January 5, 2024

Artificial intelligence advances work in VUMC’s clinical and research settings

Vanderbilt University Medical Center physicians and researchers are applying artificial intelligence in innovative ways to advance clinical care and scientific understanding of these cutting-edge tools.

Vanderbilt University Medical Center (VUMC) physicians and researchers are applying artificial intelligence (AI) in innovative ways to advance clinical care and scientific understanding of these cutting-edge tools. Some examples of the use of AI are included here but other new examples are being developed and tested.

VUMC recently began testing AI for assisting health care documentation with a 10-physician pilot involving a system called DAX Copilot, from Nuance, a Microsoft subsidiary.

DAX Copilot is an AI-powered, voice-enabled system designed to automate the creation of clinical notes. The program listens to patient encounters and leverages generative AI to generate real-time, comprehensive clinic visit notes with appropriate headings and context. These notes are then made available immediately for clinician review and editing, integrating directly into VUMC’s electronic health record system.

Dara Mize, MD, MS

A key aspect of AI-assisted documentation is its potential to alleviate physician burnout while improving patient interactions, said the leader spearheading the pilot, Dara Mize, MD, MS, assistant professor of Biomedical Informatics and Medicine and Chief Medical Information Officer for VUMC.

“Generative AI shows promise in improving both the quality and efficiency of health care documentation. The upcoming pilot marks a significant step in VUMC’s exploration of AI’s potential in streamlining clinician workflows and enhancing medical record-keeping while reducing time spent on documentation,” Mize said.

The initial pilot will involve physicians from the Division of General Internal Medicine and Public Health and the Department of Orthopaedic Surgery. The Medical Center may evaluate additional AI-powered documentation assistants, Mize said, and these tests could expand to involve faculty in other divisions and departments.

“This continues VUMC’s commitment to identifying and integrating leading health care technologies for quality, safety, and an enhanced patient/provider experience,” Mize said.

For research, VUMC has created a version of OpenAI’s large language model which is deployed inside of VUMC’s Azure tenant and is available to VUMC’s researchers. The VUMC-hosted graphical user interface (GUI) supports common prompt and response features.

Researchers can experiment using this interface as an alternative to the public version of ChatGPT which is disabled on VUMC’s networks. The Chat playground provides more controls than the public chat interface from OpenAI.

Unlike the public version of ChatGTP, VUMC retains control over all data submitted to aiChat. aiChat is HIPAA certified and its use is covered under VUMC’s Business Associates Agreement with Microsoft.

“Health IT worked closely with leaders in the Department of Biomedical Informatics to create a platform for the VUMC community to safely explore large language models,” said Travis Osterman, DO, MS, Associate Vice President for Research Informatics and associate chief medical information officer for VUMC.

More information about generative AI can be found at the Office of Research On the Horizon Town Hall Series (https://www.vumc.org/oor/horizon-series). Researchers can test aiChat at https://aichat.app.vumc.org/ (on campus or vpn required).

The Vanderbilt-Ingram Cancer Center (VICC) was an early adopter of using AI to help schedule treatment for patients with cancer. In 2019 VICC implemented the LeanTaaS iQueue system to reduce patient wait times in infusion centers, assist with nurse scheduling, and improve throughput. The iQueue platform collects actual infusion times for various treatments and leverages AI to recommend the optimal infusion appointment for patients.

“Use of the LeanTaaS iQueue for Infusion product has helped TVC 2 Infusion achieve a 50% reduction in median patient wait time, while increasing our average patient hours by 10%. Our scheduled is now level-loaded throughout the day, leading to significantly fewer bottlenecks, increased patient satisfaction, and increased nurse satisfaction,” said Cody Stansel, MSN, RN, administrative director of Nursing for VICC.

“Patients typically feel that their wait times are more reasonable and that they can get in and get their treatment faster. Our nursing staff are enabled to take their breaks on time and no longer feel the pressure caused by a consistent lobby full of patients waiting to be seen.”

As another example of AI adoption at VUMC, RapidAI is also currently being used to assist the VUMC Stroke Team in evaluating patients with stroke. These tools deliver quantified and color-coded CT perfusion maps that help physicians quickly assess salvageable brain tissue — enabling faster clinical decisions that facilitate better patient outcomes (https://www.rapidai.com/rapid-ctp).

In addition to deploying commercial AI applications, VUMC also has a history of developing its own. For instance, in 2014 Vikram Tiwari, PhD, MBA, associate professor of Anesthesiology and Biomedical Informatics, William R. Furman, and Warren Sanberg, MD, PhD, chair of the Department of Anesthesiology and professor of Biomedical Informatics, developed an AI tool to predict elective surgical case volume that is still used today. Their work is published in the journal Anesthesia.

Developing elective schedules allows anesthesiologists to predict final case volumes weeks in advance. After implementation, overly high or low-volume days are revealed in advance, allowing nursing, ancillary service, and anesthesia managers to proactively fine-tune staffing to match demand.