Rasi Bhadramani/istock via Getty
How health systems can build an AI-ready data infrastructure
Cedars-Sinai's chief AI officer shares strategies health systems can employ to create a robust data infrastructure that supports AI implementation and accelerates adoption.
In the era of AI, data readiness is more crucial than ever. Data quality, completeness and diversity across populations are crucial for powering AI models that are rapidly being implemented into the clinical and administrative frameworks of the healthcare system.
Mouneer Odeh, Cedars-Sinai Health System's chief data and AI officer, emphasized this point in a recent interview, noting that AI technology is only as good as the data backing it.
"In many ways, the fuel for AI is the data -- and garbage in, garbage out," he said. "So, data quality is really, really important, and one of the risks when it comes to deploying AI is to make sure that we have confidence in the data."
Data readiness remains a challenge for many health systems. According to a 2024 survey, nearly half of healthcare leaders (46%) report that creating a data-driven culture is a significant barrier, followed by ensuring data quality and governance (44%) and integrating data across silos (42%).
However, there are steps that health systems can take to build and maintain a robust data infrastructure to support the rapid implementation of AI.
Develop data assessment frameworks
Healthcare produces a vast trove of data, encompassing everything from clinical to financial to operational information. While this data availability benefits the healthcare industry greatly, Odeh noted that health systems must create comprehensive frameworks to ensure the data is actionable and valuable.
"You really don't know if [the data is] good until you need it, and then you start to use it to create a new metric, a new report, and all of a sudden you realize, 'Hey, there's an issue here and it's all the way up in the workflow,' and we need to work with the application team and the operational folks to fix it," he said. "So that happens still even with AI."
Cedars-Sinai has established a responsible deployment framework to address this issue. Per the framework, Odeh's team collaborates with any department that wishes to utilize AI tools, ensuring that the quality of data being fed into the tool has been reviewed and that the AI tool's outcomes are reliable.
Odeh shared that one of the most important lessons his team learned through this process is that the metadata -- the information that describes and provides context to the data -- may not be comprehensive or standardized enough for an AI tool.
For instance, there are several different ways to describe when surgery might begin in an operating room, and depending on the particular use case, staff might not use the same one. AI tools might not be sophisticated enough to parse through these differences peppered through the metadata.
"So what we've learned is that we have to enrich the metadata to make it data-ready and AI-ready," he said.
Break down data silos
According to Odeh, healthcare stakeholders have employed various strategies to address the persistent issue of data silos in healthcare.
"Years ago, we said the solution to that was to get it all data into a single data warehouse, a single physical data warehouse," he said. "And now we've moved on to say, 'Hey, federated approaches and data fabrics are a good strategy to do that.'"
Computing power has also become increasingly important from an AI perspective, according to Odeh.
"When you take those two things together, you start to think, okay, cloud strategy becomes critical to our data and AI strategy," he explained.
This is why Cedars-Sinai is in the process of migrating to Microsoft Fabric, a cloud-based integrated data and AI platform. Odeh explained that the platform connects different data assets, bringing them together at the semantic layer. The organization also builds special purpose data marts -- databases built for a single department or business functions -- as needed to support specific needs systemwide.
Balance long-term investment with short-term needs
With AI advancing rapidly, businesses across sectors are expected to adopt the technology -- and quickly. However, health systems must strike a balance between the urgent need for AI implementation and longer-term investments in data infrastructure.
According to Odeh, health systems must be cautious about implementing too many one-off AI point solutions that are challenging to scale from a data perspective. Rather, they should develop a comprehensive data and AI strategy that informs their investment in AI tools in the short term.
"The need to deliver value quickly can be a stepping stone towards the bigger picture of what you're doing," Odeh said. "So that's why I do think having a strategy in place and saying, 'Okay, if we know that that's the strategy, let's take a little time and build, use this as a building block towards that strategy.' It doesn't always work that way. But in general, I find two-thirds of the time we're able to leverage our need to deliver short-term value as a stepping stone towards the longer-term picture."
It won't always be possible to avoid vendor-specific point solutions; however, by developing an overarching strategy to manage data assets that guides AI implementation, health systems can ensure that both short- and long-term needs are met.
AI is racing ahead, and health systems will need to prepare to meet the moment. Still, there is no way to consistently predict how the technology will evolve. Thus, in addition to enhancing data and other IT infrastructures, health system leaders must be ready to change course quickly when needed.
"I think we have to have some level of humility as we move forward, and a willingness to understand and scan where the market is going and where technology is headed over the next couple of years," Odeh said. "It's hard to look beyond a two-year horizon, to be honest. Sometimes it feels like even in two years, things are rapidly evolving. So, there's a certain level of humility to say, we don't exactly know how this is going to unfold…. I think we're going to be as agile as we can in being able to pivot to where the trends are going."
Anuja Vaidya has covered the healthcare industry since 2012. She currently covers the virtual healthcare landscape, including telehealth, remote patient monitoring and digital therapeutics.
