AWS' Allyson Fryhoff on how health systems are using AI to analyze complex data

Agentic AI has evolved past the buzzword stage, with healthcare organizations using these tools to analyze healthcare data, improve workflows and research how to develop therapeutics to fight diseases.

And Amazon Web Services is playing a key role in facilitating these use cases, according to Allyson Fryhoff, managing director of healthcare and life sciences at AWS.

For example, Rush University Medical Center in Chicago is using agentic AI to study population health.

"Rush has built a series of agentic AI capabilities for patient experience and population health, so that they can really look at what's happening within that greater Chicago area, and make sure they're providing those therapies, those specialists, that are going to make a difference for those patients in the Chicago area," Fryhoff said.

Meanwhile, Memorial Sloan Kettering Cancer Center in New York City is using Amazon Bio Discovery to develop the next generation of antibodies that could become a drug to fight cancer, according to Fryhoff.

"The AI could help in many ways to really optimize and speed up that process," said Nai-Kong Cheung, M.D., Ph.D., Enid A. Haupt chair in pediatric oncology at MSK Cancer Center, in an AWS case study.

Agentic AI has been a focus for health systems, and that's why tech companies have been introducing industry-specific agentic AI solutions. In March, AWS launched Amazon Connect Health, an agentic AI solution that integrates with EHRs and allows healthcare organizations to simplify administrative tasks, such as scheduling and clinical documentation.

At last month's AWS Summit in New York City, HealthTech Analytics sat down with Fryhoff to explore how healthcare organizations are using Connect Health as well as other agentic AI tools to research new drugs and manage complex data. She also discussed how AWS maintains AI guardrails with security and privacy controls.

Editor's note: This interview has been edited for clarity and brevity.

HEALTHTECH ANALYTICS: How are health systems like MSK using Amazon Bio Discovery, which was launched on April 14, to come up with therapeutics?

ALLYSON FRYHOFF: Amazon Bio Discovery provides a catalog of over 40 AI biology models and allows organizations to get to therapeutics faster. It was used to design over 300,000 novel antibodies that were then sent as part of the lab-in-the-loop process we have worked on for 100,000 top candidates. And that was done in weeks versus up to a year that it would have taken before. What we're incredibly excited about with MSK is that this set of antibodies was for a very rare disease, and so very often the time and effort is not put into these specific rare diseases, and now you can. You can get to those therapeutics so much faster and at much less cost, so they can be brought to market faster.

How does technology help get therapeutics to market faster?

An AI allows a researcher to test hundreds of thousands to a million permutations very quickly, and then to get to the candidates that might be most efficient for tackling that cancer. When you are looking for a cure for cancer, very often it's a very specific antibody that will break down that cancer. Especially when it's a rare disease that has not been worked on before, it takes a tremendous amount of effort.

How does AI make provider workflows smoother?

We have purpose-built agents that are part of Connect Health that help take away that administrative burden. I know you've probably heard it a million times, [that benefit] of giving more pajama time back to the doctors, but that's when everyone does their case notes. Doctors can take those notes [from patient visits] through ambient listening, providing that better patient care, that eye contact with the patient that we're all starting to miss because we're all looking at our devices, and then be able to have those sent for medical coding and have AI look at some of that medical coding, so then frankly, for the provider, they can get paid faster as well. So we're looking at that entire lifecycle of the patient and caregiver experience in the provider space.

How are providers using AI to analyze complex health data?

When we look across the spectrum [of] population health within a city, a country, a county, some of the larger systems are using [AI] across very large amounts of data, and they're also looking at that from a longitudinal perspective. So if you look at organizations around the world, like the NHS, that have data from the time someone was born to the time of their demise, they know so much about that particular person. How do we now put all of that together and look at what might be possible across a population? And so that's one way that data is being used for personalized and precision medicine, but also across populations -- and to find patterns to better detect those anomalies that doctors are looking for as well as to treat those diseases.

Maintaining HIPAA compliance is essential when feeding health data into AI models. So, how do you institute guardrails to protect the patient data?

We're seeing organizations using Amazon Bedrock to provide the guardrails that are so necessary within healthcare when you are utilizing AI in a responsible way. Bedrock allows you to create an enterprise-level responsible AI that has guardrails according to the policies that you've instituted as a provider, but then it allows you, within that, to utilize the models that different parts of your organization may need to do their job. So that continues to be a conversation that we have every day around how do we as a responsible health system provide responsible AI, but go fast within the guardrails that we may want to establish, and that's what Bedrock allows you to do.

I think that's something we're starting to hear a lot about as organizations look at some of these AI models out there that are more general-purpose. You really need to think about what those guardrails are. Where are you putting that PII information? And that's what Bedrock and some of our tools allow you to do is standardize your policies. Make sure that any model that you're using is conforming to those policies.

Speaking of PII, or personally identifiable information, to what extent is patient health data used to train AI models? How are they used, or are they not?

They are not. That's not what we do. We have a shared responsibility model for security and for your data, and so that sharing means that the data stays within your organization. That is critical to making sure that those guardrails can be enforced.

But you have to have your policies as well, and so we can provide those tools to allow you to create those policies, but it's shared. So you, as a provider, you must also have the wherewithal to be enforcing your policies. We can do it through technology, but you also have to enforce it through how you're training your teams.

How do you prevent AI tools from leaking sensitive health data? How do you maintain the security and privacy controls?

What's most important to understand is that privacy and security must be built in from the ground [up]. When you start, that responsible AI, responsible data, has to be part of how you think about your data as well, and we will help you with the tools that we provide, because it's something that is job zero at AWS. We think about it every single day. But we share that responsibility with you, so the tools can only do so much, and then together we make sure that your data is secure.

What's next for healthcare providers as far as how they will use AI via AWS going forward?

Ninety-seven percent of healthcare data goes unused because it is trapped in unstructured data. With this in mind, we are continuing to explore new ways to help healthcare organizations unlock the power of their data with AI, from clinical notes to multimodal data and even social determinants of health data, to arrive at actionable insights to improve outcomes.

For example, several of the features included in Amazon Connect Health are still in preview, which allows us to see where we can continue to make updates to further enhance the use of the service for our provider customers. In addition to healthcare-specific services and solutions, we are also continuing to expand our partner network to make it easier for providers to integrate their workloads on AWS with other organizations they already use.

Brian T. Horowitz started covering health IT news in 2010 and the tech beat overall in 1996.