One of these tools, called large language models (LLMs), is trained on vast amounts of data to create connections or associations among trillions of data elements. This training allows them to process and analyze data very quickly and provide helpful, informative, human-like responses to a wide range of questions. However, their responses aren’t true intelligence, meaning these tools don’t use unique knowledge or skills to determine their answers. Instead, generative AI uses this complex map of data associations to predict answers. It’s important to understand this fact as we look for ways to responsibly use generative AI in health care.

Publicly accessible tools like ChatGPT and Bard can pose privacy and security risks, because they use the information you enter and your interactions to further train and refine these models and their output. This means that what you enter into these tools could be exposed to other users, so you never want to risk entering any sensitive information, like Personally Identifiable Information (PII) or patients’ Protected Health Information (PHI). These privacy and security risks limit its use in health care, where protected and sensitive data is often required to gain insights from AI tools.

Generative AI has the ability to analyze large amounts of patient data, reducing the need for clinicians to undertake time-consuming research and analysis. AI-driven platforms and chatbots are already helping us to connect with patients and improve efficiency. Now, we’re looking at ways to use these tools to analyze large clinical datasets to make our clinicians’ lives easier.

“Clinicians want to give their patients a great, personalized experience. As a practicing hospitalist, I do inpatient care, and it can be challenging to review a chart with loads of patient notes and data to find all the key, critical information in a very short period of time,” said Dr. Jordan Dale, HM chief medical information officer. “AI-driven technologies would help unlock massive amounts of patient data and drill down to a more complete picture of each patient’s history, so we can really get to know our patients, have more meaningful interactions and give them better care.”

“We’ve actually applied some generative AI technology that will help generate a patient summary, so I can ‘get to know’ a patient’s key, past medical history, current problems, medications they’re receiving and who’s following the patient more quickly,” Dr. Dale continued. “Before, some of those specific data elements were really difficult to find.”

Generative AI can also be used to analyze clinical workflows and identify bottlenecks and inefficiencies to help save valuable time and facilitate more efficient patient care.

Amy Premetz, vice president of revenue cycle operations stated, “We’re currently working with generative AI to assist in the creation of insurance authorization denial appeal letters. This will make the process more efficient and help expedite care for our patients.”

“For this endeavor, I’m partnering with one of our radiation oncology providers. The radiation oncologist is using generative AI tools to pull in various clinical elements, including recent studies to support medical necessity, into letters to appeal denials, that will be sent back to the insurer,” said Premetz. “Once the radiation oncologist has created the draft letter, reviewed and approved it, we’ll send it to our Ambulatory IT partners who will load it into an Epic template. Once it’s in Epic, the provider opens up the template, reviews it to make sure no changes are needed and signs the letter. This process will expedite appeals and potentially allow the patient to start cancer treatment sooner,” said Premetz.