AI in Healthcare: Challenges, Ambitions, & Strategies

Dataiku Company Renata Halim

The Dallas roadshow of the Everyday AI Conferences in 2023 included a compelling conversation on the topic of scaling AI in the healthcare industry. Led by Kelci Miclaus, AI Solutions Director for Life Sciences at Dataiku, the talk featured industry leaders Manish Motiramani of Medtronic, Chaitanya Vempati of the Memorial Hermann Health System, and Scott Sacha of CoverMyMeds.

The Healthcare Industry: A Complicated Playing Field

With a complex network of players in this space, from hospitals to pharmaceutical companies, and medical technicians to patients, this high-pressure, saturated landscape depends on massive datasets from multiple providers, along with legally-protected information that carries hefty fines if mishandled. As machine learning (ML) and AI investments are made in this space, careful consideration must be made not only towards this information, but the people the information represents.

Screenshot 2024-04-30 at 11.03.56 AM

Chaitanya Vempati, Director of Data Science at Memorial Hermann Health System, kicked things off. “I’m in charge of predictive analytics, population health analytics, and consumer digital analytics. We represent a system with more than 14 hospitals, over 250 different points of care for our patients. It’s a huge population flowing through.”

Scott Sacha, Senior Director of Decision Science at CoverMyMeds, continued. “There’s a network between so many entities: the pharmacy and the payer, the doctor and the payer, the doctor and the patient. We have a bunch of connected networks to add value.”

Manish Motiramani, Director of Advanced Analytics, Global Operations and Supply Chain at Medtronic, added, “As head of AI and ML for global operations at Medtronic, we have 100,000 employees globally. About half work for global operations and supply chain. If you think of a life cycle of a device that’s implanted in you, like a diabetic pump or pacemaker, the journey starts at Medtronic. Then we work with a distribution partner, to get it over to a hospital, where it is implanted in a patient.”

→ Go Further: Watch the Everyday AI Dallas Session

Big Ambitions Bring Big Challenges

Miclaus then moved the conversation towards challenges. “I’d challenge anyone in this room to say they’re not affected by healthcare, or how AI is going to impact that. As we think about accepting and scaling AI, what are your biggest ambitions and your biggest challenges?”

Vempati responded, “We want to reduce costs for all of our patients. That’s our biggest motivation. We want to reduce the length of stay, reduce mortality, and enhance quality for our patients. Patient satisfaction is a big thing.” As a leader of a large association of hospitals, Vempati sees his organization as accountable for patients’ outcomes. “We have a wide range of AI and ML applications in digital engagement, reducing costs, and supply chains. Fundamentally, we’ve invested in the platform, so we want to operationalize six data science models by the end of the fiscal year. We’re starting to invest our time and efforts in identifying the application and executing on the strategy to achieve that.”

The situation is slightly different for CoverMyMeds, but Sacha brought up the importance of information security. “We have the same problems, but also additional ones like PHI, which is protected health information. We have to be very careful because we don’t want to release it to the public. We process around 20 billion plus transactions a year. With AI, we need to solve for how we protect PHI. We currently use ML for predictions on how patients might pay for medications, and for analyzing legal documentation like contracts, but we want to use it more and also be responsible stewards of the data.”

Motiramani’s approach is highly tactical. “We sum up everything we do in six words: alleviate pain, restore health, extend life. The patient is our major focus.” He then spoke about two major initiatives that Medtronic is taking on with AI and ML. “We try to achieve zero backorders. If the device is needed at a surgery, it should be available at the surgery. The second thing is operational efficiency. We are expected to do more with less.”

For all of these organizations, no matter the specific corner of the healthcare industry they occupy, the same challenges and opportunities are present. AI and ML are being utilized to save costs and improve efficiency, all while keeping patient information secure and optimizing patient outcomes.

The Healthcare Industry Sets Itself Apart by Its Data

With actual patient lives at stake every day, the healthcare industry is more than just a little bit different than other global industries. Vempati, whose background includes experience in the energy industry, again mentions personal health data.

Personal health data is not shareable. Data is a governed thing in healthcare.

-Chaitanya Vempati, Director of Data Science, Memorial Hermann Health System

Therein lies one of the biggest ironies in the healthcare industry. The problem isn’t that there’s not enough quality data to build and test models–it’s that the data by its very nature is siloed, protected, and hard to access. “In healthcare, real-time data isn’t easy to get,” Vempati continued. The difficulty is multiplied when attempting to embed AI in an organization using this data. “All roads in AI/ML lead to an EMR (electronic medical record). That system is key for operationalizing everything. It’s where the rubber meets the road for us.”

Sacha emphasized the critical nature of medical decisions being made with the help of AI. “You can use AI,” he began, “but if it’s wrong, it’s a lot different than if it’s wrong in other industries. If we get this wrong, it can lead to death, or other severe challenges. You have to be extra careful when you allow AI to make those decisions for your patients.”

Every decision we make has a human life on the other side. This is very rewarding, but also very challenging.

-Manish Motiramani, Director, Advanced Analytics, Global Operations & Supply Chain, Medtronic

Vempati emphasized the importance of getting things right in the healthcare industry, especially as it relates to competition. “If we look at the top 20 hospitals, it’s probably 3-5 deaths that differentiate them. We’re talking about accuracy. Mortality is a big, big deal.” It’s likely the biggest deal there is, so it’s important that AI and ML are not only implemented correctly, but also for the right reasons.

Investing in an AI-Driven Future

Miclaus mentioned a key challenge that appears once a healthcare organization encounters PHI. “How does this challenge where you invest in AI and the use cases you can tackle now, vs. those that might be out of reach because of the high level of trust you need?”

Sacha responded, “One area that is challenging is data rights. In my group alone I have 5 firewalls of people that use this data.” Leveraging AI with such sensitive information is not only difficult, it is nearly impossible. In an ideal world, this information would be interconnected and readily accessible, so that a patient’s care journey can be connected in a linear way, and care adjusted accordingly if necessary. In the current landscape, “this is nearly impossible, legally and structurally. It’s hard to figure out where to invest because of the areas that are blocking you. Instead we’re using the aggregate. I think that type of thing will continue.”

Data governance is the key. Without it, you’re looking for trouble.

-Scott Sacha, Senior Director – Decision Science, McKesson (CoverMyMeds)

From Strategy to Application

Miclaus is impressed with what Medtronic has done in the area of building a comprehensive data strategy that includes AI and ML. When asked about barriers that still exist, Motiramani said that they have doubled down on making decisions based on data. “Dataiku is a big part of that,” he says.

There are four pillars to analytics enablement: Dataset, toolset, skillset, and mindset.

-Manish Motiramani, Director, Advanced Analytics, Global Operations & Supply Chain, Medtronic

“The biggest of them is the last one.” He argued that especially as it regards AI/ML use cases, experimentation, testing, failing, and trying again are all parts of the same process. It becomes less about using solutions that are ready out of the box, but finding room to experiment with models that might not fit perfectly. He cited a very specific example.

“One example I can think of is market basket analysis that was born in the retail world. We had this data but we had no clue how to use it.” His team took a very unique and innovative approach, retrofitting a solution designed for retail supply chains, and adapting it for a healthcare use case. “We retrofit the algorithm from this Dataiku solution to plan our output. You can’t just stick with the algorithms that are used in your industry. There are things done in other areas that you can take and make your world better.”

AI and Patient Care Decisions

It only seems natural that risk aversion towards AI and ML will appear, especially from the patient side. Miclaus asked the group about the expectations of what a patient should know about AI potentially changing healthcare decisions.

Vempati said, “The biggest thing that’s driving adoption and acceptance are data literacy and AI literacy. Patients are actually being exposed to other things in their world that’s already AI driven, but physicians, administrators, and hospital staff might not be. We need to educate them. We go full on in data literacy and AI literacy.”

“People are scared of AI because they’re scared of losing their jobs–that it’ll replace them,” Sacha agreed. Organizations need to understand that AI isn’t intended to replace skilled individuals–it should help them do their jobs faster and better than before. Implementation should just be done sensitively, with appropriate care given to the opinions and wellbeing of the teams using the available AI tools.

Motiramani mentioned perhaps the key statistic that should move the needle for most healthcare organizations as they shift from theoretical notions of AI implementation to tactical integration.

This is the overwhelming opportunity that sits in front of every organization in the healthcare industry, making it the perfect time to develop, test, adapt, and iterate with AI.

Adapting and Scaling AI in Healthcare

Going forward, how does the healthcare industry learn to navigate these uncharted areas? Motiramani recommended diving in as soon as you’re able. “If you’re not on top of Generative AI yet, you should be — this is what is going to change our world.” Vempati agreed. “There is one fundamental thing that is absolutely true: Your health is determined by the zip code you’re in. I hope we can use AI to change that.” Sacha added, “embracing technology, especially when you have a disability, is the most important thing you can do.”

Miclaus closed the discussion by mentioning the slight but powerful difference between AI and IA. “I prefer IA — intelligent automation. I like this idea because it’s scaling. It’s allowing you to have augmented intelligence.”

Organizations globally understand the immense responsibility they have as they become stewards of healthcare information, especially as much of it is highly sensitive. But this is not a reason to step away from AI and ML techniques. Quite the opposite, in fact. Now is the time to dig into the challenges that can be solved with a robust AI and ML strategy. When this is done, perhaps we can respond to questions that have remained as huge challenges in the healthcare industry for years, and in doing so improve patient health, longevity, and quality of life.

With a clear and effective data strategy in place, Motiramani says, “Life gets better. Not worse.”

You May Also Like

How Will AI Shape the Future of Life Sciences Organizations?

Read More

Improving the Delivery of Therapies in the Healthcare Market

Read More

Enhancing Speed to Market in Life Sciences Operations

Read More

Solving the Ocean Plastic Pollution Problem With Data

Read More