Health AI at CES 2025

The AI Wave Hits CES

 CES 2025 in Las Vegas showcased a tidal wave of consumer-focused AI products, signaling the industry’s drive to put AI-powered innovation directly into the hands of users. Among the buzz-gaining healthcare entries were Omnia, a combination mirror, scale, and AI assistant aimed at delivering personalized wellness insights through real-time biometrics, and FaceHeart Cardio Monitor, a device capable of detecting cardiovascular health indicators simply by analyzing facial expressions. While CES often flaunts concepts that evolve dramatically—or even evaporate—before hitting the market, the event remains as close as we'll get to a crystal ball for where consumer-focused healthcare companies see future trends evolving.

Companies appear eager to follow the success of devices like the Oura Ring and Fitbit, betting on the power of seamless day-to-day integration of biometric-centric technologies. Expect more ambitious ideas to emerge as these tools begin to leverage AI assistants (in place of nurses or doctors) for the interpretation of the results from these scans.

From our vantage point, the focus on AI wearables highlights growing consumer demand to cut through the red tape involved in managing health conditions and interfacing with the traditional healthcare system. Expect to see more where this came from.

EHRs Have to Disclose AI Details to Customers

The industry is finally seeing some movement from an initial wave of AI-specific regulatory decisions. Effective January 1, 2025, a key provision in the rule requires electronic health record (EHR) vendors to disclose detailed information about predictive models and algorithms embedded in their software.

Following November 2023's Biden executive order on the safe use of artificial intelligence (which specifically names healthcare AI usage), HHS released the Health Data, Technology, and Interoperability: Certification Program Updates, Algorithm Transparency, and Information Sharing - shortened to "HTI-1" - final rule (full rule here if you're into that sort of thing).

A core element to this rule is updates it provided to the ONC Health IT Certification Program which provides the criteria EHR's have to follow to be formally recognized as EHRs. And the downstream affect of this was to obligate large EHR vendors like Epic, Athena, and Oracle Health to follow these updated standards.

Key disclosure requirements under the new rule include divulging the following:

• Input Variables: Identifying the data elements that feed into each predictive model.

• Real-World Testing: Indicating whether the tool has been tested in practical, clinical settings.

• Bias Mitigation: Describing measures taken to address potential biases within the algorithm.

• Usage Warnings: Providing cautions regarding improper or unintended use of the tool.

The hope, is that these transparency will keep EHR companies honest by ensuring that end users, such as hospitals and healthcare providers, have a clear understanding of the predictive algorithms they utilize. You can read a good summary on the whole thing from Mario Aguilar at Stat.

Zooming out, the next six months will be telling to see if the Trump administration, eager to wipe away any Biden accomplishments, attempt to roll back some of these regulatory elements or simply replace them with their own while keeping the core effect. There's been lots of buzz on both sides but the proof will be in the pudding and all we can truly do is wait and see.

CHAI Open Sources AI Nutrition Label

Fierce Healthcare has a great write up on the contents of the Coalition for Health AI’s (CHAI) AI Nutrition Label that was unveiled last fall. We've written about CHAI consistently as an organization making great strides in their attempts to create benchmarks and standardization in how end users can compare different AI tools. And now they've pushed the tool to GitHub for anyone to use when developing a product.

Here are the core elements of the 'nutrition label':

• General Information:

  • Name of the AI solution

  • Developer's identity

  • Release stage, date, and version

• Summary:

  • Brief overview of the AI solution's purpose and functionality

• Uses and Directions:

  • Intended use and workflow

  • Primary intended users

  • Out-of-scope settings and cautioned use cases

• Warnings:

  • Known risks and limitations

  • Known biases or ethical considerations

• Trust Ingredients:

  • AI system facts (e.g., outcomes, model type, data sources)

  • Bias mitigation approaches

• Transparency Information:

  • Funding sources

  • Third-party involvement

• Resources:

  • Evaluation references

  • Clinical trials

  • Patient consent or disclosure requirements

I've said it before - organizations like CHAI are doing the lord's work in our industry. It's getting harder and harder to separate the wheat from the chaff as more AI products get pushed to the broader product. There's real risks involved in products being released that are misunderstood, poorly developed, or poorly used and better upfront disclosures on how a product was built, how it should be used, and where the pitfalls are can only be good for the industry as a whole. More guidelines mean more comfort with broader adoption.

That’s it for now. We’ll catch up again next week.

-Patrick

Enjoy this? Please take the time to forward to a friend or subscribe.