Health systems looking to use generative artificial intelligence are making investments into the cloud.
Generative AI has excited health system leaders over the technology's potential administrative and clinical use cases but it's a costly investment for health systems at a time when margins are tight. Leaders nevertheless are bearing the added infrastructure costs to tackle initiatives such as transitioning data from physical to cloud-based servers.
Related: 4 steps to shore up health equity using AI
“At the beginning of this year and late last year, more discretionary spend was possible but the early returns are not good so there’s much more of a reluctance to actually go out and buy more infrastructure on the AI side, or the data infrastructure needed to support AI,” said Erik Pupo, commercial health information technology advisory director at consulting firm Guidehouse.
Health systems such as Stanford Medicine are exploring many areas for AI, which increasingly means moving their data to cloud-based systems to accommodate the shift.
“[AI] accelerates the move to the cloud because it's much easier to access certain services if your data is on the cloud,” said Anurang Revri, chief enterprise architect at Stanford Medicine. “Most vendors are investing in the cloud, so you’re not left with a lot of choice unless you want to spend a lot of money moving data back and forth between physical and cloud-based servers.”
Half its data is in physical servers and half is in the cloud, Revri said. Stanford also has made significant investments including a dedicated data science team with 15 full-time employees. It’s also invested in infrastructure to use AI for clinical research because Revri said the cloud allows for easier collaboration in and out of the health system.
Like others, Stanford is piloting generative AI to draft clinician messages and document in the electronic health record system. The system also has developed a small language model that can help clinicians answer queries within the EHR, Revri said.
New York University Langone Medical Center has created a unified analytics and IT team, a structure that helps find the best uses for AI deployment, said Dr. Devin Mann, strategic director of digital health innovation.
Langone has used generative AI to improve the claims denials process, patient-doctor messaging and clinical documentation. The system also has its own large language model trained on a decade’s worth of clinical notes from its inpatient records. Using a private cloud server has been critical for patient data security, Mann said.
“We do everything behind our firewall. That was one of the first steps,” Mann said. “We don’t send anything to the public cloud ever.”
The costs to build AI capabilities can add up. Revri said the goal is for around 20% of Stanford’s IT budget to be dedicated to AI costs, although he declined to say how much the system spends on the technology.
“That's the challenge for us as an industry, how do we invest in these future technologies when you don't even have margins? Stanford is fortunate…we’ve done well enough financially that we can invest in these technologies,” Revri said.
Health systems can’t skimp because investing in a cloud-based infrastructure is necessary to expand or contract AI solutions, experts say. And while infrastructure costs are becoming more affordable, Pupo said AI remains expensive for other reasons.
“Infrastructure costs are actually starting to move down but the actual cost to build models, to train them, continues to get bigger and bigger,” Pupo said. “When you build a very large model, you need enormous amounts of data and you're constantly training it, which means you're constantly bringing in more and more data.”
Health systems are adopting cloud solutions from Google and Amazon to ignite their generative AI efforts. In October 2023, Google Cloud said it was working with Mayo Clinic and Hackensack Meridian Health for the rollout of new AI search functions. Amazon Web Services, a subsidiary of the big tech company, launched a generative AI tool focused on clinical documentation in June 2023.