Healthy Data Podcast_ Danny Sama (Northwestern Medicine) & Jordan Cooper (InterSystems)-20250226_160107-Meeting Recording
February 26, 2025, 9:01PM
22m 36s
Jordan Cooper started transcription
Jordan Cooper 0:03
Emma, the VP and Chief Digital executive at Northwestern Medicine for those who don’t know, Northwestern medicine is a health system headquartered in Chicago, IL, with 2700 inpatient beds and 11,000 providers across 12 hospitals and other ancillary facilities.
Danny, thank you so much for joining us today.
Sama, Danny 0:22
Thank you for having me.
Jordan Cooper 0:23
So today’s topic that we’re gonna be discussing is fire as a gateway to Gen. AI. Now everybody as you know and as everyone who’s listening to this knows who’s in healthcare dealing with data.
The word is Gen. AI.
And that’s the marching orders on the tip of everyone’s tongue.
There’s a lot of questions about is this vaporware.
How does IT support our mission?
How does it improve clinical outcomes?
How do we integrate it into our workflow?
You know what should we do?
A lot of organizations are saying we know we need to do Gen. AI.
I, but what’s next?
So, northwestern, Danny, you have a story about how you’ve leveraged fire as a means of getting there. Would you care to elaborate?
Sama, Danny 1:05
Yeah, sure.
So we’ve been on kind of our digital transformation journey for several years now, probably five or six and very early when we started to look at how we wanted to design the platform. We landed on a kind of a fire first fire centric approach for for a couple.
Of reasons one is we really thought we wanted to leverage interoperability standards to, to organize and normalize the data.
There are a lot of of of value propositions to us for that, but really the the main reason was we really saw kind of the future for healthcare data was going to be about real time intelligence and and as the advent of Gen. AI Gen. AI has come.
Up that that that of course is going to be very important, being able to make inferences within workflow based on real time data I think is going to be the differentiator whether Gen. AI, particularly in the clinical setting.
Will work and we’ll have adoption.
Because we’ve seen kind of a traditional ML prior to kind of the advent of Gen. AI that the modeling itself isn’t really the hard part, it’s getting the data right and it’s integrating into workflow.
And so I think the same is true about Gen. AI, and we think fire can be an enabler to to to support that.
Jordan Cooper 2:16
So talking about workflows, I’d like to dive into a concrete workflow that’s in use at Northwestern.
I believe you’re building a cancer registry and a fire standards are being leveraged to enhance that functionality.
Would you care to talk about how fire particulars being used to for the cancer registry?
And then also, if there’s any Gen. AI component?
Sama, Danny 2:42
Yeah. So I think one of the first use cases where we had a third party ready willing and wanted to partner with us on leveraging fire was for our cancer.
So we’re a nationally certified Cancer Center and it’s part of that.
We have to submit data to certified cancer registries that eventually go to a centralized body, and as we were looking for a new vendor several years ago, one of our first questions as we would talk with.
Third parties was, you know, can you use fire?
We were starting to push our third parties to think about fire integration.
As kind of a first step versus you know HL 7 or batch files and the vendor we end up going with was was very open and wanted to because they saw the the opportunity as well that if they could figure this out, it makes integrations a lot easier.
For the future. So that was our first real test at kind of really using fire at scale.
So we do have a fire server that’s set up.
It’s leveraging the EHR.
It’s kind of the back end for the for the fire resources.
The vendor makes API calls to get at the patients that they.
Need for the registry and it produces lab results, medications, notes, et cetera.
And the Gen. AI component of that is because they’re able to get the notes with API calls. They’re able to apply LLM models and try to extract data elements kind of automatically.
That might be embedded in those notes.
Things like pathology notes, which are are really rich source of information for cancer registries, and that’s been a great productivity boon for for the organization and for the abstractors that do that work.
And again proves the the value proposition.
Now, it’s not really kind of the real time inference use case that we have in mind, but it was a great use case to kind of dip our toe in the water.
And again, really make it work for the third party because again, that’s, that is what fire is meant is for data exchange and and that was a successful use case.
Jordan Cooper 4:25
So I appreciate you diving into that.
Some of our listeners are kind of on the front end of the journey that you just described and they have business owners of various service lines at their healthcare delivery organization that are saying, OK, what?
Please tell me how would I capture the ROI.
What is the value proposition? You know, why would I invest in this third party to build this fire server when we don’t know?
That this will actually work and maybe it works at Northwestern Medicine but not at all.
Our organization, cuz we’re unique, special and different. So and I believe you even mentioned that sometimes third parties will claim to be able to be fire native. But then once you sign the contracts kind of promises may fall by the wayside when you try to rubber hits the.
Road. So how do our listeners who have not yet gone upon a journey, begun a journey that you’ve already started? How do they?
Approach business leaders.
In a way, is saying just trust me, there will be ROI.
How do you have that conversation?
Sama, Danny 5:30
Yeah, it’s a good question.
And you bring up some great points and I think the maturity of the ecosystem and and vendors and third parties with fire has been slow. And I would say five years ago when we talked to vendors, most were not ready. I would say these days like you said.
Most of these claim to be aware and and ready and want to try it.
But there are some that still struggle.
I think the ROI is a couple of things. So so one is from a from a pure kind of hard ROI.
So we we think there’s a hard oi for for our res.
This is our technical resources in the sense of today. If we need to do an integration with a third party, we either have to write our own HL 7 interfaces or have to produce our own batch extracts and that takes resources from our internal is teams to to.
Build out those feeds. In theory, you know using fire and APIs.
Kind of puts the onus on the the effort on the 3rd part, right?
That’s why we’re using the standard. They can look up the specs that the the standards bodies have defined.
They can make the API calls assuming all the security permissions.
Properly set. I think that’s another value proposition for business leaders is we believe in a modern API ecosystem is is just gonna be much better from a security perspective, from an auditing perspective provides a lot of tools and opportunities to just again make sure that we’re securing and.
Protecting the data with as as modern of technologies as possible and so I think those are the the two main like hard value propositions.
I think there’s another value proposition that’s a bit more.
Maybe.
Altruistic or public health standpoint is we just really believe that if fire adoption continues to grow, there’s a lot of value from a public health standpoint. You know we thought about during the pandemic. If if if fire was mature, that’d be very easy to actually know the EV.
The incidence of COVID at any given moment just by making API calls to all the hospitals, and so I think preparing for the next wave of public health emergencies or how we want to deal with measuring quality.
Institutions.
Kind of embracing entropy style new standards is valuable for the broader ecosystem. It’s valuable to us to try to get ahead of it so that we’re not kind of caught at the at the too late in the game trying to catch up and and get our data and.
Our infrastructure set up to to meet those demands.
Jordan Cooper 7:45
So you mentioned three kinds of of ROI 2 hard ROI in terms of reducing the burden on your internal technical resources because third parties will be be able to make their own API calls and you have a better a modern API ecosystems better for security and auditing you.
Mentioned the public health altruistics kind of component as well.
I’d like to focus on the second one. Security, last year in 2024, there were a number of data breaches.
Most notably, change healthcare.
This really impacted cash flow at a lot of organizations, turning thin margins of 1 to 2% in the black to really being operating in the red, not being able to pay contractors, even staffing issues.
A lot of difficulties facing healthcare because of of payment issues which came back to security last year.
So how does?
Fire and API specific.
Assist health care delivery organizations with doing user roles and security and provisioning. How will investing in fire be something that a Seeso would love to see?
Sama, Danny 9:02
Yeah. I mean, I think RC, so and I think most Cso’s will appreciate that the the tools, the tool belt that comes around with leveraging APIs, the encryption standards, the tokenization’s, the user permissions, all of that is is much easier to manage and more, more easy.
To keep a centralized and monitor what’s happening with all of those integrations through APIs, through other integration methods, particularly batch file transmission, SFTP file transmissions of of large sets of data.
Lots of opportunity for risk, right?
Who knows where that file is landing at the third party?
Who knows what the security permissions are on those files, you know? And it really just kind of don’t have the the control, you’re kind of trusting either third party is going to be responsible.
That’s what contractual agreements are meant to to do in HIPAA, certificationship certification, etcetera.
But I think that’s what’s led to a lot of the breaches is kind of some some lax security practices. And I think APIs kind of.
Minimize the risk because instead of start sending.
Huge batches of data.
We’re sending individual transactions for the API calls and again really tight monitoring. Really good encryption, really good security tokens, and so I think from a technology standpoint, the just the API is a much more modern system for transmitting data and and brings with it those additional security bar.
Jordan Cooper 10:20
I appreciate you delving into that, Danny.
Security is an issue front and center on many kind of CI OS stack leaders across the country.
But they’re also thinking about cloud.
Cloud is another priority that many organizations are shifting when they say, look, I want to focus more on the healthcare business, the delivery of care.
I want to get out of.
Kind of. The data center management piece.
Which, you know, disaster recovery, high availability. I think a lot of what they’re looking to shift if moving to a cloud would be, I wouldn’t wanna focus on security of data centers anymore.
Sama, Danny 11:01
Yep.
Jordan Cooper 11:04
So I believe that northwestern medicine is made an investment in fire for cloud migrations and rebuilding the enterprise data warehouse. Previously having had an on Prem custom data model.
Would you speak about fires role in your?
For cloud transformation journey.
Sama, Danny 11:21
Yeah. So we have chosen a public cloud provider as as kind of our cloud and we are. Our goal is to move not just the data warehouse, but a full digital platform.
Our goal is to allow that platform to enable analytics as we did with the data warehouse, but also custom application development, so custom apps and software and AI and machine learning.
Those are the three main goals we’re trying to enable with the platform and the benefits of the cloud is one is we need to be able to do data.
And ml at scale.
And particularly with fire, that’s that’s been the challenge. If the adoption of fire goes where we think it could and should go, it’s very likely that we could bring down our production, eh, Rs by making the amount of API calls to it.
So our intent is to actually build the fire server, the Enterprise fire server in the cloud, leverage the advantages of scalability and kind of cheap storage in the cloud to to do that and make the API calls hit that infrastructure rather than our production DHR.
And so with that, that also comes as you mentioned much better increased security, right?
The the public cloud providers investments in security.
Both physical technology, both people is is exponentially hundreds of 1000 times more than we could ever spend on on that.
And so yeah, you know, it’s still we’re putting trust and risk in a third party to manage our data and make sure that they keep it safe.
But I think with the amount of investments, places like Microsoft and Google and Amazon are making.
In security.
It’s it’s the risk is actually less from from a data security perspective by using the cloud and that’s that’s the assumptions we’ve made as we’ve cost it out and planned out our digital transformation.
Jordan Cooper 13:01
So I think a lot of our listeners now are saying great, you know I am on board with fire.
I’m on board with cloud.
I kind of get some of the business logic behind how I might get some ROI from fire and cloud, but I’d like you to tell me more about the real clinical and business use cases for ML as enabled with these fire AP is in the cloud, can you?
Tell us, how are you?
You know, improving quality of.
There. How are you reducing 30 day readmissions?
How are you addressing your electronic clinical quality measures and improving your scores to improve your payments?
Because we’re in a value based payment program.
Can you talk about some examples?
Sama, Danny 13:44
Yeah. Just first quickly on the last point, I think that was another reason to embrace fires as we fully expect the federal government is headed towards a world of digital quality measures in which they make fire API calls and calculate the measures themselves. So getting ahead of getting.
Our data ready for that I think gets to the the points you’re making about ensuring that we have an clear and accurate medical record to respond and reflect the quality of care we provide. But the real kind of clinical use case that we’re embarking on now and I.
Will say this has been a journey.
We’re still on the path to build out.
This fire server and and really get it live and operational with with real time production data. But we started planning out a use case for rebuilding the current ML model that’s been embedded in our EHR for Ed Triage use case. So patients come into the emergency department we.
Want to provide the triage staff a quick score risk score based on the factors and of data that are in the EHR at the moment, including some textual components of notes from prior visits as well as documentation.
The nurses might put in at that moment.
And so right now we have a model that’s running on Prem, but we think there’s opportunities to leverage things like Gen. AI and LLM and things like the the kind of modern foundation models that might exist and be available in the cloud to kind of improve the ACC.
And the usefulness of that score.
And so that’s our that’s our first use case. We’re planning to make inferences using fire API data.
And we’ll train the model in the cloud based on historical data, but then run the model and deploy it again in the cloud.
Outweigh would make the API call get all the data almost needed, run an inference and then provide that in real time to the triage staff.
Jordan Cooper 15:21
I wanna circle back kind of almost to an academic topic for a second. The third value proposition, the public health COVID API to figure out the national incidence. Not going to help the bottom line.
Northwestern, but maybe good for the whole country.
That’s the thing I’m going to ask my next question in so.
The.
OK.
So for example, ChatGPT is owned by open AI and Microsoft has a stake in that.
And they license out copilot and you can take the copilot instance on your own private server.
Keep your Phi private and then train your own models, for example, to create this quick risk score in the Ed at Northwestern.
But then you know, 2500 healthcare organizations may be doing the same thing.
There seems to be a lot of redundancy of effort.
Sama, Danny 16:14
Yeah.
Jordan Cooper 16:14
Potentially this could be a more efficient process of everyone that were to collaborate. Of course the risk would be.
HIPAA violations with Phi and potentially data getting out there and not used for patient care, but for profitability in some company.
So I’d like to just ask you for kind of a fun second, just to depart from our jobs for a moment.
What would?
Be a way of is there any way you can envision of collaborating with other healthcare delivery organizations so that not each organization needs to reinvent the wheel training essentially copilot on their own?
Sama, Danny 16:42
Yeah.
Jordan Cooper 16:50
Own data to get a use case as applicable everywhere.
Sama, Danny 16:54
Yeah. I mean, I think it’s the to answer that question is another proof point for the investments and kind of the adoption of fire.
The reason?
The challenges that we have with collaborating and kind of making models reusable is is the data right?
Data is not even if you’re on the same same EHR. One hospital’s data is going to look different than other hospitals, even on the same same platform, and so fire as kind of the, you know, as was intended as an interoperability standard, should provide kind of that.
That that playground of where we can kind of look at data the same across institutions now full transparency, I don’t think we’re there today, right. You still make a fiery pick call one institution make the same AP call another one’s and you might get slightly different levels.
Of completeness and things like that.
But if we as an as a help as an industry further adopted and and move towards that kind of vision, then you could see the ability for collaboration.
Hey, we are.
We all have the same our data in the same format.
Let’s let’s bring it together with the right, you know.
Privacy and security and consent from patients.
And build the model together or and this is where I think the cloud providers are really banking on. Let’s build a marketplace of models where we may have built a model that works. It’s using fire data.
Hey, let’s make it available in some sort of marketplace for other organizations to leverage and deploy into their organizations.
They don’t have to start from scratch again.
The key there is the data has to look the same and that’s why the embrace of fire is important.
Jordan Cooper 18:15
The marketplace, actually this comes kind of full circle to an earlier conversation, which is earlier part of this conversation, which is maybe instead of having a one to 2% margin as a healthcare delivery organization, you’re getting out of the data center business, but you’re getting into the.
Software development LLM. Business and you can license out the model that you create to other.
Organizations to buy and that can generate a new revenue stream and therefore demonstrate the profitability or ROI of investing in fire care to comment.
Sama, Danny 18:44
Yeah, absolutely. I think that is an aspiration for a lot of organizations. You know, to your point, every every health system is looking to diversify some revenue streams and looking at things like innovation as a place to do that. And I and absolutely I think I can see.
A world now that again, this is where I do think fire going to become so important.
It’s really hard to do that if you have to retrain the data on every organization.
They’re not even just retrain.
You actually have to extract the data.
Individually because the data’s not structured the same.
But if we can embrace a standard like fire, there’s an opportunity to eliminate that step and kind of allow the scale of kind of that marketplace to grow.
Quickly.
Jordan Cooper 19:25
So I have kind of one doozy of a follow up question. As we approach the end of this podcast episode.
We’ve covered fire, have covered cloud, we’ve covered a wide range of of topics within those two buckets.
You know, but everyone knows about GEICO garbage in, garbage out.
How what measures is northwestern medicine taking to ensure that your data is clean, that you have normalized data, aggregated data, deduplicated data that?
Sama, Danny 19:50
Yeah.
Jordan Cooper 19:54
You know as much as these, you know, embracing these larger models and the integrations. They’re only as good as the data that they’re trained on.
So how to ensure that you have clean data?
Sama, Danny 20:02
Yeah.
Yeah, it’s it’s a huge challenge.
Can’t pretend that we have the answer, but we are focusing on data quality as part of our strategy and kind of building it upfront as the data lands and the the data lake, how do we build the right kind of quality tests and scripts and compare that data?
To kind of the source of truth data to make sure we have completeness and accuracy.
I actually think this could be an opportunity for llms to really help us. Is llms have the opportunity to help with the normalization of data.
Right. If you look at individual hospitals, medication data, right, there’s there’s thousands of rows just for this, for one, medication for Tylenol, right?
Thousands of rolls of different variations of Tylenol and LLM might be really useful to say, hey, I’m looking for Tylenol. I I’m going to read through all those. I’m going to pick, find those thousand rows rather than having to individually map row to row. I think that could.
Be a really promising next step for how LL Ms. Enable kind of AI at scale on an old scale.
Something we’re starting to explore and have didn’t done it internally, but we haven’t.
Kind of operationalized it yet, but we’re looking forward to kind of taking that tactic as we as we move forward.
Jordan Cooper 21:14
Thank you, Danny.
I have no more questions for you, but I’d like to offer you an opportunity to speak to our audience.
You they’ve just been listening to this conversation about using.
Fire as a gateway to Gen.
AI.
Any final words of Wis?
Parting words of wisdom that you’d like to impart upon your peers across the country.
Sama, Danny 21:33
Yeah. I mean, I think the one thing is that traditionally fire is thought of as a information exchange standard, right?
And that’s what it’s built for, right?
API calls to exchange information.
I think growing and we’re seeing this from from big platforms like the cloud providers, the idea of fire for a persistence, how do we also persist the data in a data lake?
Build an analytics platform.
Build a platform by which we can model create models off of fire based data.
I think is the growing trend.
So if you aren’t, you know, well versed on fire.
And are interested.
I would definitely take it from both lenses. Think about it.
How how can fire be useful for to me for exchanging data with with other organizations other entities but also how to use it for kind of as an analytics data model and and hope that the industry buys into the vision as we do in that we see it.
Kind of grow in its adoption.
Jordan Cooper 22:23
Thank you, Danny.
That for our listeners has been Danny Sama, the VP and Chief Digital executive of Northwestern Medicine.
Danny, I appreciate your time.
Thank you for joining us.
Sama, Danny 22:33
Thank you for having me.
Jordan Cooper stopped transcription
