The future is never executed exactly how it's envisioned. Using pop culture texts as examples, hoverboards didn't hit the mass market by 2015 as Back to the Future II had implied, but there is still time to reach The Jetsons' flying cars by 2062 (I wouldn't bet on it, though).
The reality is many futurists currently seek to conquer boring activities in hopes to make our daily lives easier. It's this sobering thought many in healthcare industry should square against the hype and promise that's sweeping the industry over artificial intelligence and machine learning.
Though artificial intelligence hasn't become deeply seeded in the healthcare industry (it is just still recovering from adopting EHRs on the whole), there is potential to make workflows more efficient with machine learning and automation. As hospitals and health systems are largely fixed costs businesses, administrators are no doubt interested in tech that prove ROI. Some wonder or worry if these new tools will lead to unintended consequences such as the loss of jobs. Many experts dismiss this concern, believing new technology will — interestingly — alleviate burden and allow clinicians to return to their patients, rather than planting their faces in screens.
Providers are still exploring the space (many are looking into readmission use cases) but considerations over liability still need to be firmed up as the industry enters this brave new world. Healthcare Dive spoke with some of the prominent thought leaders across the industry on the subject. Here are the main takeaways to frame future discussions of AI, machine learning and new tech tools in the space.
Once you understand [artificial intelligence] a "how," you can break down the workflow of the physician to understand where these technologies can be deployed.
Dr. Jack Stockert
Managing Director, Health2047
1. Providers are exploring potential use cases, but it's still a new area
One of the truisms in healthcare is that red hot trends can move at glacial paces in the space. Artificial intelligence is the example du jour. While it was flagged as a tech for administrators to familiarize themselves with this year by PricewaterhouseCoopers' Health Research Institute and was a major theme at HIMSS17, only 4.7% of a Healthcare IT News and HIMSS Analytics survey in April responded they are already using AI tools.
About 35% of respondents were prepping to implement AI capabilities in two years. The survey only consisted of 85 respondents, but it does point to the growing interest in the tech. A recent IDC Health Insights report also found hospitals are actively investing in robotics over the next three years.
Some of the reasons providers are slow to adopt technologies can be purchasing and implementation cycles, costs, or the evidence surrounding a product. For all the hype around AI and machine learning, there are still questions about where to prioritize efforts as the potential adoption landscape is very broad. But the potential does exist — and the interest in such tools has been gaining traction. Leading provider systems see new tech tools as efforts they should throw their weight behind. For example, Mayo Clinic recently announced its teaming up with AliveCor to help prevent sudden death while Jvion — developed in collaboration with Mayo — launched a product that uses AI to help identify vulnerable patients and interventions to reduce avoidable deaths
Daniel Barchi, CIO at NewYork-Presbyterian (NYP), noted interest in the topic may be somewhat sudden, but the technology shouldn't be discounted. He stated NYP is focused on AI and machine learning as a change agent in healthcare and the system's C-suite has frequent conversations on the topic.
It's important to remember that AI and machine learning are "hows" like a combustion engine or a camera, Jack Stockert, a physician and managing director at Health2047, told Healthcare Dive. The broader question surrounding these "hows" are where they fit into organizations and advance productivity and transform care delivery interactions. One forward-looking example could be using speech-to-text and real-time translation algorithms over Skype to change translation services in a telemedicine setting.
"Once you understand it's a 'how,' you can break down the workflow of the physician to understand where these technologies can be deployed," Stockert said.
"Generally when we look at artificial intelligence, there's a lot of great promise in primary care," Ben Isgur, director of PricewaterhouseCoopers' Health Research Institute, recently told Healthcare Dive, adding that clinicians' workflow often involves the review of data, whether it be medical history or test results. Isgur says algorithms can be helpful to review large data sets.
If you think about a physician, they wish they were spending 80% of their time with patients but much of their time is spent with medical billing, gathering data, accessing labs and summarizing patient family histories.
Daniel Barchi
CIO, NewYork-Presbyterian
Dr. Michael Oppenheim, VP & CMIO of Northwell Health, agrees that digital tools can help with data review. "The things computers do really well, and maybe better than humans, are looking at very large volumes of data and identifying mathematical or statistical interrelationships between them," he told Healthcare Dive.
Though Northwell is just starting to evaluate how to incorporate these new tools, he shared the New York-based health system is exploring use cases surrounding image/pattern recognition and using computers to sift through large volumes of discrete data to divine insights into readmission rates, length of stays and clinical deterioration. These projects focus on how a computer can help find inter-relational patterns between defined patient features and a set of mathematical rules to predict a certain outcome based on known data sets.
Readmission seems to be a common thread for providers. In addition to Northwell and Mayo, Barchi noted NYP is also looking into readmission prediction tools. Methodist Hospital uses a tool from CareSkore that boasts AI and machine learning to help with patient interventions by linking claims data and risk stratification to predict readmissions, Dr. Madison Sample Jr., vice chairman, department of anesthesiology at Methodist Hospital, told Healthcare Dive. He added Methodist was able to drive down its readmission rate from 26% to around 12% using the tool.
Joseph Fournier, SVP and Chief People Officer at Intermountain Healthcare, shared with Healthcare Dive the system is currently running simulations to help optimize how to source and schedule workers.
But for the most part, most work in the provider space regarding new tech tools is exploratory. "We're so early in this space," Oppenheim said. In part, he stated, you have to have enough data for machine learning technologies to work and evolve to be useable and make an impact. While it may be a hot topic, the industry may well hit disillusionment over these new tools while data are populated to feed the machines.
2. Most think clinician jobs won't be replaced by AI tech tools. But that doesn't mean workflows won't change.
Artificial intelligence has received more than its fair share of hype and think pieces over how "The Robot Economy" could affect healthcare.
In 2014, Vivek Wadhwa, distinguished fellow at Carnegie Mellon University Engineering, Silicon Valley, at Rock Health's Health Innovation Summit, stated physicians would be replaced by AI. While a provocative thought, many in the industry believe work in the clinical space will be augmented rather than replaced. "Eventually, computers will replace 80% of what doctors do and amplify their capabilities," wrote Vinod Khosla of Khosla Ventures back in 2012.
In other industries that have moved to advanced technologies, certain positions were replaced because they were centered on predictable manual labor. Stockert points out while most in the industry would agree that the medical profession could be helped with the automatic collecting and input of data, some can overstate the potential of AI and machine learning in the short-term, especially when the profession's task moves into unpredictable physician labor areas, such as surgery. Everyone's appendix is similar yet different, for example, and the tiny yet numerous decisions that can go into removing such an organ could prove difficult to automate.
"Augmentation could help but there are a certain set of physical and mental aspects of being a doctor which is hard to automate because you have to write an infinite number of complex rules," Stockert said.
"I think there's still a long way to go in the computer science space over how to mimic human reasoning," Oppenheim stated.
I think there's still a long way to go in the computer science space over how to mimic human reasoning.
Dr. Michael Oppenheim
VP & CMIO, Northwell Health
"Even though technology continues to be introduced into the workplace, there will always be a role for people," Fournier said. "Healthcare is always going to be a people business because it involves people. Though the way people interact with each other in healthcare will be vastly different 10 years from now.“
Instead, artificial intelligence is seen as the opportunity for a physician to partner with technology to make a quicker, accurate diagnosis and spend more time with patients, Isgur says.
For example, James Golden, managing director of PwC Health Advisory, shares that some of the low-hanging fruit around AI and machine learning is with radiology/imaging and pathology. If a radiologist can go from reviewing an average of 30 images a day to 60 images a day or help physicians go from seeing 20 patients a day to 30 a day, then new technology can add opportunity costs or revenue to the organization, Golden notes.
"If you think about a physician, they wish they were spending 80% of their time with patients but much of their time is spent with medical billing, gathering data, accessing labs and summarizing patient family histories," Barchi says. "[If] a machine can take mundane tasks away from them, that'd be a great win. We think [these tools will eventually] change everybody's jobs" and will ultimately make clinicians more efficient.
Oppenheim says the ultimate goal at Northwell is to have workers operate at the top of their license by allowing physicians to focus on cognitive and clinical tasks while duties that require different levels of clinical knowledge and skills can be offloaded to different staffers. Technology will help drive the shift toward top of license where computers can pick up certain data-oriented clinical tasks so clinicians can focus on being clinicians, Oppenheim stated.
Golden notes tools such as Alexa are being looked at in the healthcare space to help with care delivery design. Voice assistants could be used for environments where the ability to write or take notes is limited, such as in an operating room.
An exciting aspect to deploying such technologies, according to Stockert, would result in looking at how to work with automation to enhance productivity and create new clinical workflows and roles for physicians as a result.
3. Wrinkles like liability still need to be ironed out
The possibility for productivity is exciting, but there are some caveats to figure out before widespread adoption will likely occur. Last month, Elon Musk stumped for more regulation in the field. “AI is a rare case where we need to be proactive about regulation instead of reactive," Musk was quoted in The Verge. "Because I think by the time we are reactive in AI regulation, it’s too late.”
While Musk's comments may be more cerebral than its potential uses in healthcare, there are certain considerations for AI as it advances through the industry. For one, because machines will use data points to make decisions, physicians could find themselves in situations where they may feel uncomfortable with machine-enabled decisions. "How do you put in checks and balances in place to maintain a level of civility and ethics around care decisions that we currently hold physicians to, which machines won't adhere to unless you write rules around it or unless you audit how they're making decisions?" Stockert asked.
A question of liability also arises if and when a machine takes an action that harms a patient.
The traditional path of tech firms in Silicon Valley is to write long terms of service waiving liability rights, Stockert notes, adding healthcare works differently as doctors carry liability in their practice. The question, Stockert posits, is if the industry will see a different tone from companies bringing machine learning products to market and, if not, who is liable in a world where an AI engine diagnoses a patient with a particular lesion in their lung and recommends a lung resection when no lesion is actually present. "It's an important question that's easy to lose sight of, but it's a big open question for what happens."
Whatever the actual impacts of AI and machine learning turn out being, one thing is for certain: Technology will change the workflows of administrative and clinical workers.
