Editor's note: A previous version of this story incorrectly referred to Paige as Paige.AI. The story has been updated to reflect the correct name.
Major players in the health care industry are betting big on artificial intelligence (AI) to revolutionize the way providers care for patients, especially as the world grapples with the new coronavirus pandemic.
Some evidence suggests those bets could pay off, but there's also research suggesting that AI isn't quite mature enough for providers to rely on—and particular skepticism about how effectively AI can be used to battle Covid-19. Daily Briefing's Ashley Fuoco Antonelli outlines what we know—and don't know—about AI in health care.
Investors, providers, and disruptors are betting big on AI
A recent global funding report on AI by CB Insights showed investors spent $4 billion across 367 deals in the AI health care sector in 2019, up from $2.7 billion across 264 deals in 2018. What's more, investments in health care AI outpaced AI investments for other industries, the report found.
The report also showed that investments in health care AI surged toward the end of 2019, with companies raising nearly $1.6 billion across 103 deals in the third quarter alone.
Further, some big corporations are teaming up with AI startups on health care products. FierceHealthcare's Heather Landi notes, for example, that Microsoft has joined forces with KenSci, which has created a risk prediction platform based on AI and machine learning systems; NVIDIA has teamed up with Paige, which uses AI to study cancer pathology; and Google has partnered with Suki, which has a voice-enabled digital assistant for doctors that runs on AI.
A survey released this month by the audit, tax, and advisory services firm KPMG found that health care CEOs also are adamant about integrating AI into their systems—and about AI's potential to improve health care. Melissa Edwards, managing director of digital enablement at KPMG, in the report said, "The pace with which hospital systems have adopted AI and automation programs has dramatically increased since 2017. Virtually all major health care providers are moving ahead with pilots or programs in these areas."
And a majority of health care leaders believe AI can have a valuable impact for their health systems, Advisory Board research shows. Advisory Board found 37% of leaders in 2018 expected AI technologies could present transformative value to their systems and 27% expected AI would have some incremental value for the systems.
So it's not surprising that, faced with the extraordinary task of fighting the United States' Covid-19 epidemic, providers are turning to AI as a potential tool. Stanford, for example, is evaluating whether AI can help identify Covid-19 patients who are likely to require intensive care. New York University researchers have embarked on a similar effort, and they've found that an AI tool helped to identify three factors that researchers could use to predict whether a patient would develop a severe case of Covid-19 with up to 80% accuracy.
Hospitals also are using AI to help screen patients and frontline medical workers who might be infected with the new coronavirus, to differentiate Covid-19 from other respiratory conditions, to track hospitals supplies and capacity, and to monitor patients outside of the hospital setting.
Research shows AI's promise in health care
Some research suggests health care leaders could be right about AI's potential.
For example, a study recently published in Nature found that an AI system developed by Google in some cases can detect breast cancer better than radiologists. As part of the study, researchers asked six radiologists in the United States to look at 500 mammograms and compared their responses to that of the AI—and the researchers found the AI system generally outperformed the radiologists in determining whether a woman would develop breast cancer.
Google's had some other early AI successes, as well. For instance, Advisory Board's Jackie Kimmel writes that "one Google-created algorithm was shown by Stanford researchers to diagnose skin cancer as well as a dermatologist, while another algorithm was as effective at diagnosing certain eye diseases as ophthalmologists." According to Kimmel, research showed another Google algorithm was 99% accurate when detecting breast cancer in lymph node biopsies, and a separate study "found Google's lung cancer screening algorithm outperformed all radiologists in the control group at correctly diagnosing the cancer—detecting 5% more true positives and cutting false positives by 11%."
And it's not just Google that's seen success with health care AI. For example, the Associated Press' Matt O'Brien and Christina Larson write that, as 2019 came to an end, the HealthMap AI system at Boston Children's Hospital "sent out the first global alert about a new viral outbreak in China" that has evolved into the current coronavirus pandemic.
Is AI always right?
But evidence also suggests AI can sometimes fall short.
For instance, while the Nature study on Google's AI system found that the system in some cases was better than radiologists at detecting and predicting breast cancer, it also found that radiologists in some cases outperformed the AI system. All six radiologists in the study at some point caught a cancer case that the AI missed.
And in the case of HealthMap's coronavirus alert, O'Brien and Larson report that New York epidemiologist Marjorie Pollack had begun working on an alert about the virus four hours before HealthMap's notice went out. O'Brien and Larson also note that HealthMap "ranked [its] alert's seriousness as only 3 out of 5," and "[i]t took days for HealthMap researchers to recognize its importance."
Some evidence also suggests AI technologies, if applied incorrectly, could worsen existing health disparities, Dhruv Khullar, a physician and researcher, argues. Khullar in a New York Times opinion piece writes that AI may be trained with narrow, unrepresentative data, as well as "real-world" data that perpetuates real-world biases. In addition, Khullar writes that even if an AI system's underlying data is "ostensibly fair" and "neutral," the technology still "has the potential to worsen disparities if its implementation has disproportionate effects for certain groups."
Barriers to adopting AI
Further, some health care CEOs say there are barriers that have slowed their efforts to adopt AI. Specifically, health care CEOs in the KPMG survey cited privacy issues and a lack of workforce training as barriers that have stymied their efforts to use AI. And Advisory Board's survey found that health care leaders viewed uncertainty regarding the costs and maturity of AI technologies as key challenges.
And as the Washington Post's Meryl Kornfield writes government regulation of AI could be coming. She notes that federal lawmakers last year introduced legislation that would give the Federal Trade Commission the authority to oversee how AI companies collect and use Americans' personal data—though the legislation hasn't yet advanced.
In the meantime, some states have taken steps to regulate the use of AI, and the White House last month released draft principals intended to guide federal agencies in regulating AI technologies. The Trump administration said the draft principles are intended to balance regulatory decisions regarding the technical and ethical issues related to AI with efforts to invent new AI technologies—and some stakeholders praised the draft principles as a positive step.
Further, Alex Engler, a Rubenstein Fellow in governance studies at the Brookings Institution, writes that although AI might be able to play a significant role in addressing future disease outbreaks, AI's role in addressing the coronavirus pandemic may be limited. He notes that, currently, "AI is only helpful when applied judiciously by subject-matter experts," it "needs tons of prior data with known outcomes," which can be hard to come by with such a new virus.
What's to come?
But the recent investing boom in health care AI—paired with health care leaders' excitement about AI technologies and their current applications to the Covid-19 pandemic—suggest providers will continue integrating AI into their businesses.
Health care leaders are beginning to look beyond workflow efficiencies and toward the role AI can play in patient care. About 90% of health care CEOs in the KPMG survey said they were confident AI will improve patients' experiences, particularly when it comes to diagnostics.
Moving beyond diagnostics, Advisory Board experts note that there's been "rapid development" of AI technologies focused on chronic disease management, which "could be a game changer" for health care systems across the globe. Advisory Board experts also have flagged opportunities for population health leaders to use properly trained AI and deep learning systems to address inequities in care, particularly among people of color.
However, Advisory Board experts caution that providers will need to be smart about how they train and use new AI technologies, especially when it comes to verifying the technologies' accuracy. Particularly, they warn that clinical decision making "is often quite messy and highly dependent on doctor intuition—and understanding this fact is essential to understanding the strengths and limitations of AI."