Integrated data, generative AI, and a new era of health breakthroughs

Um, I will have a cup of coffee glued to my hand because I need coffee today. So thank you for bearing with me.

As Adrian said, my name is Valerie Delva. I lead data AI/ML strategy and solutions for healthcare and life sciences. And I'm excited about our discussion. It's a very timely topic. And as you think about it, it's really on how do you build an integrated data strategy and think about some of the fundamental requirements in an end to end strategy to unlock ML and generative AI use cases.

So all those interesting things from what are the trends driving the need for an integrated data strategy today? Um, we know that a key theme in things that we're hearing is that data is your differentiator for generative AI. So I wanna go through exactly what does that include? What does that mean?

First, I'm gonna talk about challenges and opportunities in the space and then, and the key components that go into building a data strategy. We'll also discuss the AWS health portfolio for data. So really thinking about our fit for purpose, services and products that we have around data AI/ML.

Um, also looking at what are some of the partner offerings and solutions that we have that support that as well.

Um, before we start though, I'd like to take a step back and have everyone think about this question that you see on the slide here. Um, keep it in mind as we go throughout this discussion, how can we together scale and accelerate the future of health? And I'm defining that as one where genomics, machine learning AI can drive more predictive, preventative, personalized and participatory medicine. For all.

I very much believe that a lot of the things that we'll talk about today will make this progress not in 20 years or 10 years. But right now, we have a real opportunity to think a bit differently about how do we define health and wellness and how do we leverage the power of health and a uh data and AI to do that.

So let's dive in. So I want to start with some trends over the past several years, we've noticed that there are 44 or so trends that have created both new challenges and new opportunities for how healthcare and life sciences organizations in terms of how you create store and use data.

Um and until quite recently, you couldn't easily address many of these challenges or leverage these opportunities with the technologies available today. So the first one and this may not be, be too different to some folks but it's a familiar one. Many of you have been asked to greatly speed up the discovery and development of new therapies. It now takes almost a decade, about nine years and $3 billion to develop a new therapy. In addition to that, think about the tremendous patient impact the time and also how economically unsustainable that is over time. That's how we're developing medicines. But all of this means if you think about it, you need to make better use of the data to speed up early research, clinical development and clinical trials and you have to prove efficacy to different bodies such as payers including through the use of real world data. So the first challenge that you see here, it's really thinking about how do you find and use data faster while at the same time reducing costs.

So the second one, the volume and complexity of data are both exploding. So for example, if you're in life sciences, your labs can now generate multiple terabytes a day from newer essays like cell painting, RNA sequencing and cryo EM yet. And this is a step that I think we've all heard quite a few times. 97% of the data generated today goes unused because it's not stored in a manner that makes it easy to find use or make actionable. Plus scaling up storage for these data types would be hard enough if they were all the same type of data, but they're not, they're often in different modalities as you see here on the right.

Yeah. So now as we continue with these observations, another trend that we're seeing is that every year, a growing percentage of new therapies that reach the market originated in some form of collaboration between a bio pharma company and a third party, such as a start up a university lab or population research partner. For example, more than 50% of drugs that are launched today originate from such partnerships. Think about that more than half.

So challenge number three is how do you more efficiently discover what data your collaborators have that you're allowed to access and then make appropriate use of that data. So this all means you need powerful tools for data cataloging uh data governance and data sharing and that's an area i'm going to spend a bit more time on in a little bit.

So the fourth one and Adrian when he introduced me, call this out, do you remember the fourth challenge? Uh J of AI? It's an option 280. It's a challenge. It's a buzzword. It's what we're all talking and thinking about. And for uh a quick definition, think of generative AI for those who are unfamiliar as a type of artificial intelligence that allows you to create new content and ideas. Uh whether that's conversations, stories, images, videos, even music, the promise of generative AI in healthcare and life sciences is vast.

Um it has the potential to accelerate innovation. It can increase efficiencies across the healthcare and life sciences value chain. However, in addition to uh efficiently managing large volumes of data for such models, they are important considerations that all organizations, especially if you're in healthcare and life sciences need to consider.

So with patient data and proprietary data, you must address keeping the data private and confidential and use in a secure manner and ensuring only the right access and controls to it. So it's also say that generative AI and the opportunities there only intensify some of the challenges that I've talked through so far.

So I do want to take maybe a couple of more minutes to talk about. Well, what do you need to think about when it comes to generative AI plus data?

The first thing to keep top of mind is and I'm gonna put my coffee down because I get excited about this part. Data is the difference between a generic generative AI application and one that knows your business, your patience, uh your workforce deeply.

So every company at some point can have access to the same foundation models. But companies that will be successful in building generative applications with real business value are those that are doing so with their own valuable data and assets. And the companies who have not yet found ways to effectively harmonize and provide ready access to their data will be unable to for example, fine tune generative AI to unlock uh the transformative potential it has in healthcare and life sciences.

Um this is incredibly relevant for health care and in life sciences. When you think about um where data and the use cases are so specialized, the amount and types of data are growing exponentially. If you think of just genome data alone, it's predicted and expected to reach 40 exabytes, not tera exabytes in the next 10 years.

The other point I want to make is that generative AI does not promise instant insights. People think of i go from data to a model to insights that will work. Um the key to delivering healthcare and life sciences use cases with impact is quality data.

So if you're trying to get to insights, it matters the type of data you're bringing in um and how relevant it is to your use case. And i i would say that this is the number one challenge for many organizations in realizing the potential of generative AI data systems are often sprawling siloed complex uh diverse data uh sets spread across different data lakes, data warehouses, uh cloud databases, a applications it devices and on prem systems as well.

So to get meaningful insights, you really have to be clear about what is your data strategy. What is your infrastructure strategy and always keeping top of mind that data is your differentiator.

So now that I've highlighted here are the trends and the opportunities in the space and how you should be thinking about your data to unlock. Uh ml and ja i use cases what goes into developing and building an integrated data strategy.

So to start an integrated data strategy is not just about breaking down the internal data silos, but doing so in a manner that puts in place secure government access. So only those who should have access, have access and then identifying which use cases are critical to driving your business outcomes and the types of data that are needed to execute on those use cases in an impactful way.

Uh I say this because as we know, all data are not created equally, so you always have to be thinking about that. So perhaps you're a bio pharma company, you're collaborating on clinical research with an academic medical center and you need to augment your own data with insights hidden in patient medical records and research data sets.

So your data strategy, for example, has to include seamless integration of rural data sets, for example, but in a secure privacy preserving manner um that's purpose built for third party data collaborations of this type. In addition to a comprehensive range of analytics, ml and generative AI capabilities to allow you to go from data to insights generation.

So i set up challenges, set up opportunities and said here are all the things you must consider. If you're going to build a data strategy, how can aws help you do this, right? And if you've heard me speak on this topic before, you know, something that we always uh note and mention is we were reflective ourselves. And to be honest, we realized that building an end to end data strategy that unlocks use cases like this. Um you can't continue to use general purpose cloud services and it wasn't enough to get the job done specifically storing, analyzing, cataloging, sharing, multimodal health and life sciences data requires purpose built solutions and services and not just from aws but also our partners and other customers that we work with.

So i'm going to cover a few of those today. Um we have many other sessions that re invent where we'll dive deeper into them. But i wanna highlight some of our core services but also talk about some of the newer features that we have upcoming.

So let's start with uh storing and analyzing multimodal data. Um we've developed a range of offerings that you'll see on the screen starting here with healthlake uh health link um allows you to uh store analyze text based clinical data such as electronic medical records and claims data. A key feature that i do want to call out is it automatically converts unstructured clinical data, like physician notes, for example, into structure data and provides a chronological view of individual or patient population health data.

Additionally, our next service that you see here, our health imaging service can reduce uh imaging storage costs by up to 40% while also allowing multiple users to access the same copy of that image at the same time with sub second image retrieval.

Now the third pillar uh healthomics, you can transform genomic transcript omic and other omics data at scale to generate insights, improve health and advanced scientific discoveries.

Um and perhaps the most important of all is all of these services can make their data available in uh for analytics and machine learning in a matter of minutes. So not just as separate data types, but in a multimodal manner.

Finally, on the right, the thing that i'll call out is health bribe, a newer service that we just announced hipaa eligibility for less than 10 days ago. Uh it, it empowers health care software vendors to automatically generate transcripts, summarize notes, clinical insights by analyzing patient clinician conversations um in their clinical applications.

So now let's talk about data collaborations and specifically cataloging linking and accessing data. So across all teams in research and clinical development, for example, users often start with three questions and i get these three questions all the time. Who is the data that i need? Am i allowed to access it? And if all those things are true, is it in a format or stored in a manner that makes it useful?

So to start and what you see here, i'm excited to share that. Just last week, we announced a new service uh aws entity resolution as a hi a eligible service. Um it allows you to easily match link and enhance related records across multiple applications. So multiple applications systems, data stores using a flexible and configurable workflow that takes only minutes to set up.

We also often hear from teams in research and development and clinical organizations that they need access to rural data, whether that's the study biomarkers or disease pathways or design a clinical trial. Um but many of our customers spend hundreds of millions of dollars per year licensing real world data.

Um and what we also hear from them is that to build a single cohort, they told us it can take off in many weeks from the time someone decides what cohort they need to then find sample and license the data. And then once it's in house to convert the data to the format or data model, their team uses, we're in a mission to simplify this. How do we take this from weeks to days to even hours? And that's why we create a data exchange that you see here.

Um or how we refer to ad x now while ad x is helping more and more customers every day. Uh researchers and health care organizations sometimes do not want to or are not able to put their data in exchange. So we announced clean rooms to address exactly those needs with clean rooms. You can securely analyze and collaborate without sharing or revealing your underlying data.

And besides clean rooms, uh many of my conversations, and i say this because even more recently, they start off with, you're talking about accessing data in other organizations. How do i even unlock data within my own organization? I get a sigh, i get head scratches. And as most of you already know, uh finding out what data your organization already has and only allowing appropriate access to that data can be incredibly hard.

And what i'll add to that is that the reality is there are several great data catalogs that exist today and many on aws. However, meeting health care and life science customers and use cases are not sufficiently addressed with general purpose catalog services. Here is the remaining transcript formatted for better readability:

So that's why we create a data zone that you see here, customers can leverage datazone to build an enterprise data ecosystem like a data mesh platform with federated governance where multiple personas across domains like R&D and commercial can collaborate at a project level.

Couple of more slides, I'm almost there. Um what I'll call up here and I won't go through all of them individually for the sake of time is to say we also have a suite of purpose built uh generative AI and ML services and solutions that we offer.

I will call out the one on the left because I often get questions about it. So, so Amazon Bedrock is a fully managed service that offers your choice of high performing foundation models from leading AI companies. So uh AI21 Labs, Anthropic, Cohere, Meta, Stability AI and also our own Amazon models as well.

So I went through quite a few different solutions and services and I often get the question of how does it all come together in an integrated strategy? And how do we pull it together? And how do we think about it?

I'll say what you see here on the slide is it means going from data to action to actionable insights. It means that how do you ensure that you can find access, collaborate and perform multimodal analysis in a way that is secure, makes sense and is aligned to a use case and those use cases could be any of the ones you see on the right uh detecting diseases earlier designing battery clinical trials or even developing medicines faster.

So I often like to close on uh maybe not this slide. But the next one to bring back to where I started, there are many challenges when we think of the data AI landscape and the opportunities that we have. But I think some of the things that we've highlighted today is we have an opportunity to think about impactful solutions to drive outcomes for patients.

So think about all the work that you're doing today across precision medicine diagnostics, targeted drug discovery. We're making real progress. How do we ensure that not just keeping patients and researchers and clinicians and caregivers at the center. But we're fully leveraging the power of data to do that.

And I put this slide in closing, um we're here to help. So whether the question is, how do you get started if you need peer level executives, sparring partner to think through your data strategy and what you should put in place.

Um if you're thinking about building a data vision um and executing against that strategy, or even you know what we need to start migrating and modernizing and taking that first step. So wherever you are in your journey, we have teams and solutions to help.

So with that, I will say thank you.