Cedars-Sinai CIOs Tips to Ensure GenAI is Fair, Appropriate, Valid, Effective and Safe
Craig Kwiatkowski, senior vice president and CIO at Cedars-Sinai, and his team have pushed the famed healthcare system forward in the field of artificial intelligence.
Their most notable achievement is the AI-powered virtual primary care app Cedars-Sinai Connect. Launched in late 2023, the app has already increased primary care capacity by 11% – the equivalent of building three new clinics. It has helped more than 6,900 patients – from San Diego to Sacramento – access care through more than 9,200 virtual visits.
But that’s just one of AI’s achievements. Kwiatkowski, who has a doctorate in pharmacy, is steeped in technology and finds ways to improve outcomes, enhance the patient and caregiver experience, and reduce physician burnout.
Below is the eighth interview in our series on top voices in healthcare IT discussing AI. In this part one we talk to Kwiatkowski about his views on current topics AI in healthcare in general. Tomorrow in part two we will discuss AI projects at Cedars-Sinai.
Q. How do you decide whether to build or buy AI, and what are the biggest challenges in integrating AI technologies into existing systems?
A. The question of building versus buying is situational, and we start by looking at each with fresh eyes and trying to understand the problem we are trying to solve. And where possible, we like to look for solutions within our existing systems and platforms.
So we’re an Oracle store for our ERP and we’re an Epic store for our EHR. If the workflow is enabled on one of those platforms, we will typically start working with that functionality where possible. And those suppliers and other major suppliers are building a nice roadmap of tools, and we want to lean on those that we can.
An example could be from Epic. We’re starting to use in-basket message response technology, something others are using, where a draft is queued for the clinician to further edit and send.
Another thing we’re starting to work on is graph summary capabilities, which use AI to sift through all the information in the graph and start a course note or maybe even a resignation letter. Within these supplier solutions we try to find a solution for physician burnout and well-being.
If that isn’t solved well with one of those platforms or solutions, we’ll look to other vendors to support a solution and consider whether we should build something ourselves. There are clearly many variables in that decision and perhaps too many to discuss here, but thematically the most important variables would center on capabilities and resource availability, feasibility and efficiency.
Sometimes we can talk our way into accelerating abilities. Ambitious documentation solutions are a good example. We’re not going to try to build that ourselves. That would be too difficult, too time-consuming and too expensive. But sometimes we are willing to invest and build when we see a gap in the market, or when it just makes sense for us to do it ourselves.
DIY gives us a little more control and flexibility to create something tailored to our specific needs. I think as much as we like to think, and people in healthcare in general like to think, that we’re reaching a point where healthcare is standardized and scalable, but there are still a lot of customized workflows and processes that do not always work well. to buy, or you buy and you still spend a lot of time on local configuration.
So there will probably always be some kind of trade-off.
The second part of your question was about integration and workflows. I alluded to that a little bit, but the more we can do to integrate these tools into existing workflows, the better off we’ll be. Honestly, especially from a doctor’s perspective, taking them out of their workflow is a non-starter. No separate logins, no side-by-side portals or duel screens, no extra keystrokes, ideally fewer keystrokes.
And that is what will lead to success in integrating and adopting these solutions, both in the short and long term.
And sometimes, whether we like it or not, these solutions are simply necessary workflow redesign. In some cases there is no escape. And so in those cases we need a good ‘why’ story and all the necessary change management to support people who will have the most impact on the front lines.
Q: How are patients and physicians responding to these tools, and can healthcare do more to accelerate their use or broaden adoption?
A. We have been using AI for many years, mainly in the rule-based and machine learning varieties, and we have had great success incorporating these types of tools into the workflow and good adoption in all kinds of use cases – patient risk predictions, deterioration, length of stay capacity, patient flow.
But as for the newer one generative AI tools, it’s still very early. I think we’re taking a thoughtful approach, as are most people across the country, in validating these tools to make sure they’re safe and effective. We focus on organizing our approach around the FAVES principles: fair, appropriate, valid, effective and safe. And ensure that we understand how these tools work and function in the short and long term.
There’s a lot we’re still learning in the early innings, so to speak. And we investigate whether these tools work. And a few examples I mentioned. Are the concept notes in the shopping cart complete? Are there any keywords missing? Where can context be lost? Where could additional information be inserted that did not exist there initially? Hallucinations that people are aware of and want to be careful with.
We’re intentionally limiting adoption a bit and proceeding cautiously to ensure we deploy the solution safely and responsibly, which actually helps from an adoption perspective.
As we build the trust of the early adopter user community, they can become evangelists to help share the story of how these tools work, how they help, and where we might have opportunities. It is very helpful to have colleagues who can share this information. We’ve learned that word of mouth is an incredibly powerful tool to speed up or slow down adoption.
But there is certainly a lot of interest in these tools and excitement about their potential. Frankly, I don’t see a silver bullet to solve many of the challenges we face in the short term, but there is incredible potential in these tools, the generative tools, in the medium to long term. So it will be fun to work through that.
Editor’s Note: This is the eighth in a series of articles from top voices in healthcare IT discussing the use of artificial intelligence in healthcare. To read the first part, about Dr. John Halamka from the Mayo Clinic: Click here. To complete the second interview with Dr. To read Aalpen Patel at Geisinger, click here. To read the third, with Helen Waters of Meditech, click here. To read the fourth, with Epic’s Sumit Rana, click here. To read the fifth, with Dr. General Brigham’s Rebecca G. Mishuris, click here. To read the sixth, with Dr. Melek Somai of the Froedtert & Medical College of Wisconsin Health Network, click here. And to read the seventh, with Dr. Brian Hasselfeld of Johns Hopkins Medicine, click here.
Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki
Email him: bsiwicki@himss.org
Healthcare IT News is a HIMSS Media publication.