The AI 'landscape is vast' in health care, but which products are available?

Key takeaways:

• Generative AI products have the potential to save clinicians time and reduce burnout.
• Expert recommends deliberate deployment of AI-based tools in health care, with governance and oversight.

Health care providers know of the broad potential AI offers for reducing burnout, but which products are actually available for use can be difficult to ascertain, especially considering electronic health record system compatibility.

AI has the potential to aid primary care providers in a variety of ways according, to Deepti Pandita, MD, FACP, FAMIA, an associate professor of medicine, vice president of clinical informatics and chief medical information officer at the University of California, Irvine. For example, some tools can enhance diagnostic accuracy, improve patient engagement, streamline administrative tasks and “help with decision support via predictive analytics,” she said.

“The global population is aging, and with this, the prevalence of chronic diseases is rising, leading to an increased demand for health care services,” Pandita told Healio. “PCPs face growing patient loads, making it challenging to provide timely and personalized care. There is also a growing dearth of PCPs — because these are often lower paying specialties — and trainees tend to gravitate to super and subspecialties that are higher paying.”

Enlarge 

All of these factors contribute to a mounting stress in the specialty, which means one of the most exciting prospects for AI is its potential to reduce burnout rates.

“PCPs often spend a significant amount of time on administrative tasks, such documentation, billing and coding,” Pandita said. “This reduces the time available for direct patient care. Increasingly, PCPs need to interpret large datasets not only from the EMR but also from devices such as remote patient monitoring devices, BP and blood glucose machines, etc., adding to the amount of work and cognitive load of the PCP.”

Ambient technology like chatbots, scribes and virtual assistants have quickly become popular, and it can be challenging for PCPs to keep up with what is available, particularly given the added complication that electronic health record (EHR) integration may pose.

“The landscape is vast, ranging from ambient documentation tools, AI scribes, coding assistance, chatbots and virtual assistants, automated claims and prior authorization processing,” Pandita said.

Scribes, chatbots and virtual assistants

Pandita mentioned several specific AI-powered medical scribes, among them Abridge (Abridge), DAX Copilot (Nuance Communications, Inc), Nabla (Nabla Copilot) and Suki (Premier, Inc).

“AI-driven chatbots can handle routine inquiries, schedule appointments, provide medication reminders and answer general health questions,” Pandita said. “This improves patient engagement and ensures timely communication, including answering in basket patient messages within the EMR, which typically are a huge time drain for the PCP.”

Sean McGunigal, director for artificial intelligence at EHR software provider of Epic, told Healio that the company has seen “rapid adoption of ambient technology.” Currently, more than 200,000 visits each month are documented in Epic with ambient technology, he said, and the EHR has “interfaces that are used by many of the common ambient voice vendors, including Abridge and DAX Copilot.”

“At University of Michigan Health – West, PCPs save an average of 10 minutes per day using ambient listening technology to draft notes,” McGunigal said. “This brings joy to the practice of medicine. By freeing providers from the need to take detailed notes during a patient visit, it allows for more human-to-human interactions between providers and patients.”

Additionally, Epic has developed “several in-house” virtual assistants or chatbots, McGunigal said.

“Our approach is to embed AI assistant features directly into the workflows that users follow today,” he added.

For example, one of the new technologies, “In Basket Art,” is used to create a first draft response to patient messages, “so that clinicians have a starting point for writing back, ... [which] helps providers create richer responses in less time,” he said.

“We've seen rapid adoption of In Basket Art,” McGunigal said. “Art has saved nurses at Mayo Clinic 30 seconds per message, and Mayo Clinic projects organization-wide savings of 1,500 hours per month. UC San Diego Health Care published a study showing that time-crunched physicians who may only have time for a brief, facts-only response found that generative AI is helping to draft longer, more compassionate responses to their patients.”

Another tool called “Patient Summaries” can “distill clinical information into a brief summary” for clinicians to review, “significantly reducing the amount of time needed to prepare for a high-quality patient visit,” he said.

There is also, “SlicerDicer SideKick,” which “combines the power of our SlicerDicer ad-hoc analytics tool with generative AI,” McGunigal said. This means that users can “ask plain-English questions (eg, ‘how many kids are using inhalers?’) to query their Epic data,” he explained.

Plus, there are many more virtual assistant functions enabled by AI that are currently in development, he said.

Prior authorization and predictive analytics

Tools for prior authorization and predictive analytics are also attractive options for providers.

“AI can streamline the prior authorization process by quickly analyzing patient data, medical necessity and insurance requirements, reducing the time and effort required by health care providers,” Pandita said. “Predictive analytics can identify patients at high risk for specific conditions, enabling early interventions and personalized care plans.”

Epic is developing generative AI applications that identify “relevant clinical information to support medical necessity for authorizations and to answer payer questionnaires needed for authorization submission,” McGunigal said.

“Our Payer Platform supports connections between providers and payers,” he added. “These connections, combined with generative AI tools to assist with supporting documentation, will drive a more automated workflow for authorizations.”

For predictive analytics, McGunigal said the company has more than 30 prebuilt predictive models that are part of the Cognitive Computing Model Library, and their customers have created at least another 200 custom models using Epic’s tools.

“These models cover acute care, population health, capacity forecasting, revenue cycle and more,” he said.

Today, more than 440 organizations use machine learning models in Epic, and many have reported success, according to McGunigal. For example, he noted that Jefferson Health uses models to predict the risk for an ED visit or hospital admissions.

“These alert care managers of at-risk patients [offer support] so that they can intervene and prevent the need,” he said. “Jefferson reduced 30-day readmissions by 13%, saving $2 million in 3 years by avoiding penalty payments.”

Additionally, as part of remote sepsis monitoring, Ochsner Health uses a predictive model that has led to a 56% reduction in sepsis mortality.

Novant Health identifies hospitalized patients at risk with a patient deterioration predictive model, and over an 11-month span, the program was associated with an estimated 153 saved lives with a mortality reduction of 22%. Stanford Health Care similarly launched an AI model that identifies patients with clinical deterioration. McGunigal said that, in the pilot, clinical deterioration events were associated with a 20% reduction, and participating nursing staff left positive feedback.

Importance of humanity

McGunigal said that Epic’s approach to AI development “is to embed these tools directly into existing workflows, which makes them more accessible and easier to learn.” Additionally, the software provider works with PCPs and their organizations to train users on how to use new tools.

“We've worked hard and thoughtfully to rapidly develop, deploy and implement generative AI solutions that address needs across the health care industry, as identified by our customers,” he said. “We have over 60 generative AI projects in the works.”

Pandita stressed that, although there is a wide range of ways in which AI can help PCPs, it is important to use caution when deploying these tools and focus on augmenting rather than replacing a provider’s work.

“AI uses need to be validated and have governance and oversight,” she said. “Keeping the ‘human in the loop’ is always best.”