Artificial intelligence (AI) and machine learning are thrown around so often in healthcare tech circles that they begin to sound like mere buzzwords after a while. It can be hard to make sense of every tech startup and software developer claiming to have the next big solution for how these technologies can improve healthcare.
But it is increasingly clear that AI is going to change healthcare in a tangible way in the near and long-term.
Use cases for AI have been a hot topic in recent years, particularly at conferences such as HIMSS18. Vendors are eager to show off their latest AI and machine learning developments, and researchers unveil new ways they’ve found to put the technology to use in various care settings.
The areas of potential disruption were outlined at a recent HIMSS education session titled, “How AI and Machine Learning are Disrupting the Current Healthcare Ecosystem.” They are:
- Care delivery and patient diagnosis
- Claims collection and payment
- Clinical decision support
- Population health
- Precision medicine
In order for these areas to turn the promise of AI and machine learning technology into a widely practiced reality, use cases are identified and with a very specific focus, the new tech is applied to a specific problem.
Real Life Use Cases
There are a number of areas already being affected by AI and machine learning, so much so that major tech giants are jumping at the opportunity to work on projects in this arena. One such example is imaging analytics and the effect that could have on radiology.
Scientists have been looking at deep learning technology, through which a computer aims to mimic the observational power and decision-making ability of the human brain. A radiology use case discussed at the HIMSS session demonstrated a machine “learning” information that scientists looking at brain scans could not.
It examined MRI texture features as biomarkers in patients with a specific type of brain tumor. It proved, as former executive of Alphabet Inc., Eric Schmidt, put it in his opening keynote at HIMSS18, “computers can see things better than we do.” It identified the tumor based on the texture of the brain more accurately than its human counterparts could.
Complementing the skills of humans, AI can identify even the most minuscule changes to imaging scans, aiding radiologists for quicker diagnosis and thus, doctors in faster clinical decision making. This has led to Microsoft becoming directly involved in cancer research and IBM in heart disease.
Uses in Precision Medicine
Most commonly, AI and machine learning are seen as a vital component of data analysis. With the ability to comb through a volume of data larger than any human could process, machines can deliver insight that would normally be far more difficult to attain.
For physicians, the gift that a machine, such as IBM’s Watson, provides is time: more time to see patients or keep up with advances in medicine. Watson can provide analysis of patient information and spit out a range of potential treatments.
This process of clinical trial matching has been a challenging project that has played out over many years. Yet, for patients in need of new therapies that could prove more effective and play into genetic solutions, this an important part of the overall precision medicine effort.
“For an oncologist, let’s say, someone working in a smaller town where there is maybe one other clinician in that specialty, they have to be focused on all types of cancer, whereas someone in a more-populated area can focus on one type of cancer and just keep up with the research and advances in that type of cancer,” said Cris Ross, Chief Information Officer of Mayo Clinic, during the session. “But for the doctor in that rural area, AI can supplement their knowledge and provide guidance in areas they can only try to keep up with.”
James Golden, Managing Director of PricewaterhouseCoopers, was involved in the early days of a partnership between MD Anderson Cancer Center and IBM’s Watson that sought out the development of an “Oncology Expert Advisor” for clinical decision support. The project has since been put on hold, but according to Golden, the work done in that moonshot has been vital to the growing success of clinical decision support systems, clinical trial matching technology and has helped advance the conversation around the democratization of care.
“Pioneers catch all of the arrows,” Golden said. “What that MD Anderson project did was incredibly bold, and it was early days for Watson doing deep analysis. But the work that was done was excellent from a technology standpoint. It was a big moonshot, but that democratization of care is, from a provider perspective, is the most exciting and first achievable things using AI.”
Golden has also seen AI at work in the payer world, where recently he conducted an AI workshop. He believes that there is low hanging fruit in the payer field and we could see AI put to work there sooner rather than later.
“I was stunned when I saw how many people have to touch a claim,” Golden said. “We can call it AI, I’m going to call it thoughtful automation and it includes Natural Language Processing. We can do more for payers immediately that will improve the lives of physicians immediately, simply through thoughtful automation of processes. We can do that now.”
AI, Machine Learning as a Service
Looking at the immediate future, the cost and expertise involved in implementing AI and machine learning can offset its benefits in the eyes of healthcare executives looking at a budget. But 2018 promises to be a major development in vendor solutions on this front, as machine learning as a service creates a low-cost starting point for launching these efforts.
This allows providers to pass the intricate work involved in collecting data and storing, moving and analyzing it onto a third-party vendor which often uses a “data lake” approach to analytics. This means that there are no infrastructure or talent requirements from the hospital or healthcare system to concern itself with, allowing it to move more quickly toward implementation of precision medicine and population health initiatives.
The “as a service” market for machine learning has huge potential for growth and is expected to be worth around $20 billion by the year 2025, according to Transparency Market Research. Healthcare could, by some estimates, account for as much as a quarter of that.
“I’m not sure we’re going to see AI and machine learning used for diagnosis and treatment first,” said the Mayo Clinic’s Ross. “We’ll see it used in all the stuff that leads up to diagnosis and treatment over the next two to three years, but it will take the next seven to eight years to see it in diagnosis and treatment. Most technology throughout history has been implemented incrementally, not through giant moonshots.”