Assuring the Future of Health AI

Thanks to the all too well-known breakthroughs in generative AI of recent, when coupled with the risk implications and resource constraints, the need to assure and accelerate trustworthy AI has never been more critical.

The Coalition for Health AI (CHAI) has been coalescing over the past year to do just that. CHAI seeks to harmonize standards and reporting for health #AI and educate end-users on how to evaluate these technologies to drive their adoption.?The outcome of our efforts will be a framework of guidelines for the ever-evolving landscape of #ArtificialIntelligence tools that extend across the product lifecycle and tightly aligns with the NIST AI Risk Management Framework and the White House Blueprint for AI Bill of Rights .

Earlier this week, in what has turned out to be yet another personal career highlight working alongside John Halamka, M.D., M.S. who had an early influence on my career, I had the privilege of helping to host members of the Coalition for Health AI (CHAI) on the 赛仕软件 campus in Cary, NC. SAS is a co-founding member of CHAI alongside organizations like Change Healthcare , Duke AI Health , 谷歌 , Johns Hopkins Medicine , John Snow Labs , Mayo Clinic Platform , 微软 , MITRE , Stanford Health Care , 美国加州大学伯克利分校 UCHealth , UCSF Health , and Vanderbilt University Medical Center . Additionally, multiple U.S. Federal agencies are closely observing including the Agency for Healthcare Research and Quality , Centers for Medicare & Medicaid Services , FDA , Assistant Secretary for Technology Policy , The National Institutes of Health , and the White House Office of Science and Technology Policy .

All of which is to say, we have the right people in the room. ????

"In the sense of trustworthy AI, when competing tech companies win, we win. If they lose, we lose. The fact that we can get into the same room and have these very open and transparent discussions for the sake of health care AI and make sure that we form guidelines that we can all agree on, that is a massive force." - Allie DeLonay

How CHAI will assure Trustworthy AI

This group of nearly fifty top voices on the use of AI in #healthcare is digging deep into the work of assuring that #TrustworthyAI is deployed in a manner that is secure, fair, transparent, useful, and safe.?David Page, Duke University Chair of Biostatistics, said it best, "We must make AI worthy of trust." To accomplish this, CHAI aims to produce the following outputs:

1. Assurance Standards Guide: Step-by-step plan for considerations at each stage of model development and implementation.
2. Evaluation Rubric: A checklist validation tool to be adopted either for self-review by the technical community or for independent review by health systems.
3. Community-Driven Transparency Information: a catalog including Model Cards, technical documentation, and documentation describing data provenance.

What I'm most proud of is this group's action imperative. Advanced #analytics cannot just be academic exercises, they must be operationalized. According to Udo Sglavo , that means we have to "ask the questions of what problem(s) we're trying to solve, then figure out the technology. Not the other way around." To put a finer point on it from Reggie Townsend , "Think before you code." Ethical considerations must be considered throughout the product lifecycle, especially during the upfront concept phase, something the Data Ethics Practice (DEP) at SAS has been researching and developing into our Viya platform already:

Reflections from the Convention

After a morning of keynote presentations that included a briefing from Kathryn Marchesini , Chief Privacy Officer at ONC, members broke into workgroups to begin developing considerations and evaluation criteria across each of the core principles:

1. Fairness, Equity, and Bias Mitigation
2. Privacy & Security
3. Transparency
4. Usefulness
5. Safety

One clarifying premise from Michael Pencina that pervaded all workgroups was the distinction needed between the technology/algorithm and the solution/strategy. Operationalizing AI is as much about #ChangeManagement as it is about #DataScience . Human Factors will become increasingly important as we incorporate more and more AI into workflows.

Second, end users must be involved throughout the AI lifecycle. This doesn't just mean #Physicians either. #NursePractitioners and #PhysicianAssocciates will be key users, as will #Nurses . Even #Patients need to be in the loop.

I have pages upon pages of more notes and I look forward to helping CHAI consolidate this work into outputs for the broader industry and policymakers. The challenge will be simplifying the sheer volume of important considerations so as to not overregulate and create undue burden and cost.

We closed out the day with Troy Tazbaz , Director of Digital Health for the FDA and Dr. Halamka summarizing the day's findings and discussing the future of CHAI.

"The only way forward is together" - Dr. John Halamka

My brain has never had to think so hard to try to keep up with the brilliant minds of John Halamka, M.D., M.S. , Shauna M. Overgaard, Ph.D. , Michael Pencina , Nicoleta Economou , Armando D. Bedoya MD MMCi , Sophia Bessias , David Page, Tim Suther , Matt Elmore, Nancy L. Walden , Marcella Dalrymple, MPH, MBA , Richard Shannon , Eric Poon , Ivor Horn, MD, MPH , Lisa Lehmann , Christine Dymek, David Talby , Mwisa Chisunka , Hailey H. , Matthew Lungren MD MPH , Lee Fleisher , Mary Greene, MD MPH MBA , the MITRE team that's been quarterbacking the coalition Brian Anderson, MD , Mary Jo Sullivan , John Morgan , Stephen Ondra , my SAS colleagues Gail Stephens , Stephen Kearney , John Esham , Hamza Ghadyali , Allie DeLonay , James DeNofrio , and Meral Tunc-Ozdemir , and the many others that joined us virtually for the day.

I’m hopeful about our future, I'm proud of SAS’ leadership in #dataethics , and I couldn’t think of a better group of people to be working together with to assure that future.