Today, tools of artificial intelligence are primarily in the hands of data scientists and developers. However, the era of AI exclusivity in healthcare is waning. As advances in technology make designing and using AI systems less complex and more citizen-friendly, clinicians and hospital administrators are gaining new and increasing opportunities to engage with data analytics.
This content is sponsored by Microsoft.
Becker's Hospital Review caught up with Elena Bonfiglioli, Microsoft's senior director of health and life sciences for Europe, the Middle East and Africa, and Tom Lawry, Microsoft's director of worldwide health, about opportunities and challenges for artificial intelligence in healthcare over the next three to five years.
Note: Responses are lightly edited for brevity and clarity.
Question: First off, what does it mean to democratize artificial intelligence in healthcare, and what does that process look like?
Elena Bonfiglioli: Democratizing AI means empowering every person in an organization with systems of intelligence to provide better care for patients, better access to the right data at the right time for the right people, and to overall provide a better healthcare experience and better outcomes.
We talk about democratization of AI because we believe AI solutions should not be limited to a few key players, but the underlying technology should be integrated and infused into every application and service — into the very infrastructure of healthcare — and made available to every provider. This is really an important philosophical and technological approach to health reform. That is the way in which we can enable everybody to change how they interact with the surrounding environment as that environment increasingly becomes an ambient computing environment, which we call the intelligent edge. AI solutions will not replace clinicians, but they can augment their capabilities to make them more productive and improve patient care.
Tom Lawry: AI is used to automate and make decision-making better across clinical and operational processes. Applying that definition to healthcare, democratizing AI is about making the power of AI available to everyone in the care process. Democratization is moving from this concept of using intelligence to improve patient care to actually living and breathing AI by putting it in the hands of anyone and everyone in the organization.
Q: What is required to truly make AI accessible and meaningful to everyone in the healthcare system, beyond data scientists? What are some things Microsoft partners are already doing with AI in healthcare?
EB: In short, three elements are required to democratize AI: necessary skills, governance and an interoperable ecosystem. First, we need to democratize skills related to AI, including what it takes to code and design algorithms. If these skills are not pervasive, then we don't have enough developers coding for these solutions. Also, we need to consider what skills are necessary in health professionals who need to be able to use these solutions every day. The experience between the machine and human needs to be simplified and more intuitive and user-friendly.
Second is governance. We need to be sure AI are understood and trusted and that we make the governance of those guiding algorithms ethical, open, transparent, and have no risk of harming humans. Governance makes sure the solutions we trust and the solutions we know are the solutions we use.
Finally, we need to ensure these systems are fundamentally interoperable, meaning these machines and systems can connect irrespective of vendor. If we are to infuse intelligence into every agent, application, service and infrastructure, we need to have a need to have IT environment developing health solutions that can work together. It cannot be one company doing it all; it needs to be a new AI economy and set of services.
Many of Microsoft's partners are using AI in interesting ways to augment physician's skills and work alongside them, rather than replace them. For example, AI tools help delineate tumors in shorter time and with highest precision that physicians alone cannot. Microsoft researchers in Cambridge developed an AI solution called InnerEye that creates a 3-D model of a tumor from a CT scan, which helps radiation oncologists prepare treatments and deliver better care to cancer patients.
TL: Democratizing AI in healthcare organizations is gated by three things: old processes for how to do analytics, old data and old-world thinking by traditional-minded leadership. Increasingly we are seeing tools to drive advanced analytics and AI to higher levels, but a key challenge is getting healthcare leaders to understand how to embrace and use AI. Consider the process for using analytics in hospitals today. If someone wants to gain better knowledge through analytics, he or she has to meet with informaticists, data scientists and analysts who have the necessary tools and skills. Basically, every analytics project passes through them. If you're going to talk about democratizing AI, this is a process that cannot keep up with growing demand. New tools and solutions are increasing the ability to provide things like predictive capabilities and “research on demand” to those on the front lines of managing care.
Right now, predictive capabilities are big in AI among both our healthcare clients and technology partners. Top use cases we are currently seeing include predicting readmissions, hospital acquired infections and falls. We're also seeing a growing number of use cases around predictive care guidance. Operationally, AI is being applied to the areas of fraud and denials management.
Beyond that, we're seeing a lot of interesting things our technology partners are doing with Microsoft's AI. One of my favorite projects is a partner using AI to innovate care in the intensive care units by providing real-time assessments and predictions to mitigate common risks such as ventilator-associated infections. By providing “suggestive analytics” to ICU caregivers the data shows AI reducing mortality rates as well as lowering average lengths of stay.
Q: Personal health data are the currency for AI in healthcare. What are some of the ethical considerations for using medical data in AI and some of the associated opportunities and challenges?
EB: In theory, more data means more trust. But the reality is that, at least in the beginning, when people do not understand what can be done with their data, people are skeptical and they don't want to share data. The more you can show people how their own data can save their life, or the lives of people like them, or the population at large, the more people will want to share.
Studies by groups like Wellcome Trust show the moment you tell people what can be done with their data, they begin to share everything. We believe when you start unfolding scenarios in precision medicine and genomics and explaining what can be done if data are aggregated for population health, then people will understand and will want to share their data.
It's not a question of having technology advanced enough to provide the right safeguards and protections to ensure privacy and security — we have that technology. The challenge from an ethical standpoint is to be clear regarding the new questions, paradigms and choices presented by some of these algorithms and solutions, and then bringing together the right people to consider those questions. We have the right technology to provide the right safeguards, and we know better education and knowledge drive sharing, so then we need to have the right balance between innovation on one side, and privacy and security safeguards on the other.
TL: Patient data is always sensitive. It's probably the biggest impediment to people moving into the world of AI. Healthcare leaders want to make sure data is safe, secure and in line with the standards we all have to operate within. Having said that, the upside of AI and PHI is our ability to predict and manage the things most consumers care about. If you're a woman and you just received a diagnosis for a malignant breast lump and you have the ability to use your PHI and other data to collaboratively decide with your care team the best treatment option, it empowers everyone to have more information to make better decisions.
Q: Do you encounter common misconceptions regarding AI that make its democratization particularly difficult?
EB: The key is telling people what can be done with their data. The moment you start sharing outcomes and scenarios when data can improve real lives, people's skepticism goes away. Basically, you start by explaining to people what sharing their data can do for them and for others, and once you do that, people can think about data donation as a way to help specific communities, like rare disease populations. Through transparency and conversation, we can be clear as to what sharing your personal health data can do for you, your children, your family, people like you and the population overall. Then data sharing becomes a common good.
TL: When it comes to democratizing AI and making it real in healthcare, the journey is new. However, the technology is there and it's getting better, and today it allows for what I call agile innovation. Through better technology, healthcare organizations can harness the power of AI to drive innovation in clinical and operational processes faster. With growing demands for quality and cost improvements AI becomes a competitive marketplace delineator.
A misconception among health leaders today is that AI is somehow “future-state” rather than something they can leverage today. In working with health leaders, a simple exercise is to have them define what they would like to predict that would make them better at something they care about. This often becomes the starting point for evaluating how AI can be applied.
Q: If you could change one thing in the realm of machine learning and data science overnight, what would it be?
EB: I'd democratize the understanding of coding and data science. I'd love for data science to be a must for children today, and accessible science for people overall. Then we would demystify it. It's similar to when people didn't understand chemistry and feared alchemists. I would love for people to have a better understanding of data science so they could use it and not be afraid of it.
I would also love for our technology people to interact more with clinicians and have conversations about improving clinical processes. I would love to see that partnership become stronger.
TL: We will reach a point where virtually all knowledge workers — from direct care providers to an employee in the business office — will have at their fingertips the ability to use predictive capabilities to make them better at what they do, no matter what that is.
For more content on AI and the digital transformation in healthcare, download this whitepaper here