ORLANDO – At HIMSS24 on Wednesday, officials from the Office of the National Coordinator for Health IT offered some reminders and refreshers on a significant final rule that took effect for many healthcare organizations just over a month ago February 8.
And ONC is also previewing what to expect for the next related rulemaking.
This past January, the agency published the final rule for the (deep breath) Health Data, Technology, and Interoperability: Certification Program Updates, Algorithm Transparency, and Information Sharing Final Rule (much more pithily phrased as HTI-1) to the Federal Register, with an effective date of March 11, 2024.
Read more of our coverage of HTI-1, here and here.
The rule implements more provisions of the landmark 21st Century Cures Act and makes updates around standards, implementation specs and other criteria for the longstanding ONC Health IT Certification Program.
Among them: new interoperability-specific reporting metrics, revised info blocking rules, promotion of USCDI v.3 as a baseline standard and, most significantly, algorithm transparency – posting new requirements for how artificial intelligence and other predictive algorithms that used in certified health IT products.
At HIMSS24 this week, Jeffery Smith, deputy director of the certification and testing division at ONC offered some thoughts on the agency’s approach to HTI-1 and its approach to AI-enabled clinical decision support.
“When we update certification criteria, sometimes it’s because there’s new standards available,” said Smith. “Sometimes we want new functionality available, and sometimes it’s because there’s a deep policy driver that is needed.”
Three years ago, ONC was tasked by HHS Secretary Xavier Becerra to take a look at bias perpetrated and propagated by algorithms in healthcare.
“Our entry point was really looking at: How does this technology help and hurt?” said Smith.
“You can walk down the show floor and see that there are just literally hundreds, if not thousands, of different applications that can help doctors do a better job, help nurses and other clinicians do their job more efficiently,” he said at HIMSS24. “But we also understand that there is an opportunity for harm to enter into the equation.”
And so, ONC spoke with stakeholders and examined peer-reviewed literature to explore the various ways AI can cause problems in healthcare settings.
The technology “can do a lot of good, but it also can do a lot of bad – and it can do bad at scale,” said Smith. “We’re not talking about one patient getting hurt but hundreds, thousands and in some cases millions. There are very well-known peer-reviewed studies that look at the use of technology being used across populations to determine who is going to be sicker in the future, predict who’s going to need more care in the future – they were systematically biased, and that had an impact on literally millions of patients.”
ONC is not a safety regulator like the FDA, which is exploring approaches to AI-based software as a medical device. But it does have certification authority over IT products.
“Viewed of through the prism of our certification program, we thought ‘OK, what can we do?'” said Smith. “How can we help people understand the quality of information or the quality of the predictive model, and what kind of information would go into understanding that? So, we came up with an acronym.”
That handy acronym? FAVES.
High quality predictive algorithms and decision-support interventions ought to have five characteristics, according to ONC. They should be:
Fair.
Appropriate.
Valid.
Effective.
Safe.
“That’s our guiding principle, our quality framework for predictive algorithms in healthcare,” said Smith. “It was something that we thought about early, and it really drove the rest of our work with HTI-1, and specifically with the decision-support and intervention criteria.”
Since then, the five FAVE criteria have been used in other instances, such as the Biden Administration’s recent announcement that it had secured pledges on AI safety and ethics from some healthcare heavy-hitters – Boston Children’s, Geisinger, UC San Diego, Wellspan and two dozen other providers, payers and consumer health companies – promising their AI and machine learning models will be geared toward fairness, appropriateness, validity, effectiveness and safety.
With the HTI-1 rule, “we thought if we could bring more transparency to the data that are used as part of the decision support intervention, if we could bring more transparency to the performance of the decision-support intervention, and if we could bring transparency to the organizational competencies of the organizations that develop these things, then we can instill more trust and we can help optimize the use of these algorithms,” said Smith.
“Because at the end of the day, we are AI optimists,” he said. “We are the ONC. We believe in technology and its ability to improve patient care. And so we’re trying to think, how can we lower the barriers, and how can we increase the quality and use of these things.”
‘We are not slowing down’
That same day, National Coordinator Micky Tripathi posted a blog on the ONC site offering some foresight into what stakeholders can expect from the still-in-development rules for HTI-2.
That stands for (again, deep breath) the Health Data, Technology, and Interoperability: Patient Engagement, Information Sharing, and Public Health Interoperability Proposed Rule.
“This will be the second rule in ONC’s series of Health Data, Technology, and Interoperability rules,” Tripathi writes. “In HTI-2, you can expect to see a strong focus on interoperability and a specific focus on how strategic standards adoption can further interoperability.
“In fact, we see the impact of standards adoption every day in our ongoing work on United States Core Data for Interoperability (USCDI). We are thrilled to see the uptick in USCDI use across the healthcare sector and amongst our federal partners.”
With HTI-2, ONC also seeks to address new advancements for public health through certified IT requirements, he said, and expects to propose new provisions related to API certification focused on use cases, including electronic prior auth, patient engagement and care coordination.
“We are not slowing down,” said Tripathi. “Our work to support patients, to help providers as they seek to provide care, and to create clear expectations for the entire care continuum when it comes to health IT is ongoing.”
Mike Miliard is executive editor of Healthcare IT News
Email the writer: mike.miliard@himssmedia.com
Healthcare IT News is a HIMSS publication.
