized dictating a summary after each visit to "anchor" the note and better reflect clinical reasoning. 

"To use AI effectively, the quality of the output often depends on how well the prompt is structured. Context is everything," said Dr. Toro-Tobon. He shared two complementary frameworks—P.U.L.S.E. and V.E.T.—to help in this area. 

P.U.L.S.E. 

The P.U.L.S.E. framework—Persona, Universe, Logic, Structure, Expectation—can help guide clinicians in crafting precise, clinically useful prompts. 

P — Persona. Define the role of the AI. Example: "Act as an endocrinologist ..." This ensures the response uses appropriate clinical language and reasoning. 

U — Universe. Provide the clinical context. Example: "This is a 68-year-old patient with type 2 diabetes, A1C 11%, BMI 36, and established cardiovascular disease." The context transforms generic responses into patient-specific recommendations. 

L — Logic. Specify the reasoning task. Example: "Explain the medical necessity of starting a GLP-1 receptor agonist based on cardiovascular and glycemic benefits." This directs the AI to justify decisions rather than simply describe options. 

S — Structure. Define the output format. Example: "Write this as a formal prior authorization appeal letter." Structured outputs can be used directly in clinical workflows. 

E — Expectation. Clarify the goal. Example: "The goal is to overturn an insurance denial." This ensures that the response is purpose-driven and actionable. 

V.E.T.

The V.E.T. framework—Verify, Evaluate, and Transparently disclose—can help to mitigate risk.  

V — Verify. Confirm accuracy and sources. "Never act on a claim from an AI tool without verification," said Dr. Toro-Tobon. 

E — Evaluate. Apply clinical judgment. An AI-recommended therapy may be guideline-based, but clinicians must assess whether it fits the patient’s individual needs.  

T — Transparently disclose. Use AI responsibly and protect patient data. De-identify patient information when using AI tools and disclose AI assistance when required in academic or clinical documentation. 

Dr. Toro-Tobon reported no conflicts of interest pertaining to his presentation.

AACE Endocrine AI is published by Conexiant under a license arrangement with the American Association of Clinical Endocrinology, Inc. (AACE^®). The ideas and opinions expressed in AACE Endocrine AI do not necessarily reflect those of Conexiant or AACE. For more information, see Policies.