Rankings, Rules, and Real-World Applications

Collaborative Push to Rank AI Models in Healthcare

Leading health systems, including Mass General Brigham (MGB), Emory Healthcare, and others, have formed the Healthcare AI Challenge Collaborative to evaluate and publicly rank AI tools from tech giants like Google, Microsoft, Amazon, and OpenAI. The goal is to help healthcare providers make informed purchasing decisions by directly comparing AI tools' performance in clinical simulations.

Key Takeaways

• Problem: Healthcare providers struggle to assess and compare AI tools due to a lack of standardized evaluation metrics. Current methods rely on anecdotal evidence or user surveys.

• Solution: The Collaborative will test nine AI models in areas such as diagnostic accuracy, report readability, and differential diagnosis generation. Results will be released as a public "leaderboard" by year-end.

• Participants: MGB is joined by Emory, the University of Washington, the University of Wisconsin, and the American College of Radiology. More participants may join.

• Potential Impact: Transparent rankings could democratize access to AI insights, particularly benefiting smaller, resource-constrained systems that lack the capacity for independent testing.

Why This Matters

The surge in healthcare AI tools has created an urgent need for reliable evaluation methods, and initiatives like the Healthcare AI Challenge Collaborative and the Coalition for Health AI (CHAI) are tackling this challenge from complementary angles. While CHAI focuses on establishing national assurance labs to independently validate AI tools and ensure safety, efficacy, and equity, the Healthcare AI Challenge provides a real-world testing ground where clinicians assess tools directly through simulated use cases, producing public rankings to guide purchasing decisions.

This dual approach—CHAI's structured, regulatory-centric model and the Collaborative’s hands-on evaluation and feedback—reflects the necessary chaos of navigating a new and fast-moving field. It’s crucial because the healthcare market is still grappling with a lack of standardization in AI performance metrics, and these efforts offer distinct, yet interconnected solutions. The Collaborative’s clinician-driven feedback loops aim to benchmark utility in clinical practice, while CHAI’s framework is more foundational, working toward long-term governance and assurance.

The novelty of these tools and the fragmented evaluation landscape underscore why such experiments are needed. Both efforts highlight a broader truth: while chaotic, this era of trial and exploration is essential for shaping the responsible adoption and meaningful application of AI in Full Article | Read The Announcement

Challenges in Building a National AI Assurance Network

At the Coalition for Health AI’s (CHAI) Global Summit, stakeholders explored the framework for a national network of AI assurance labs, intended to evaluate healthcare algorithms for safety, bias, and performance. These labs aim to serve a wide range of use cases, from pre-FDA validation to algorithm testing for health systems, but key questions remain unresolved about funding, scope, and equity of access. Participants also discussed the potential for a national registry of validated AI tools to reduce redundancy and promote shared standards.

Key Takeaways

1. Primary Use Cases: Assurance labs could validate algorithms for vendors prior to FDA review or for health systems evaluating tools for deployment, but funding models remain contentious. Options include vendor-paid validation, buyer-paid testing, or shared costs.

2. Addressing Equity: Concerns about accessibility for small and rural health systems were raised, with ideas floated to streamline engagement and share validation responsibilities across better-resourced institutions.

3. Scope and Specialization: Labs will need to balance broad coverage of AI types with deep expertise in specific models, raising sustainability concerns for labs focused on niche technologies.

4. Conflict of Interest: Transparency requirements will need to address potential bias, particularly if larger institutions validate their own or competitors’ technologies.

5. Pooling Resources: There is potential to reduce inefficiencies if health systems collaborate on evaluations, but concerns about intellectual property and competitive dynamics complicate such efforts.

Why This Matters

With the flood of healthcare AI tools entering the market, it’s vital to have CHAI’s network of independent assurance labs to help healthcare providers and systems compare tools and to build trust amongst the larger ecosystem. CHAI’s task is daunting. Among the key hurdles: defining who pays for validation, balancing thorough testing with financial sustainability, and ensuring a clear definition of ‘independence’ to avoid bad actors tipping the scales.

Our perspective: perfection shouldn’t be the goal. CHAI’s priority should be to get these assurance labs operational, showing the healthcare ecosystem that even a foundational evaluation system can offer meaningful insights. By providing tangible tools to evaluate the burgeoning AI marketplace, CHAI can create momentum, fostering innovation while helping providers make more informed decisions about the tools they adopt. Full Article

Presbyterian Partners with RhythmX AI to Ease Primary Care Burdens

 Presbyterian Healthcare Services has partnered with RhythmX AI to deploy a generative AI platform aimed at reducing the administrative and cognitive burdens on primary care clinicians. The pilot project will launch in December, utilizing AI copilots to assist with clinical decision-making, risk stratification, and patient care orchestration, ultimately hoping to streamline workflows and improve patient outcomes.

Key Takeaways:

1. Primary Care AI Copilots: RhythmX AI’s platform integrates with EHR and payer data to provide clinicians with predictive insights and patient-specific recommendations for holistic care, reducing the time spent on manual chart reviews.

2. Pilot Program Launch: Presbyterian will test the platform in select primary care clinics, focusing on improving clinician efficiency, patient outcomes, and access to care in rural New Mexico.

3. AI-Driven Collaboration: The health system and RhythmX AI are tailoring the platform to address specific clinical needs and workflows, emphasizing clinician input and on-the-ground observation to optimize usability.

4. Expanding Reach: RhythmX AI plans to partner with up to 15 health systems by 2025, leveraging its extensive data assets and the backing of SAIGroup to scale its generative AI solutions.

Why This Matters

Presbyterian Healthcare’s partnership with RhythmX AI underscores a growing trend in healthcare AI announcements: health systems are seeking pragmatic applications of AI to address immediate, tangible challenges, like reducing bedside workflow burdens and supporting clinicians in their day-to-day tasks. These efforts focus less on futuristic promises and more on delivering real-world improvements where they’re needed most.

What’s particularly striking is the diversity of vendors securing hospital agreements—reflecting a market still in its exploratory phase. Not only does this highlight the absence of a dominant AI player, but it also shows how health systems are stepping outside their traditional EHR relationships to pursue innovative solutions. This variety is a positive sign, as it fosters experimentation and allows the industry to test different tools and strategies, helping to shape a more practical and sustainable future for AI in healthcare. Full Article

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.