For years, we’ve heard about the enormous potential of artificial intelligence in healthcare — not as a replacement for physicians, but a way to improve outcomes by extracting meaning from data. AI, according to Alex Madama, DO, Healthcare and Life Sciences Chief Technologist at HPE, can help “amplify the human capabilities and turn this exponentially growing data into insight, action, and value.”
The question, it would seem, is how this can be achieved. That, however, is just one of many questions which needs to be addressed in order to unlock AI’s potential. During a recent discussion, Madama explored this issue along with co-panelists Zafar Chaudry, MD (CIO, Seattle Children’s Hospital) Aalpen Patel, MD (Medical Director for Artificial Intelligence, Geisinger & Steele Institute for Health Innovation) and Zachi Atti, Co-Director of Artificial Intelligence in Cardiology, Mayo Clinic.
Although there is no secret sauce when it comes to building a successful AI program, there are steps that can point organizations in the right direction. It starts by having some critical discussions and outlining specific objectives, said Patel. The first is in determining how a particular model can positively affect patient care, while keeping expenses in check. Then, “if we’re successful in getting the data or the insights that we need, how are we going to change patient care? If you don’t have that answer, you have to go back and find it before we go ahead with modeling.”
These are big questions that can’t all be tackled at once, according to the panelists. Instead, they advised breaking it into chunks by taking the following steps.
- Involve the right stakeholders. To Chaudry, this means not just involving the right stakeholders when creating AI models, but ensuring those stakeholders understand the output of what is being created. “We learned that we needed a set of analysts in our team that can help explain what all of this means,” he noted. In other words, “if we were to move the needle, what would it mean for that particular individual or that particular patient cohort, or for our business in general? And so it’s a combination of the technology, the people and how you change the processes.”
- Skills, then technology. At Seattle Children’s, the strategy has been to bring data scientists who specialize in building predictive models, and have them work side-by-side with physicians and nurses to determine what models are built. “We’ve embedded AI tools and predictive models in all facets of our operational business so that we can predict when we’re going to have census spikes, when we have to cancel surgeries, and what will be the impact of Covid-19 and flu, all in real time,” he noted. “We use those data scientists to model what the physicians want us to do, and then we present those dashboards in real time. We’ve embraced the approach of skills first, then technology.”
- Align engineers and physicians. According to Atti, failing to do so can result in AI models that are only good on paper. “You need to translate the needs of the patients to the physicians, which requires a lot of collaboration between the physician and the engineer,” he noted. “It’s helping them ask the right questions to guide the development of these models.”
- Understand business needs. Taking the time to understand the needs of the business, whether it’s the physician or patient, is critical, according to Chaudry. “If you don’t do that, you get into a model where technologists are building things that are of no use in the real world.” On the other hand, by collaborating with the business, IT leaders can help ensure AI initiatives provide “real value” to users.
AI in Action
The challenge is that with so much data available — Geisinger, for however, has three petabytes — it can be difficult to figure out where to start. Because AI offers so much potential, there’s often a tendency to want to solve every problem, when in fact, the key is to identify a specific area, according to Patel.
At Geisinger, one of those areas was intracranial hemorrhage detection, a condition that is quickly spotted in ED patients, but can go unnoticed in the outpatient setting. By applying intelligence, the team was able to equip CT scanners to uncover episodes of intracranial bleeding and prioritize them, which in turn helped providers act quicker. As a result, they’ve been able to reduce patient turnaround time for the positive bleeds by about 96 percent, noted Patel.
For Mayo Clinic’s cardiology department, one of the key use cases has been in detecting diseases using signals that humans haven’t been able to interpret, according to Atti. Specifically, they created a model that leverages an EKG to predict possible events in a way that is quicker and more cost-effective.
And while these examples are certainly noteworthy, it’s only a sampling. There are, in fact, numerous uses of AI that extend beyond clinical outcomes, the panelists noted.
At Seattle Children’s, for example, AI tools are being leveraged to predict spikes of Covid-19, which can help calculate patient census by unit and more effectively plan for surgery cancelations. But that’s just part of it. The organization is also utilizing AI to extract information from building management systems and sensors and predict things like when the next airflow handler or device will fail, and provide notification for early maintenance.
What this does, according to Madama, is enable organizations to do preventative maintenance as opposed to having an unplanned downtime. “They’re using AI and predictive modeling to see when a machine is going to fail and divert to a different manufacturing line, for example, so that they could do maintenance on one while the other one continues running,” he said. “That’s a space we believe is going to explode in the next few years.”
For that to happen, however, leaders need to get past another critical roadblock: finances, which have become particularly scarce in 2020. For CIOs, the question has become, how do you prioritize innovation and leveraging AI to transform care delivery versus keeping the lights on?
Perhaps it’s time to look at it another way, according to Chaudry. “Maybe we need to spend more time employing these AI tools and algorithms to improve the cost of delivering health care. We’re getting to a point where we have to find that balance between fostering innovation and keeping our business running.”
Part of that is in determining which projects are truly mission-critical, and that is where governance comes in, noted Patel, whose organization created a steering committee with representation from various stakeholders, including the chief data informatics officer, CMO, and CFO. “That gives us guidance from a system perspective, which direction we want to go in, and which direction we don’t. That’s been really helpful.”
Other areas that also require focus include growing the data scientist workforce, providing better training for those inputting the data, improving the data cleansing process, and shifting away from the silo mentality, said Chaudry. During the discussion, he advocated for a standardization collaboration model that enables easier access to intelligence and more data-sharing.
“The vendors that help us create data need to inter-operate data more clearly and adhere to standards,” he said. However, that’s one just facet. “It’s a combination of things that we need to look at if we are really to move this forward, otherwise we will be siloed for the next five to 10 years.”
To view the archive of this webinar – AI in Healthcare: An Exploration of Practical Perspectives & Use Cases (Sponsored by HPE) – please click here.