Stanford Health Care deploys AI agents to access personalized RWE

How Stanford uses AI agents

ChatEHR allows Stanford physicians to ask questions about a patient's medical history. The AI tool translates EHR data into plain language. Physicians can ask for a summary or specific data points. For example, a physician might ask ChatEHR if a patient is allergic to a medication.

"We're fine-tuning within ChatEHR the right set of questions that should be extracted from the note," said Anurang Revri, vice president and chief enterprise architect at Stanford Health Care. "So the actual result, which is the evidence-based studies, is a set of questions. We are anticipating those questions and sending those without the patient data leaving our premises."

Physicians are looking for RWE to enhance patient notes to help them more effectively communicate with the patient during clinical visits. Previously, health systems would create notes based on historical data. Now, AI agents can generate personalized RWE and embed it in an EHR. Physicians can then use a large language model (LLM)-based application, like ChatEHR, to chat with the medical record.

The Atropos AI agent works in the background to gather information about the patient and the visit and determines key treatment decisions, according to Hyde. 

"While the doctor swivels their chair over to the EHR, that evidence and summarization is inserted right in the EHR next to the note," Hyde said.

When physicians review notes, they can access clear evidence to guide decisions, he added.

"It saves them time, they get a clear answer on what to do, and don't have to stop and use other tools or change their workflow," Hyde explained. 

For instance, physicians may use the Atropos Evidence Agent if they're looking for studies on side effects of weight loss medicines, such as Ozempic, Revri explained further. They may also collect RWE on treatments for COVID-19 symptoms, such as loss of smell. Physicians then share these findings with patients.

An agent could also collect a list of appointments from a physician's schedule, provide a summary of each patient and indicate if patients are eligible for a clinical trial. You can call up records of a single patient or in a batch, according to Revri.

"Instead of a provider's assistant, or a nurse, or somebody else doing that administratively, we now have agents doing that work," Revri said.

Atropos has a platform called the Generative Evidence Acceleration Operating System, or GENEVA OS, that houses its evidence-generation tools in the cloud environments of health systems such as Stanford. GENEVA OS rapidly incorporates healthcare evidence from a network of real-world data.

"Running a study in seconds is not easy, and GENEVA is what enables us to do that," Hyde said.

Atropos provides a Green Button that physicians can press to get evidence-based answers to clinical queries. Or they can choose for Atropos to automate evidence generation and have it incorporated automatically, Revri says.

"It also gives us transparency into how that agent is making decisions, monitoring the agent, evaluating the agents, to provide all sorts of governance around the agents," Revri said.

"ChatEHR calls Atropos with the right set of questions to get the results that we desire," Revri said. "That's the Atropos secret sauce. They have access to all sorts of studies that we would have a hard time getting all on-prem ourselves."

AI agents aid treatment decisions

Using Dragon and Atropos allows Stanford to personalize medicine, according to Hyde. While Dragon is recording data from patient visits, the Atropos Evidence Agent collects data on studies, such as which patients are suitable for GLP-1 vs. Metformin.

"We insert that new study and evidence right into the notes, so the doctor stops their visit and looks over at Epic, and right in there is evidence from the Atropos Evidence Agent on what to do, and they can make a bigger decision," Hyde says.

The tool inserts the study data directly into a patient's record to not only help with clinical decisions but also promote health equity.

For example, health systems could use Evidence Agent to collect real-world evidence for a physician treating a female Hispanic 45-year-old patient with a history of chronic kidney disease.

"By producing a study that is only run on similar patients, you can get a better view of how therapy might perform in that patient group and understand what the best choice is for patients that look like that," Hyde said.

Thousands of physicians also use an Atropos product called ChatRWD, a generative AI application that incorporates an AI-based chat interface to deliver observational studies.

"It's sort of a world now where agents talk to agents in the background, getting work done for the clinician," Hyde said.

The real-world evidence from Evidence Agent saves physicians time searching PubMed or around the internet, he added.

"When we have evidence for what to do with patients, doctors are more confident in their decisions," Hyde said.

In addition, Stanford is working with Microsoft to develop a healthcare agent orchestrator, which will extend the capabilities of its ChatEHR platform.

Addressing other healthcare problems

Combining Dragon with an Atropos AI agent not only adds a layer of automation to evaluating evidence for treatment decisions as part of a physician's workflow, but they can also help with determining pharmacy policy and seeking prior authorizations for drugs, Hyde said.

It allows health systems to decide whether to keep therapies on formulary, he added.

"We are in a moment in the technology industry where AI is helping to address many of healthcare's most urgent problems," said Peter Durlach, corporate vice president and chief strategy officer for Microsoft Health & Life Sciences, in a news release. "By collaborating with Atropos Health, we are working to help healthcare providers unlock new levels of productivity, reduce administrative burden and empower them to leverage new clinical decision-making capabilities in the care of their patients."

Going forward, Atropos plans to work with other health systems on AI agents as it is doing with Stanford.

"Being able to scale that across the health system and bring this value to physicians everywhere -- that's the goal," Hyde said. 

Brian T. Horowitz started covering health IT news in 2010 and the tech beat overall in 1996.