Artificial intelligence (AI) seems to pop up in headlines nearly every day. With applications ranging from individual nutritional guidance to prediction of premature death, ever-evolving computer-based algorithms present intriguing possibilities for the future of healthcare.
Depending on how proactive your facility is and how rich in resources, you may already be seeing or hearing about the impact of AI. In many healthcare organizations, AI applications may be more of a theory than a reality, but considering the rapid advances in robotics and other technology, AI may be adopted faster than you might imagine.
In this issue, we begin a series of articles on AI to present developments to date and projections for where the technology may soon take us (cover story). The series will continue in the next several issues of OR Manager.
Inspiration and incentives
The Centers for Medicare & Medicaid Services (CMS) recently jumped on the bandwagon with its launch of the CMS Artificial Intelligence Health Outcomes Challenge–a new competition that "aims to develop artificial intelligence-driven predictions that healthcare providers and clinicians participating in CMS Innovation Center models could use to reduce the burden to perform quality improvement activities and make quality measures more impactful," according to a CMS press release.
Participants who are selected for Stage 1 of the challenge will develop algorithms that predict health outcomes from Medicare fee-for-service data. The competition will have a Stage 2 (with more information to come later this year), and up to $1.65 million in total will be awarded to participants in both stages.
Of course, well before the advent of AI, reducing variation in clinical care was linked to cost savings and better patient safety. Many OR Manager articles have addressed ways to standardize protocols, and more will be forthcoming.
Meanwhile, Flagler Hospital, a 335-bed community hospital in St Augustine, Florida, recently leveraged AI to improve adherence to order sets. By uploading data from a variety of sources into AI software, the hospital analyzed pneumonia, which is often the culprit behind readmissions.
For pneumonia patients with chronic obstructive pulmonary disease, the data showed outcomes could improve by starting nebulizer treatments as soon as possible. When clinicians saw the data, compliance with order sets rose to nearly 80%, thereby reducing the readmission rate from 2.9% to 0.4%.
Another promising application comes from the University of Nottingham in the UK, where machine learning AI algorithms predicted premature death more accurately than did traditional models like Cox regression in a study of more than half a million people aged 40 to 69 years.
Proceed with caution
In March, the Cleveland Clinic launched a Center for Clinical Artificial Intelligence to develop AI research projects designed to improve patient outcomes and lower costs. "It's likely that healthcare will be the [industry that is] most disrupted by AI over the next decade," Aziz Nazha, MD, told Business Insider Intelligence. Nazha, director of the new center, notes that there's high interest and activity around AI in healthcare but as of yet not a lot of value or implementation. Although 53% of US providers say their organizations are using AI and data visualizations for clinical decision support, just 7% of respondents claim their organizations are "extremely" effective in using AI, according to a recent HealthData Management article.
And, like any new technology, AI comes with certain caveats. Researchers at Harvard Medical School and Massachusetts Institute of Technology in Boston recently warned about possible "adversarial attacks"–manipulations that can use digital data to change the behavior of AI systems and thus lead to false diagnoses. Of even greater concern, they say, is the potential for corruption among healthcare providers who might be tempted to manipulate data to increase payments or get regulatory approval for a new device.
Impact of individual data
Regardless of where your facility is on the AI spectrum, here's an interesting tidbit on how the technology may evolve. If you've ever failed to lose weight, even after trying multiple different types of diets, take heart: It turns out that response to food is highly individualized, and one diet does not fit all.
In a recent study of glycemic responses, for example, researchers used machine learning (a subtype of AI) to analyze a vast amount of data and found that gut bacteria, not food, played the bigger role in determining blood glucose levels. In other studies, these researchers also found that many healthy people have high glucose levels after eating. Furthermore, depending on a person's individual gut microbiome, some foods ordinarily considered as healthier choices might actually cause glucose spikes.
So if you've ever wondered why just thinking about a doughnut tips your scale the wrong way while a relative/co-worker/friend can eat doughnuts every day without gaining an ounce, now you know. Here's hoping our AI series provides food for thought that benefits all of us, despite our individual differences! ✥
References
Bazzoli F. Providers begin using AI to improve clinical decision making. HealthData Management. January 22, 2019.
Cohen J. Flagler Hospital combines AI, physician committee to minimize clinical variation. Modern Healthcare. March 9, 2019. https://innovation.cms.gov/initiatives/artificial-intelligence-health-outcomes-challenge/.
Lineaweaver N. Cleveland Clinic has launched the Center for Clinical Artificial Intelligence. Business Insider. March 18, 2009.
Metz C, Smith C S. Warnings of a dark side to A.I. in health care. New York Times. March 21, 2019.
Topol E. The A.I. diet. New York Times. March 3, 2019.
University of Nottingham. Artificial intelligence can predict premature death, study finds. ScienceDaily. March 27, 2019. https://www.sciencedaily.com/releases/2019/03/190327142032.htm.