I was playing a video game the first time I realized that I was interacting with artificial intelligence (AI). On a virtual battlefield in northern France, I was pitted against AI characters (bots) who, in the form of Axis infantry, were determined to kill me. These formidable adversaries adapted their tactics to my behavior and I learned to never underestimate them.
Fast-forward 20 years and the bots at Amazon, Google and Apple are now studying everyone’s behavior, learning from enormous data sets and seeing things we never saw before. As you read this, hundreds of companies are developing medical AI applications with billions of dollars in venture capital. Adoption is in the early stages – helping physicians with image interpretation and diagnostic support. But make no mistake, AI is poised to transform the health care landscape.
I recently had the opportunity to ask Dr. Joe Cummings* some questions about the use of AI in health care.
NH: “You have written that health care is primed for the introduction of AI. Within the context of health care, how would you best define AI?”
JC: “AI is a broad term that encompasses a lot of different things, but basically it means using computers to do things that humans do through higher reasoning; for example, learning, pattern recognition and problem-solving.
“As you narrow down the definition and activities within health care, there is a subset of AI that is currently the furthest along in development: machine learning (ML). Machine learning is a set of techniques that allows computers to learn from experience. It is fundamentally different from the way computers were programmed and used in the past, and this has set the stage for very powerful capabilities that are truly transformative.”
NH: “So, machine learning is the ‘secret sauce’ of AI. Can you give an example?”
JC: “Yes. In particular, ML is well-suited to image analysis and pattern recognition. So, anywhere in medicine that involves reading images is being impacted by ML. Radiology is obviously one area of focus and there are currently more than 100 companies developing ML-based radiologic capabilities. For example, there are applications that can read x-rays, diagnose fractures and spot bleeding or occlusion on CT scans during stroke.”
NH: “How should hospitals consider whether or not to purchase AI products as they come to market?”
JC: “There may be instances along the clinical care pathway where an AI algorithm can result in adverse events, patient harm or reduced treatment efficacy. When AI is used in the clinical care pathway, it is in fact a medical technology; therefore, it should be subject to the same government, payer and provider approval standards used for other medical devices. This means hospital committees need to rigorously evaluate it just like they would for any new emerging medical device to ensure that they are providing safe, effective, high-quality, cost-effective care to their patients.”
NH: “Aren’t these products already approved by the time they are brought to market?”
JC: “Because AI is so new to many hospitals, there hasn’t been a lot of time for them to determine how best to evaluate it. The initial reaction may be to consider it similar to software or information technology and therefore go through that approval process. My position is that when the technology is used to diagnose or guide patient treatment, you need the clinical perspective. Clinician input is critical.
“Similarly, the FDA has also been struggling with how best to consider AI and their approach continues to evolve over time. They are involved in the marketing approval process and have developed a safety-based approach to approval standards. As we know, however, just because a device is FDA approved, that doesn’t mean it should be used. Analogous to medical devices approved through the 510(k) process, the level of evidence may be sufficient for FDA approval, but insufficient to determine where and how the technology fits into clinical treatment. Again, this speaks to the need for hospitals to use a clinically integrated evaluation process.
“There is variation in the way an AI product may come into a hospital. Some AI products may be treated as a new product request and go to a value analysis committee. Some may enter as an IT application where they would be routed through a technical vetting process that emphasizes software compatibility and competing IT priorities; while yet other AI product requests may be recognized only as a new feature on a piece of capital equipment and slip under the clinical vetting radar.”
NH: “Can you share an example of a well-executed AI product evaluation?”
JC: “One example is in stroke. In some hospitals, the workflow for processing stroke patients through the emergency department can take too long. The adage in stroke is that “time is brain,” so reducing delays can be very impactful. Some hospital quality improvement initiatives have targeted their door-to-treatment time metric and found that for their situation, using AI for rapid CT-scan analysis and patient triage helps them reduce their average stroke notification time from about 45 minutes to five minutes.
This is expected to improve acute stroke outcomes and save most if not all of the downstream costs associated with long-term recovery as well. This is a good example of the implementation of innovative technology to provide value-based care by integrating both clinical and economic analyses.”
Artificial intelligence is primed to unleash enormous potential in health care. A rigorous evaluation process for the devices, products and technology driven by AI should include stakeholder clinician input to achieve this potential and put health care’s best foot on the path forward to better patient care.
For more information about how Vizient Advisory Solutions helps our members deliver operational, clinical and financial improvements that optimize the total cost of care, contact us today. Interested in learning more about how your organization could benefit from participating in a member-driven path to improvement? Read about our performance improvement collaboratives and benchmarking studies.
About the author. Neil Horton is consulting director for clinical advisory solutions at Vizient. He brings more than 30 years of experience in value analysis, perioperative nursing, surgical services, account management, contract negotiation and functional assessment to his role at Vizient. His background in various clinical areas, combined with Lean Six Sigma certification has provided him with a keen understanding of how to implement a robust, evidence-based, transparent process to drive growth and help achieve occupational objectives.
*Joe Cummings, PhD, is the technology program director for performance improvement collaboratives at Vizient. His aim is to help Vizient members stay at the forefront of technology assessment, acquisition, management and rational clinical use. Part of this mission is to scan the horizon for innovative new technologies that improve patient outcomes. Dr. Cummings has been at Vizient for more than 25 years and has been involved with hundreds of evidence-based evaluations of various high-impact medical devices and procedures. He holds advanced degrees in biomedical engineering.