The Technological Frontier: AI-Driven Healthcare

Megan Rolfzen, MD

A central theme of the 2026 Annual Meeting, presented by IARS and SOCCA, is the integration of advanced computation into clinical workflows. The International Science Symposium, “Algorithms to Clinical Innovations: Harnessing Artificial Intelligence (AI) to Transform Anesthesia and Perioperative Medicine,” held on Friday, May 1, explores the transition of AI from a theoretical curiosity to a clinical reality.

Integral to the technical feasibility of AI model development is the foundational layer of data standardization; a challenge and opportunity tackled by Meredith Adams, MD, MS, FASA, FAMI, the IARS Secretary and Treasurer and professor of anesthesiology at Wake Forest School of Medicine. As a recognized leader in health IT, Dr. Adams has most recently pioneered frameworks for harmonizing complex intensive care unit variables into common data models. Her team has validated a large language model (LLM)-based tool for automated mapping to Observational Medical Outcomes Partnership (OMOP) common data model. Not only does this align with NIH’s data-sharing standards, but it also supports the future of collaborative, multi-institutional biobanks needed for this type of research. As a key architect for the NIH HEAL Initiative data standardization, Dr. Adams is leading the way in enhancing collaboration by making data digestible for foundational model development.

One standout model that highlights the power of advanced computation is “TITAN,” a vision-language pathology foundation model that uses self-supervised learning to analyze pairs of whole-slide images and pathology reports, enabling automated report generation and advanced slide retrieval for challenging cases. Andrew H. Song, PhD, an assistant professor in the Department of Translational Molecular Pathology at MD Anderson Cancer Center, illustrated how training foundation models on broad data can facilitate a variety of outputs. Implicitly stated, integrating Dr. Song’s computational tools into perioperative anesthesia might have the power to minimize the risk of cancer recurrence by tailoring techniques to individual patients’ spatial proteomic profiles.

At the core of the AI healthcare transformation lie foundational models and rigorous validation. Despite the surge in AI research, anesthesiology still lags behind other specialties in translating knowledge and data into practice. Referencing an Anesthesiology 2025 article developed by the Anesthesia Research Council steering committee titled, “Strengthening Discovery and Application of Artificial Intelligence in Anesthesiology”, Hannah Lonsdale, MBChB, associate professor of anesthesiology at Vanderbilt University Medical Center, pointed out that anesthesiology ranks second to last among medical specialties in AI-related publications. Furthermore, implementation of advanced computation in clinical workflows remains an aspirational goal rather than a current reality. Despite the vast potential of these technologies, the practical implementation in the operating room remains a significant hurdle.

To bridge this gap, Nima Aghaeepour, PhD, professor of anesthesiology at Stanford University, focuses on utilizing machine learning to synthesize “multi-omics” data with traditional EHR data to create high-fidelity predictive models. Using a foundational model trained on intraoperative data, Dr. Aghaeepour’s team identified a high-risk cohort: patients whose surgeries were short, but whose “biological signatures” predicted a long anesthetic course. These individuals were at higher risk of postoperative complications. Perhaps most provocative was the mention of an AI model that reproduces canonical drug-response patterns and has outperformed practicing anesthesiologists in forecasting intraoperative mean arterial pressure (MAP) across 70 universities within the Multicenter Perioperative Outcomes Group (MPOG). While Dr. Aghaeepour acknowledged that LLMs currently struggle with quantitative data, his group is working to refine these tools for creating anesthetic plans. The implication for the reader is a future in which the “anesthetic plan” is generated by an algorithm that considers the patient’s unique biological signature, potentially reducing the incidence of adverse events and optimizing recovery.

The session concluded with a vision of the anesthesiologist evolving into a “human-in-the-loop” supervisor, overseeing automated systems that manage routine titration. With the rise of “ambient dictation,” as explored by Rahrisch A, et al. JCA 2026, and high-fidelity “digital twins,” ultra-high fidelity reconstructed systems supporting digital biomarkers of health and disease, the role of the perioperative clinician is being redefined. However, the panel warned that this technological leap requires accountability for equity, privacy, bias adjustment, and even environmental impact. If we are at the peak of inflated expectations for these technological advances, only time will tell what and who make it to the plateau of productivity in the AI hype cycle.