The integration of artificial intelligence into clinical practice has transitioned from a speculative future to a tangible necessity, a shift that took center stage during the ENDO 2026 annual meeting. In a high-capacity session held on Saturday, June 13, researchers and clinicians gathered to dissect the evolving role of AI in endocrinology. The session, titled “Artificial Intelligence in Endocrinology: Practical Uses, Lessons Learned, and What Comes Next,” provided a comprehensive roadmap for how the specialty is navigating the complexities of digital transformation. Led by Juan P. Brito, MD, MSc, a professor of medicine and director of the Care and AI Laboratory at Mayo Clinic, the presentation underscored a critical message for the medical community: AI is no longer a novelty to be observed from the sidelines but a tool that requires active engagement to avoid professional obsolescence.
The session served as a bridge between two distinct but complementary spheres of the AI revolution. While one perspective focused on the immediate, practical tools available to clinicians to enhance daily productivity, the other examined the systemic challenges of implementing these technologies within the rigorous demands of a real-world healthcare environment. Dr. Brito, who also serves as the medical director of the Shared Decision Making National Resource Center at Mayo Clinic, emphasized that the goal was to evaluate how AI can improve clinical decision-making while remaining anchored in evidence-based medicine and patient-centered care.
The Rise of Everyday AI in Clinical Practice
David Toro-Tobón, MD, a Mayo Clinic Scholar and assistant professor of medicine, opened the session by introducing the concept of "Everyday AI." This term describes the suite of generative AI tools and chatbots that clinicians can leverage across various stages of their workflow. These include AI-powered scribes for documentation, clinical decision support tools like OpenEvidence for real-time data retrieval, and administrative assistants designed to streamline research and academic writing. Dr. Toro-Tobón argued that the primary value of these tools lies in their ability to alleviate the "cognitive load" that currently burdens endocrinologists, many of whom face significant administrative hurdles that detract from direct patient interaction.
According to data from recent physician burnout surveys, administrative tasks and electronic health record (EHR) documentation remain leading causes of professional dissatisfaction. By offloading these repetitive, low-complexity tasks to AI, Dr. Toro-Tobón suggested that clinicians could reclaim time to focus on the nuanced aspects of patient care—such as understanding a patient’s personal values and managing complex psychosocial factors—that machines cannot replicate. He framed the relationship as a partnership rather than a competition, noting that human and artificial intelligence possess complementary strengths and pitfalls.
However, the adoption of "Everyday AI" is not without significant risks. Dr. Toro-Tobón warned that an insufficient understanding of AI limitations, such as "hallucinations"—where a model generates plausible but factually incorrect information—could lead to increased workloads or, more severely, patient harm. The stakes in healthcare are uniquely high, requiring a much lower tolerance for error than in general business applications. To mitigate these risks, he advocated for a rigorous approach to AI literacy among clinicians, ensuring they can identify when a tool is providing unreliable outputs.
The Regulatory and Privacy Labyrinth
One of the most critical discussions during the session centered on the distinction between business-compliant and HIPAA-compliant AI tools. Dr. Toro-Tobón highlighted a pervasive misconception in the medical community regarding data security. Many health systems have adopted AI tools that feature "green shields" or language suggesting data protection. However, he clarified that "business-compliant" often only means the data is not used to train the underlying models; it does not necessarily mean the tool meets the strict regulatory requirements of the Health Insurance Portability and Accountability Act (HIPAA) for handling protected health information (PHI).
The complexity of this landscape is compounded by the fact that some tools, such as Doximity Ask or OpenEvidence, are HIPAA-compliant by design but require personal National Provider Identifier (NPI) logins. This creates a friction point because the data technically belongs to the institution, not the individual physician, leading some healthcare systems to restrict their use. Dr. Toro-Tobón’s own pursuit of a master’s degree in AI at Mayo Clinic was motivated by this widening gap between data science and clinical expertise. He noted a surge in "AI spoofing" in academic literature—studies that report exceptional performance but contain methodological flaws that render them impractical for immediate clinical change.
Implementation Science and the Three Pillars of Adoption
Naykky M. Singh Ospina, MD, a professor in the Division of Endocrinology, Diabetes, and Metabolism at the University of Florida, shifted the focus toward implementation science. Her research examines the social and systemic factors that determine whether an AI tool successfully moves from a laboratory setting to a clinic. Dr. Singh Ospina organized the barriers to AI adoption into three essential pillars: technical, clinical, and implementation.
The technical pillar requires AI models to mature enough to handle the longitudinal and context-dependent questions common in endocrinology. The clinical pillar demands proof that the tool actually improves patient outcomes, rather than just providing a high-accuracy prediction. Finally, the implementation pillar involves the social challenge of building trust among patients and clinicians and integrating the tool into existing institutional workflows. Dr. Singh Ospina noted that progress is often circular; without evidence of clinical benefit, health systems are hesitant to invest in implementation, and without implementation, it is difficult to gather large-scale evidence of clinical benefit.
Case Studies: Retinopathy vs. Thyroid Ultrasound
To illustrate these pillars in action, Dr. Singh Ospina compared two high-profile applications of AI in endocrinology. The first, a success story, is the FDA-approved AI tool for diabetic retinopathy screening. This technology has successfully navigated all three pillars. Technically, it provides a binary "yes/no" assessment for high-risk findings. Clinically, it has been shown to improve patient adherence to follow-up care by providing immediate results during a routine clinic visit, thereby bypassing the need for separate ophthalmology appointments. From an implementation standpoint, it is supported by clinical guidelines and specific billing codes, making it a viable part of the healthcare ecosystem.
In contrast, AI tools for thyroid ultrasound assessment face a steeper climb. While several models can flag nodules suspicious for cancer, they have yet to demonstrate a clear improvement in overall patient care or a reduction in unnecessary biopsies. The underlying clinical problem is more complex than retinopathy screening, involving significant uncertainty and the need to contextualize findings within a patient’s broader medical history. Furthermore, concerns about model accuracy for rare thyroid cancer subtypes have hindered widespread clinician trust. Dr. Singh Ospina argued that for AI to trigger a true revolution in this area, it must move beyond narrow-scope questions and assist with these more complex, multi-variable decisions.
A Chronology of AI Integration in Endocrinology
The discussions at ENDO 2026 reflect a broader timeline of technological evolution within the specialty.
- 2018–2021: The "Exploration Phase," characterized by the first FDA approvals for AI-based diagnostic tools and high levels of speculative hype regarding the replacement of human doctors.
- 2022–2024: The "Technical Refinement Phase," where generative AI and large language models (LLMs) entered the mainstream, leading to early pilot programs for AI scribes and administrative assistants.
- 2025–2026: The "Implementation and Evaluation Phase," marked by a focus on rigorous standards, HIPAA compliance, and the integration of AI into multidisciplinary workflows, as evidenced by the ENDO 2026 session.
Broader Implications and the Future Landscape
The consensus among the speakers was that AI will not replace clinicians, but rather that clinicians who utilize AI will replace those who do not. This shift has profound implications for medical education and professional development. The Mayo Clinic’s master’s program in AI for healthcare represents a growing trend of "clinician-scientists" who possess technical fluency in both data science and medicine. These individuals will be essential for "co-designing" future tools, ensuring they address actual points of friction in patient care rather than being built in a vacuum by developers without clinical experience.
From a systemic perspective, the successful integration of AI could significantly impact rural and underserved areas. Tools like the retinopathy screening AI can bring specialized diagnostic capabilities to primary care settings, reducing the burden of travel for patients. However, the speakers cautioned that if the "implementation gap" is not addressed, AI could inadvertently widen healthcare disparities if only well-funded, urban academic centers have the resources to adopt and maintain these technologies.
The session concluded with a call to action for all stakeholders—clinicians, AI developers, and implementation scientists—to collaborate from the earliest stages of tool development. As Dr. Brito summarized, the field is moving from a period of exploration toward one of thoughtful implementation. While the "unglamorous" parts of the job, such as administrative and routine cognitive tasks, are seeing immediate benefits, the tools that will directly influence diagnosis and treatment decisions still require rigorous evaluation. The promise of AI in endocrinology remains vast, but its safe and effective realization depends on maintaining a steadfast focus on the patient at the center of the technological surge.
