Artificial intelligence is becoming more of a priority for healthcare systems as executives look to streamline processes, save clinicians time and advance the patient care experience, Todd Norwood, senior director of clinical services for virtual care provider Omada Health, told an American College of Healthcare Executives webinar.
Norwood said survey data show that more than half of US health systems consider AI a top technology priority, but nearly 95% of actual projects end up not reaching objectives. Still, some AI use cases do meet with success, and the technology’s effectiveness is improving.
Data published in the Journal of the American Medical Association showed that AI scribes, typically used to document patient encounters, are associated with:
-
13.4 fewer minutes of electronic health record time per eight hours
-
16 fewer minutes of documentation time per eight hours
-
About 0.5 additional visits delivered per week
-
An added $167.37 in revenue per clinician, per month
When clinicians use AI scribes for 50% or more of their visits, they experience:
-
21.3 fewer minutes of EHR time per eight hours
-
27.3 fewer minutes of documentation time per eight hours
-
One additional visit delivered per week
-
An added $343.50 in revenue per clinician, per month
“We see a ton of potential with this technology,” Norwood said. He emphasized giving AI platforms time to have an impact, evaluating them to determine long-range effects, and setting realistic expectations about outcomes. He shared charts showing that the most significant results from new healthcare AI tools tend to occur 12, 14 or 16 months after implementation. “The really big changes are far down the line,” he said. “It’s something to consider as you are looking at piloting AI systems in your environment.”
Potential pitfalls
Norwood said Wall Street Journal research showed inconsistencies between executives’ and workers’ perceptions of the time AI tools can save. Executives often use AI platforms to quickly draft memos and other documents, and they tend to feel the tools do an efficient job. But, the data show front-line workers feel less positive about the time savings so it’s important to set realistic expectations tailored to different groups of users.
Other challenges include data siloing and fragmentation, inconsistent data formats and standards, and under-utilization of available information. “Ninety-seven percent of healthcare data go unused,” Norwood pointed out, noting that the quality of data is just as important as quantity. If the statistics behind an AI tool are not high quality and consistent, the output will not be accurate or useful, and patient outcomes may be affected, he said.
Real-world examples
Norwood offered an AI case study from his employer that was designed to improve patient care and satisfaction. The organization was interested in encouraging home exercise and reducing pain in a cohort of patients. Staff found that more frequent messaging and video visits between clinicians and patients improved the odds of exercise regimens being completed. They evaluated data and found that patients were more engaged if their first follow-up visit happened within eight days. Any more than that, and engagement levels dropped. Also, they used an asynchronous assessment tool to evaluate patients’ exercise technique. There was a dashboard so staff could see the data and use it to be more effective in their work, and internal training was held to ensure employees were comfortable with the technology.
Among the results of the program were a 61% increase in follow-up visits and a 7% improvement in pain outcomes. “And 92% of physical therapists said they found significant value in the use of asynchronous assessments,” Norwood said. “Providers have to see the value.”
Norwood said that leadership buy-in is one of the most important factors in the success of a healthcare AI program. “I’m not saying that as a senior leader, you need to know every single statistical test under the sun,” he noted. “But you’ve got to be able to hold your weight.”
Regular surveys can help measure senior leaders’ levels of competency and support, as well as staff members’ perceived access to needed data. It’s also important to set a goal for the percentage of major decisions that will be based explicitly on data, and to consider using high-quality data to track and improve staff literacy and performance.
Take-home tactics
Norwood said he and his colleagues gained valuable insights about AI as a result of their project. They included the importance of showing clinicians the technology’s value, demonstrating impact instead of just describing it, connecting the provider’s actions to patient outcomes, and giving multiple examples of an AI tool’s capabilities. It’s also important to start with smaller-scale projects with a narrower scope, he said. Ones that are too broad or ambitious aren’t likely to be as successful.
He pointed out that different age groups may have varying levels of acceptance for AI, so it’s helpful to meet them at their own comfort level. Veteran staff members may be less familiar than younger ones, so the key is to make the experience positive. “Give them places where they can ‘play in a sandbox’ and test it,” he said. “Then, start a trial-and-error process. They probably will say it’s not so bad.”