MICHAEL BIRD
Aubrey, Aubrey, Aubrey, how you doing?

AUBREY LOVELL RESPONSE
I am well, how are you Michael?

MICHAEL BIRD
I'm very good, thank you. Now, I feel like I know the answer to this, but are you the sort of person who likes to prep things in advance, or are you more of a last-minute kind of person?
AUBREY LOVELL
you 100 % know the answer to this. I definitely am a planner. I absolutely like to do things in advance if I can. And I imagine that you are the same way.

MICHAEL BIRD
I'm a planner. I love a spreadsheet

AUBREY LOVELL
And you have kids, right? So, and I have a kid on the way, so I'm assuming, like, you don't have a lot of time to prep things, right?

MICHAEL BIRD
Yeah, firstly, congratulations. Secondly, it's a bit of a mixture, I think. I mean, you have to incessantly plan anything that you're going to do, but you have to also be aware that all of your plans are going to suddenly change, but, we obviously know that there are some jobs which require huge amounts of prep work to get off the ground, even if the job itself only takes a fraction of the time to complete

AUBREY LOVELL
100%. But, you know, this week we're exploring more of a slow burn story, right?

MICHAEL BIRD
Yeah, that's right. Although nowadays it's maybe less of a slow burn and more of a roaring inferno because this week we're exploring a very specific AI revolution, an overnight success that was decades in the making. I'm Michael Bird.

AUBREY LOVELL
I'm Aubrey Lovell and welcome to Technology Now from HPE.

MICHAEL BIRD
So Aubrey, we briefly touched on the use of AI in healthcare in older episodes, but we often gloss over quite an important part of that story because, pop quiz Aubrey, I know you love a pop quiz, what's the one thing that every single AI needs to run? What's one thing that AI absolutely needs?

AUBREY LOVELL
Hmm.

Data?

MICHAEL BIRD
Yeah, exactly. They need huge amounts of data to train and run AI models. And if you don't have the data, as we've talked about many times on past episodes, you won't have very good AI.

AUBREY LOVELL
And all this data has to come from somewhere right?
We have, and you know, all of this data has to come from somewhere, right?

MICHAEL BIRD
Yeah, And people have spent decades digitizing old medical data and records, which means that by now it can be used to train AI, which means that now it can be used to train AIs for use by doctors, but also patients. Now, a little later in the show, we'll be hearing from Derek B. Howard, the program manager for the HPE Digital Health Foundry program about the ways AI is being utilized in healthcare in unexpected ways.

AUBREY LOVELL
Ooh, I am intrigued. But before we hear from Derek, it's time to look backwards at one of the more classic uses of AI in healthcare because it is, of course, time for, technology then.

AUBREY LOVELL
Okay. Okay, so Bird, cast your mind back and tell me if the name MYCIN means anything to you. And it's okay if you don't remember.

MICHAEL BIRD
I have heard that but I have absolutely no idea what it is. It sounds like a drug. Is it a drug?

AUBREY LOVELL
Getting closer, getting closer. So, Mycin was one of the first medical AIs all the way back in 1972, and it was built by university researchers in California and was designed to diagnose and suggest treatments for blood infections. It was pretty good at it too with reports that it was about as reliable as specialist doctors.

And while MYCIN never caught on as a major player in the medical field, it was certainly not the last expert system to be used.

while other people were taking the proof of AI's usefulness and building machines to play chess and the game of Go, researchers in Massachusetts were building on the ideas Myson had laid down to create their own AI, which was designed to help with diagnosis at their general hospital. Now, development began in 1984, and when it was released two years later, the system could identify around 500 diseases.

. And not just the blood infections that mycin was restricted to. Today, the hospital still uses the same albeit updated system, which now can recognize over 2,500 diseases . That's pretty cool. And of course the AI bubble continued to expand. It took only a further five years for LA cardiologists to utilize AI to help identify patients at risk of repeated heart attacks . more and more data would be fed into medical AIs over the next few years to improve accuracy and diagnosis and even discover new symptoms and markers for diseases . However, this is all still based around AI doing the diagnosis itself. And I believe your chat with Derek covered some slightly different topics, right, Michael?

MICHAEL BIRD
. It did, yes, but just to set the scene a little bit, the first thing I wanted to know from Derek... was the current state of play and how AI is being used in healthcare today.

DEREK B. HOWARD
Well, the first thing we should look at when we're looking at AI and healthcare is that it's an overnight success that's been decades in the making. Healthcare now accounts for 30 % of the world's data according to Bessemer Ventures, but the sad fact is that up to 97% of the data that hospitals create aren't actually even utilized. So, it's a big opportunity, but we really need to leverage AI to gain more efficiency.

And one of the very first use cases I think of is clinical decision support. And that is when you look at like radiology and the number of images, pathology, detecting things based on that, then, you know, accurate diagnosis. That's one of the key use cases. Another use case is kind of the medical documentation and workflow to reduce the burnout. There's so much documentation. So automating a lot of that is really critical to making sure that we can keep doctors doing what they need to do.

MICHAEL BIRD
So when you say patient data, is that just like x-rays or like just… talk about the sort of data that gets created by hospitals.

DEREK B. HOWARD
Yeah, the interesting thing is that medical imaging is one of the most used diagnostic tools today and there's upwards of 600 million plus images that are created every year now. these images have got to be stored and oftentimes the way the medical system is, you could have these pockets of data based on different institutions, different hospitals, different places. And so it makes it very difficult to gain access and aggregate all that data.

MICHAEL BIRD
And this data is potentially being created in different hospitals, different networks, different places, and it's not all being put in one place.

DEREK B. HOWARD
Sadly, we don't have a massive healthcare data lake that kind of collects everybody's information, so it is very disparate.

MICHAEL BIRD
So how do you train a large language model for this sort of work? Because presumably this is really sensitive data and people aren't maybe… maybe they're a bit more nervous about their data being trained in AI models.

DEREK B. HOWARD
Yeah, no, it starts with obviously a large language model. A lot of folks, there's a lot of models out there that has basic healthcare information in it. And then the next step is really kind of the fine tuning. And that's where we take a lot of data and, you know, take out the personal information, right? So if you think about the HIPAA compliance and things like that, the personalization is removed and then aggregated so that you gain that kind of broader level of efficiency.

MICHAEL BIRD
So these tools are being used by medics, doctors, nurses, other healthcare professionals, or is the plan to replace them?

DEREK B. HOWARD
No, I mean, if you think about it, you know, when we went through COVID, there was a huge amount of burnout. We actually lost a lot of healthcare workers. And so we actually were in a staff shortage situation. We still can't even make it up. So we can't even bring people on fast enough. So really what this is about is making things more efficient, providing tools so that people can actually automate a lot of the things that can be automated and allow the human component to come in and really provide the specialized care where it's needed.

MICHAEL BIRD
Let doctors, nurses and healthcare professionals do the thing that they're good at and the stuff that they don't necessarily need to do but maybe as part of their job sort of automate that.

DEREK B. HOWARD
We recently started working with a healthcare provider in our digital health foundry program and their focus is on providing a large language model that allows the aggregation and pulling together a singular patient's records from multiple different data locations and it provides it very rapidly so that it allows faster time to diagnostic and allow doctors to maybe they have a full picture of a particular patient's life cycle across many different images and that allows them to diagnose much much faster and more accurately.

MICHAEL BIRD
So let's just dive into that. So what is the problem with healthcare data at the moment?

So it's in lots of different locations in lots of different formats and so as a patient if you're going from hospital A to hospital B you may not necessarily have the data from hospital A to take to hospital B so hospital B may not have some of your your sort of medical history so they're maybe not making the best decisions because they don't have the full picture.

DEREK B. HOWARD
That is the crux of the problem is really data fragmentation. It's really everywhere. obviously, privacy is a major concern. So how do you basically make sure that you protect folks' but still enable it to be captured and aggregated and provide a more comprehensive picture? And that's exactly what we're trying to accomplish.

MICHAEL BIRD
Presumably you could turn out to a hospital and they can do all of the tests and that's how they get a full medical picture, or is it actually sometimes historic data helps to make a fuller picture? Yeah.

DEREK B. HOWARD
Yeah, absolutely, where we are today is really took looking at the past and the present and so what we can do is look at past images present images do a very quick comparison pull in you know through this large language model all of the clinical notes all of the additional data that enriches the information around these images and make that faster more accurate precision personalized you know diagnostic situation.

I think where we're going right is in the future we'll be able to kind of predict things. And what we need to do to be safe and secure in that place is make sure we've got the right guardrails so that we're not making crazy recommendations. And we also have the human element, right? So once we have some sort of predictive idea is to draw in the doctor to provide the validation and recommendation.

MICHAEL BIRD
Presumably there's quite a lot of agentic in this if we're making sure there are some guardrails there, making sure the right processes and decisions are being taken.

DEREK B. HOWARD
Absolutely, and there's a lot of companies that are out there now that are deploying different agent models that tap into different large language models based on particular requirements, right? So it could be a cancer, it could be a particular body part. And these different sources of data can be very easily tapped. We're going to get to the point where it's so precision care that your own history will be able to be reviewed so that if someone prescribes you medication that you're allergic to, the system will be able to automatically pick up onto that and say, a minute, we understand based on this other set of data that this isn't the right treatment course.

MICHAEL BIRD
And this sort of system will work with maybe different data sets in different formats. I guess at the moment it's like, know, data cleansing is quite important.

DEREK B. HOWARD
Yeah, it quite is and the issue often times with healthcare data is that it's not necessarily representative. And so a lot of the data is skewing so you have to be careful of the unconscious bias. That's within the healthcare data, right? It's an interesting dynamic is that you know Heart disease is a huge killer of women yet. Most of the data is based on middle-aged white guys So you have to understand that to understand you can't just go based on blindly based on what the LLM says because you have to understand the information that's is drawing from and make sure that that data is representative of the person.

MICHAEL BIRD
So capturing all this data from lots of different people, I mean, this could be used for medical research.

DEREK B. HOWARD
Absolutely. That's a couple of the other amazing use cases are around drug discovery, specific genomic precision care based on that. So there's a lot of drug discovery. That's another use case that is being leveraged.

MICHAEL BIRD
So for our non-US listeners, could you just give a bit of an overview of how the US healthcare system works and why this problem of data in lots of different places and lots of different formats, maybe it's not unique to the US but it's very prevalent in the US

DEREK B. HOWARD
Yes, I mean we've got a very, you know, payer-provider you know interesting dynamic with the insurance companies and everything is very scattered right now and the model is very rapidly undergoing a change and it will not be sustainable for the future because a lot of the younger folks that are coming in are not buying into the insurance policy which historically has funded some of the elderly older folks who needed the care and so the younger folks are kind of have taking a more active health lifecycle in the own journey and so it's breaking the model down and so that's why I think we're on the cusp of whole new different business models being created in the healthcare space and it's needed. It's ripe for disruption.. And there are true visionary leaders who this is personal passion project to really revolutionize healthcare. because it is just such a significant opportunity.

MICHAEL BIRD
I mean and the problems you identified about not being enough staff, know, not being enough healthcare providers, that's worldwide. As you said, since the pandemic, that's worldwide. actually these sorts of tools are going to benefit, you know, medical professionals, humans around the world.

DEREK B. HOWARD
Yeah, and a lot of the operational efficiency is just, I think I was looking at the different use cases and almost for every one, I'm aware of a new AI ISV healthcare application that is investing to radically transform the efficiency. We've got another ISV that's actually monitoring the operating room to ensure absolute efficiency in terms of the cycle time because that's one of their most hospitals' expensive assets is that operating room generates the most profit.

So how do they make sure that that's utilized to the maximum effectiveness in terms of aligning the right doctor to the right hospital, you know, the right operating room and ensuring that everything is moving very smoothly. absolutely efficiency is going to be a huge key gain that we get out of the AI being deployed so that we can essentially make greater efficiency in terms of decision and more rapid and accurate care.

MICHAEL BIRD
How do you convince people whose data is potentially going to be used in these models to actually willingly give their data? Because presumably they may feel a bit uncomfortable about it. Are there guardrails in place? Is their data anonymized?

DEREK B. HOWARD
it's interesting because a lot of folks are doing this today, and they may not necessarily even be conscious of it, but your smartphone, a lot of folks are into the rings. There's a lot of ways that that data is being collected today, and it's being anonymized. And I think a lot of the larger consumer companies unwittingly are becoming some of largest healthcare companies in terms of their visibility of this information. And so I think over time we're going to see is a continued focus in terms of security and privacy making sure that the data is protected. Right now, healthcare data is one of the most sought after information on the dark web, so it's very important that we continue to provide the security around the data and ensure that things are anonymized and that they're protected.
And that's what a lot of companies are striving to do because they don't want to be discriminated against the data that they have. And so it's very important that we continue to make sure that the compliance regulations are met, particularly in healthcare data.

MICHAEL BIRD
so you have this agentic AI that's sifted through the records at all the hospitals you've been to, it's aggregated the data. What happens next? Like, it flag it if it thinks there's a problem with you or is it only, you know, when you come into the hospital something happens?

DEREK B. HOWARD
Well there was a very specific use case where someone had an operation several weeks ago and they come in because they're complaining of a symptom. Very quickly the doctors need to understand is this something that was based off of the surgery? Is this a normal result or is this abnormal? Do we need to take some sort of action? And the fact that the system can very quickly through the LLMs aggregate all of the data allows them to very quickly understand is this something that should be rapidly moved on and that's part of the value proposition is the instantaneous action of pulling all the data along with the clinical notes to make that quick decision because sometimes this can be a matter of life and death whether you decide do I need to operate on this person or send them home.

MICHAEL BIRD
so do you think we'll be seeing more of AI in healthcare in the future or do think this is it?

DEREK B. HOWARD
Oh absolutely. The use cases are just endless. know even worse even like if you think about it like robotics. You know I don't know if you see the surgical augmentation now allows for more delicate procedures that are less invasive that reduces the healing time. It's just amazing right. So if you think at the automation, the workflows, there's no area of healthcare that will not be impacted by AI.

AUBREY LOVELL
I just, you know, I think we hear this story a lot and how much of a change that you're seeing with AI and how it can make a difference, right? I mean, that stat up front, talking about 97 % of, you know, hospital data is not even utilized is incredible. can you imagine if it was and the opportunities that that could, input for, you know, breakthrough innovations and cures and, you know, healthcare?
MICHAEL BIRD
both you and I have family members who work in healthcarewe briefly touched on the pandemic and like the impact that that had on our family members. mean, it was massive. And actually people getting burnt out and there not being enough staff and then maybe not being enough information. Actually, if we can utilize some, if you can utilize this data better if there is a way that we can like leave the medical professionals to get on with what they're good at. I see that as a good thing.

AUBREY LOVELL
If AI can help with that, it helps us free us up to do what we need to do to save lives in this case, you know, focus on your job, essentially innovate. And then you have the also component and the nuance with healthcare, especially, but it's also in banking as well, right? you know, having that data secure and making sure that it's not out in the world because it is things that are very private that you don't want people to see, like your sensitive data or your medical records, things like that.

MICHAEL BIRD
I love the idea of being able to pull in all of my historic data plus new tests if I had an illness or something, plus aggregated with data from other people similar to me to be able to say, yes, well, in your case, this might be this or this might be this or we've treated somebody with this.

AUBREY LOVELL
Absolutely.

MICHAEL BIRD
so Aubrey, while AI in healthcare is nothing new, as we've talked about on previous episodes, the speed of change and disruption in the space has meant that things are changing quite dramatically. So I wanted to finish this episode with Derek's thoughts on where the world of AI in healthcare is heading in the next few years.

DEREK B. HOWARD
There is a lot of innovation in the startup space, and what we're seeing is that there's an opportunity for kind of early stage AI healthcare companies to partner with the larger institutions, right? And the larger organizations just cannot move fast enough. And so what I see over the next several years here is going to be a real partnership with larger healthcare organizations partnering with these new, innovative, fast companies that are creating these things and then expanding them and putting not only in pilot models but getting them to vast production and so I think we're on this renaissance where we're going to have great opportunities with these small companies, Davids working with the Goliaths and overcoming this healthcare challenge.

AUBREY LOVELL
Okay that brings us to the end of Technology Now for this week.

Thank you to our guest, Derek B. Howard,

And of course, to our listeners.

Thank you so much for joining us.

MICHAEL BIRD
If you've enjoyed this episode, please do let us know, rate and review us wherever you listen to episodes. And if you want to get in contact with us, do make sure you send us an email to technologynowathp.com. Subject line, I need some data, stat. I need 20 ccs of data, stat. And don't forget to subscribe so you can listen first every week. How's that joke? Is that terrible?

AUBREY LOVELL
I liked it.

MICHAEL BIRD
Technology Now is hosted by Aubrey Lovell and myself, Michael Bird
This episode was produced by Harry Lampert and Izzie Clarke with production support from Alysha Kempson-Taylor, Allison Gaito, Beckie Bird, Alissa Mitry and Renee Edwards.

AUBREY LOVELL
Our social editorial team is Rebecca Wissinger, Judy-Anne Goldman and Jacqueline Green and our social media designers are Alejandra Garcia, and Ambar Maldonado.

MICHAEL BIRD
Technology Now is a Fresh Air Production for Hewlett Packard Enterprise.

(and) we’ll see you next week. Cheers!