An Insider Guide to AI Innovation at Large Health Systems

Q&A | From Skepticism to Success Stories

Healthcare is a complex world, often resistant to change yet brimming with potential for groundbreaking innovation. It is often intricate, governed by strict regulations and complex hierarchies. At the forefront of this dance are individuals like Sandra Bossi, who bridge the gap between the promise of AI and the realities of its implementation within healthcare.

During our recent encounter with Sandra, she generously offered to address a few pressing questions -which were meticulously crafted to serve as a valuable playbook, drawing from someone deeply embedded in the healthcare innovation system. 

She shares candid reflections on the excitement and skepticism surrounding AI solutions and the critical steps for successful integration. Her insights offer practical advice on introducing AI into healthcare, enriched with personal anecdotes and unfiltered takes.

‎‎

Q1.

Can you recall the very first time you encountered an AI solution being proposed for use within the system? What was the overall atmosphere surrounding this new technology? Looking back, how dramatically has the landscape transformed since that first encounter with AI?

Absolutely! My first major ML/AI project was in 2016, developing a computer vision system for fracture detection using radiographic images with a startup. Back then, the atmosphere was a mix of skepticism, curiosity, and fear. Many healthcare professionals and colleagues lacked full understanding of AI’s potential, questioning its capabilities but also fearing it could replace radiologists. We had to take time to explain how this emerging technology works, and how it could augment clinicians' skills and be responsibly integrated into clinical practice.

Fast forward to 2021, when we launched our latest and most ambitious project—the first-ever Innovation Center for Artificial Intelligence in Robotic Joint Replacement at HSS, in partnership with a leading med-tech company. By this time, AI had become integral to healthcare discussions. The question was no longer whether these technologies would be used in healthcare, but determining the optimal ways to leverage them. The development of future robotic systems heavily relies on AI algorithms and real-time intraoperative data collection. It’s completely transforming the field of surgery.

The initial apprehensions and  fears have shifted into challenges related to implementation and optimization. It’s thrilling to witness AI now embraced as a valuable ally in accelerating precision in surgical techniques and improving patient outcomes.

‎

Q2.

You've mentioned the increasing responsibilities and scrutiny involved with AI in healthcare. Can you elaborate on the concerns and how they impact administrators and clinicians when implementing AI technologies?

While AI has proven that it can significantly improve diagnosis and care, leading to better outcomes, it also brings new challenges. As we handle increasingly larger datasets, our reliance on technology grows, yet the crucial human element remains – the final say on understanding and approving these reports and decisions rests firmly with us.

Like I said, back in 2016 when we developed our first computer vision system for fracture detection, there was skepticism and fear of job replacement. Today, the focus of leading healthcare systems has shifted to integrating AI tools, training staff, and using these tools effectively. The key challenge isn't about willingness to adopt these tools, but in ensuring their appropriate use, providing adequate training, seamlessly integrating them into healthcare workflows, and maintaining continuous QA and monitoring.

One of the biggest hurdles in healthcare regarding AI implementation is the limited window for innovation and experimentation due to the risk-averse nature of the industry and the challenges in disrupting clinical operations. This slows down the deployment and development of AI solutions.

‎

Q3.

What would be the evaluation process for innovators/ startups wanting to get into a healthcare system’s innovation programs? Can you outline the exact roadmap for them to get a peek through the window?

The evaluation process is meticulous and transparent. I would evaluate over 100 opportunities and technologies annually. It starts with evaluating the clinical potential of the idea, its impact on the clinical landscape, and the level of commitment from the inventor or the developing company. Additionally, I would look into the team behind the idea, their execution track record, and the resources needed for project completion. It’s important not to be misled by visually appealing decks that overpromise. To level-set, my initial question would be: What can you do today versus what do you plan to do in the future?

Next, I would assess the entire project/product roadmap to ensure alignment with our goals and whether it's something we can realistically take on. Identifying internal stakeholders who can contribute expertise and take responsibility for the initiative is crucial. I would look for tangible proof of concept and practical integration capabilities.

And finally, we meticulously examine their ability to work within our infrastructure, verifying their ability to meet our IT and information security standards, and seamlessly integrate their solutions without disrupting operations. Their adaptability to a health system’s environment and previous experience in healthcare are significant factors. 

‎ ‎

Q4.

Is the healthcare system closed to outsiders, especially those without healthcare experience, given the rigorous triage process? And, how can virtual companies collaborate with health systems considering the guarded nature of data integration within system premises?

Navigating healthcare may seem like a tough nut to crack for outsiders, especially those without prior experience in the field, but it’s not entirely off-limits.

In countries with centralized national patient databases, startups may find it easier to access data, but in the U.S., partnerships with multiple healthcare systems are often necessary to access diverse datasets. Typically, startups are required to develop and train their models within the secure infrastructure of healthcare systems. The Mayo Clinic model exemplifies this approach, allowing startups to run and train their models on the healthcare system’s platform without transferring data offsite. This setup enables startups to enhance and validate their algorithms while ensuring data security and usage alignment with the intended project purpose. While healthcare systems are in principle open to collaboration with external entities, specific protocols regarding data access and usage must be adhered to for successful partnerships and project realization.

‎

Q5.

How can innovators effectively demonstrate the value of their solution to healthcare systems, especially when they are still in the early stages and can't rely solely on traditional financial metrics?

By showcasing how their solution can generate real value and improve efficiency, innovators can build a strong case for support without relying solely on traditional financial metrics through a vendor type relationship.

Startups often encounter challenges when expecting healthcare systems to invest in a product that is still in its early stage of development. It's important in my opinion for them to secure funding from alternative sources and view the healthcare system's involvement in early product utilization as an invaluable contribution of expertise to their product development and validation.

Healthcare systems offer critical clinical insights and best practices that are essential for creating viable clinical products. Instead of charging healthcare systems upfront for products or services that are not off-the-shelf, startups should recognize the indispensable role of healthcare systems in co-developing the product. Based on a healthcare system’s potential contributions, startups can offer incentives such as milestone payments, grants, discounted product pricing, even licenses on IP incorporated into their products.

‎

Q6.

Given the slow pace of healthcare adoption, what advice would you offer to health-tech startup founders on managing expectations and staying committed to long-term relationships despite initial hurdles and slow progress?

When selecting healthcare partners, startups should identify organizations that are faster and more experienced with innovative initiatives. However, even the most forward-thinking healthcare systems have to navigate regulatory hurdles, policies, and politics, which can significantly slow down the innovation process. 

Choosing partners who understand the startup world and have already made investments in innovation can make a difference.

Having a strategic and flexible product roadmap that can be adjusted as they progress through the development and integration process is also essential.

‎

Q7.

Legacy technologies often hold a monopoly in healthcare, making it challenging for startups to break through. Are there any impressive AI technologies that have managed to overcome these hurdles

In the field of robotic surgery, an interesting new entrant is a unicorn startup CMR Surgical that has developed the Versius robotic surgical system. It allows for better surgical precision through use of AI-empowered surgical guidance tools while integrating into clinical workflows without major disruption and with much lower cost. I am a deep admirer of the Da Vinci system that has truly transformed the field of surgery. It’s exciting to see how innovation continues to push the boundaries of surgical precision and accessibility. 

In the field of medical diagnosis based on imaging interpretation, a number of AI startups have raised the bar in enabling early disease detection and improving diagnostic accuracy, such as US’s Viz.ai, South Korea’s Lunit, Israel's Aidoc, France’s Owkin, India’s Qure.ai, etc. Contributions of AI technologies in the medical imaging interpretation are undeniable and it’s only a question of time before AI-assisted diagnosis becomes a standard practice.

We still await a breakthrough in this area, and I believe future advancements will significantly enhance how we handle and leverage healthcare data.

‎

Q8.

What questions are you most curious about regarding the future of AI in healthcare? If you had to pick the top two questions you’d like to pose to an AI expert, what would they be?

One burning question I have is how they envision the future beyond electronic healthcare records and whether it would allow us to effectively address the present challenge of data availability. While AI boasts exceptional computational capabilities that enable endless hypothesis exploration, accessing and integrating all necessary data remains a significant obstacle. I'm curious if AI experts foresee a future where we have a ubiquitous healthcare data platform that, seamlessly integrates diverse data sources overcoming the current complexities of data privacy and ownership, and truly unlocking the transformative power of AI to advance personalized medicine and revolutionize healthcare.

Another is about the evolving role of healthcare professionals in an AI-enabled world. As AI trains on larger datasets and provides deeper insights, how should we train clinicians to work utilizing these advanced tools?

I'm eager to learn more about the strategies they think are the best for educating healthcare professionals so they can harness AI's full potential and continue to transform patient care.‎

And there you have it!

Our deep dive with Sandra Bossi revealed some eye-opening insights into the present and future of AI in healthcare. From overcoming the hurdles of data accessibility to reimagining the role of clinicians in an AI-enhanced world, it’s clear that the intersection of technology and medicine holds endless possibilities.

But this is just the tip of the iceberg. Curious about how AI might revolutionize patient care even further? Stay tuned for our next edition, where we'll continue exploring the cutting-edge developments shaping the future of healthcare.