Health system leaders are viewing generative artificial intelligence as a way to cut costs and pare relationships with certain third-party vendors.
Excitement is palpable for AI in healthcare. A survey released last week by the Center for Connected Medicine at UPMC and market research firm KLAS Research showed AI was “dominating the thoughts of many executives at health systems.” AI was identified as the most exciting emerging technology by nearly 80% of health system respondents.
Read More: Dr. ChatGPT: A guide to generative AI in healthcare
Some executives see AI replacing third-party vendors operating in areas like medical transcription and patient communications. For health systems dealing with tight financial margins, the potential exists for the technology to be a cost saver.
"To be frank, the [software-as-a-service] subscription model that vendors use is financially not sustainable for health systems,” said Michael Hasselberg, chief digital health officer at Rochester, New York-based University of Rochester Medicine. “It may not be the most cost effective or personalized approach to go with an external vendor.”
Hasselberg has already seen the potential of generative AI. He said a team of data scientists and primary care clinicians spent over a year in 2019 attempting to build a natural language processing model to accurately triage patient messages to providers.
“We were unsuccessful and could not solve it,” Hasselberg said. “I had multiple data scientists [for whom] this was their sole project."
Hasselberg said there was too much variability in the system’s EHR data. His team did not have the resources to build a large enough AI model. Then came OpenAI's GPT-4 AI model this past March. He tried again to triage patient messages to providers but this time used GPT-4.
The results were staggering.
“It literally took two days,” Hasselberg said. “Two days to tune [GPT-4] to have 97% reliability, 86% accuracy of grouping those messages into a staff, nurse, or physician. [This] is a higher reliability rate and accuracy rate than a human nurse triaging those messages.”
Hasselberg is among a group of health executives who think GPT-4 and other AI large language models could empower health systems to develop their own use cases to solve clinical and administrative problems. This would essentially cut out the need for some third-party vendors.
Third-party technology vendors and point solutions have typically bridged the gap between clinicians and technology. Anil Saldanha, chief innovation officer at Chicago-based Rush University System for Health, said if his health system wanted to stratify risk for patients with high blood pressure, they'd traditionally have to use a third-party vendor.
Vendors will even hire their own clinicians to help plug the solution into a health system's EHR. But through the advancement of generative AI as well as many clinicians getting trained in informatics, the investment may be unnecessary, Saldanha said.
“We won't need these outside point solutions or third-party vendors to solve this for us,” Saldanha said.
An open-source AI library
Executives like Saldanha and Hasselberg envision an open-source library where specific AI models have been tested and could be plugged into any health system's EHR and tech infrastructure for free. Hasselberg said digital health companies with narrowly focused tech solutions would largely be replaced by the library. He sees open-sourced AI as the knockout punch to many point solutions.
“As a vendor, I want you guys working on higher-level problems,” Hasselberg said. “We’ve got so many ambient documentation vendors now. We've got so many vendors that are in the revenue cycle space or prior authorization space.”
Dr. Nigam Shah, chief data scientist at Stanford Healthcare, predicts a consolidation will occur among tech vendors partly because of AI. But he said it is unclear which entities will govern or own the space. Private companies, medical associations and even government stakeholders could all play a role, he said.
Shah said startups are likely to continue honing AI models for specific use cases and as the industry evolves, health systems could begin developing and deploying their own models, he said.
Sara Vaezy, chief strategy and digital officer at Renton, Washington-based Providence, takes a slightly different view, cautioning that maintaining these models is expensive. Vaezy also said solutions operating deeply within technologically complex areas or those with a large scope could remain beneficial to health systems.
“I don't think [third-party vendors] will go away all together,” Vaezy said. “Especially if you think about it, some point solutions go really deep into a specific domain.”
