Inside Hackensack Meridian's first-in-nation Joint Commission AI certification

Exploring HMH's application process

To apply for Joint Commission's AI certification, health systems must be located in the U.S., operated by the U.S. government, or operated under a U.S. congressional charter. They should also comply with federal laws, such as those from CMS governing participation in Medicare and Medicaid or conditions for coverage. In addition, when the system has multiple healthcare organizations within it, such as physician groups and regional hospitals, one entity or subdivision of a health system must manage and govern AI activities.

Also, the health system must have processes already in place for the responsible use of AI, as well as data security and data use protections. The commission further requires quality monitoring and voluntary reporting of AI safety-related events.

Luckily, HMH already had established a series of AI governance committees when AI and LLMs first emerged, Klein recalled. The health system had this governance structure in place before bringing AI into the clinical or business space.

One of those structures covered HMH's approach to data security when using AI, so HMH was prepared when commission surveyors were interested in how the health system protected patient data. Sometimes third-party vendors insist on using a hospital's data for their own advantage -- to improve a product or monetize their product line -- but HMH takes a strong stance against this, according to Klein.

"We protect data through contractual restrictions," Klein said. "When HMH works with vendors, we negotiate agreements that the vendor cannot use our patients' data for their benefit unless they demonstrate how the data sharing would benefit the care of our patients."

HMH, which has 18 hospitals, received the certification within three weeks. It underwent the same level of scrutiny that hospitals typically undergo to achieve other Joint Commission accreditations, such as an inspection of operating rooms or emergency departments, Klein said. 

As part of the AI certification process, the commission specified which people at the hospitals it wanted to speak with and noted the technology tools it needed to assess on an inventory list. Klein shared that HMH uses various AI tools, including Microsoft Dragon Ambient eXperience (DAX) for ambient AI documentation and some Epic AI tools.

The commission wanted to examine how HMH evaluated, stratified and scored its AI tools as well as how the health system evaluated risk. The surveyor was not only experienced in evaluating AI but also knew how to assess health systems' processes.

"It was quite a bit of firepower being brought to the table," Klein said, noting that the surveyor had extensive experience evaluating health systems in the acute care space.

Tips for health systems when applying for the certification

Klein advised that health systems interested in earning the certification first look at Joint Commission's roadmap on implementing AI technologies. Joint Commission has also published a Survey Activity Guide that prepares hospitals for accreditation surveys.

In addition, health systems should have a checklist for evaluating AI technology, which could include steps for tracking the amount of protected health information collected or how far from an organization's IT environment data should flow to be useful. Another step on the checklist could be determining the degree of a clinician's involvement in the development of AI tools. Deciding what items go into the checklist is not a "black-and-white" process and "always [has] shades of gray," Klein explained.

Health systems could also use a quantitative tool to evaluate clinician unease with AI use, with a score of zero meaning no concerns, two indicating moderate concerns and three signifying serious concerns, according to Klein. Scores could then get added up to create an overall score on AI risk, which is an area Joint Commission surveyors will assess for the certification.

Klein added that health systems should create a framework for high-risk tools to help them decide whether the tools effectively address a clinical or business problem, whether they should be implemented and how to monitor them.

In the years to come, HMH will continue to evaluate how to integrate technology into care workflows safely, especially as the number of AI capabilities available skyrockets, Klein said. Already, many of the vendors HMH works with -- including Epic, Google and Workday -- offer AI features.

"We're following the [vendors'] roadmaps carefully so we can make good decisions on incorporating these AI tools into our workflows," Klein said. "We will do this with potential new partners as well. We will always consider a strong solution that helps us take better care of our patients."

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