AI and Machine Learning in Healthcare and Anesthesia: Where Are We Going?

​Summary 

The growth of big data, advances in software and hardware, and the development of cloud-based business models are fueling an explosion in the use of machine learning (artificial intelligence), particularly in healthcare. All clinicians, including anesthesiologists and nurse anesthetists, are likely to find themselves incorporating ML tools and capabilities into their practices in the not-toodistant future. We offer an overview of ML's current and future applications in healthcare and medicine, including its strengths and limitations, as well as strategies for avoiding pitfalls. 

Medical researchers, informatics specialists and digital entrepreneurs have been exploring the use of artificial intelligence (AI) in the healthcare sphere for decades, but it is only within the past couple of years that the technology has really begun to take off. Indications are that in healthcare, AI, now commonly known as machine learning (ML) is set to explode, in fact. Imagine an environment in which machines capable of cognitive computing and processing vast amounts of data can support you with unprecedented accuracy, efficiency and patient-specificity on everything from monitoring the depth of anesthesia, determining the amount of anesthetic gas to administer, somatosensory evoked potential monitoring, classifying patients and mitral valve analysis to coding and billing and generating music with the psychoactive and analgesic properties of opioids. 

Gear up, because experts say the era of ML has arrived. 

It's an aspect of what Klaus Schwab of the World Economic Forum calls the Fourth Industrial Revolution, "a range of new technologies that are fusing the physical, digital and biological worlds, impacting all disciplines, economies and industries, and even challenging ideas about what it means to be human." 

Three factors are fueling the trend, in ML writes Steve Ranger on ZDnet: 

The rise of big data. The amount of digital data from sources including everything from journal articles and electronic medical records (EMRs) to results of clinical trials and genomic profiles is doubling every two years. The 4.4 zettabytes of data available in 2013 (equal to 4.4 trillion gigabytes) will reach 44 zettabytes by 2020, and ML will be needed to keep track of it, according to medical futurist Bertalan Mesko, PhD. The burgeoning of big data is facilitating the emergence of, as well as driving the need for, ML. 

Software and hardware advances, including the development of deep learning software that facilitates MLusing neural networks, a structure as close as possible to the human brain; and parallel computing, a type of computing in which many calculations or processes are carried out simultaneously. 

Cloud business models. "Before the cloud, most AI work was isolated and relatively high cost, but the economics of the cloud mean machine learning capabilities, such as recognizing faces or translating languages, will be cheap and easy to use," writes Ranger. "It is this realization that is triggering both the explosion of highly specialized MI startups, as well as the major machine intelligence pushes at Google, Facebook, Microsoft, Apple, IBM and their various global rivals." 

Machine Learning Defined 

Unlike robotics—the field of computer science involved with creating devices that can move and react to sensory input—ML encompasses such sub-areas as vision, natural language processing, and knowledge representation and engineering. WhatIs.com defines it as "a type of artificial intelligence that provides computers with the ability to learn without being explicitly programmed" and "focuses on the development of computer programs that can change when exposed to new data." 

Many, if not all, of these areas of ML have potential applications in robotics; however, ML does not necessarily involve robotics, and robotics does not necessarily involve ML. While the two are often closely linked, they are also distinct areas that do not necessarily overlap. 

As Ernest Davis, a computer scientist at New York University, describes AI in an article in Live Science, "There are a number of cognitive tasks that people do easily — often, indeed, with no conscious thought at all — but that are extremely hard to program on computers. Archetypal examples are vision and natural language understanding. Artificial intelligence, as I define it, is the study of getting computers to carry out these tasks." 

The technology still has a way to go, notes Katherine McKeown, a professor of computer science at Columbia University, in the Live Science article. "AIs that can handle certain tasks well exist, as do AIs that look almost human because they have a large trove of data to work with," she said. And yet, "computer scientists have been less successful coming up with an AI that can think the way we expect a human being to, or to act like a human in more than very limited situations. . . I don't think we're in a state that AI is so good that it will do things we hadn't imagined it was going to do," she said. 

One thing some experts believe ML is not likely to do is to take over the work of physicians. IBM's AI platform Watson Health, offers a case in point. The platform doesn't replace oncologists, but shows potential as a powerful decision support tool. A test of Watson at Memorial Sloan Kettering Cancer Center found that the software was capable of predicting lung cancer with 90 percent accuracy, compared with human rates of 50 percent. Driving the superior performance was the suite of tools' ability to synthesize and apply massive amounts of data, including 600,000 medical findings, 1.5 million patient records and 2 million pages of medical journals—a level of information absorption that surpasses human capacity. 

"One of the biggest challenges in the field of ML right now is that these models can't explain themselves the way a doctor can," adds Doug Marcey, vice president of technology for Plexus Technology Group, LLC, which is spearheading the development of decision support products for ABC that incorporate predictive modeling capabilities and elements of machine learning. 

"If you ask a doctor why they prescribed a drug, they can explain their thought process," Marcey said. "A lot of ML models are just statistical. You can't ask the model why it made a diagnosis or recommendation. That's why, at least in the short term, we're not going to replace people. We need someone at the helm who's going to take responsibility. At this point, you're not going to want to turn your care over entirely to those engines, even though some people with high ideas want that to be possible." 

Tipping Point

And yet, ML technology is far enough along that by as soon as next year, approximately 30 percent of health systems will be using cognitive analytics alongside patient data and real-world evidence to develop personalized treatment regimens, according to industry analyst International Data Corporation, whose work is cited in a May Forbes article.

In addition, a survey of healthcare executives published in May by AI market research firm TechEmergence reports that more than 50 percent of respondents predicted ML will be ubiquitous in healthcare by 2025. Participants ranked decision support systems for improving patient outcomes as the most likely application for ML, improving health outcomes and reducing costs as the two most important drivers for ML product purchases, and improving chronic conditions, notably diabetes, as the most important type of patient care improvements achievable through the use of ML. 

The time to embrace ML is now, suggests an article in Healthcare IT News. "AI, cognitive computing and deep machine learning are still nascent technologies but consultancies are suggesting that healthcare organizations begin working these technologies now rather than waiting," notes the article, citing a report from Gartner Group. The research and advisory firm concludes "the risk of investing too late in smart machines is likely greater than the risk of investing too soon." 

On his blog, Bertalan Mesko predicts that AI will be applied in healthcare to: 

Mine medical records. The Google Deepmind Health project, for example, is mining medical records in cooperation with the Moorfields Eye Hospital National Health Service Foundation Trust in England to improve treatment. "AI systems can be trained to learn how to interpret test results for themselves. In time, they should also be able to learn which types of treatments are most effective for different patients," the project's website explains. 

Design treatment plans. Watson, for example, can analyze the meaning of data in clinical notes and reports to help select an oncology treatment pathway. The program identifies potential treatment plans by combining data from the patient's file with clinical expertise and external research. 

Perform repetitive jobs. "Cognitive assistants" with analytical and reasoning capabilities and a wide range of clinical knowledge will be able to support clinical decision-making, freeing clinicians to focus on more difficult cases. 

Support consultations. British online consultation service Babylon Health has launched an application that allows users to report their symptoms, which are then checked against a database that uses speech recognition. The app also reminds patients to take their medication and follows up with them to ask how they are feeling. "Through such solutions the efficiency of diagnosing patients can increase by multiple times, while the waiting time in front of doctors' examining rooms could drop significantly," writes Mesko. 

Help manage medications and provide health assistance. A virtual nurse developed by medical startup Sensely uses machine learning to support patients with chronic diseases between physician visits, with an emphasis on customized monitoring and follow-up care. 

Help patients make healthier choices and decisions. Open AI ecosystems, named by the World Economic Forum as one of the top 10 emerging technologies of 2016, will harness vast amounts of data to provide patients with information on adopting a healthier lifestyle and help health systems design and improve services based on patients' needs and habits. 

Develop drugs. Atomwise, which uses supercomputers to identify therapies from a database of molecular structures, found two drugs predicted by ML to significantly reduce Ebola infectivity. The analysis, which typically would have taken months or years to complete, was finished in a single day. 

Facilitate precision medicine. The company Deep Genomics, for example, uses huge data sets of genetic information and medical records to look for mutations and linkages to disease. The company is developing technology that will tell physicians what will happen within a cell when DNA is altered by natural or therapeutic genetic variation. 

An article in the May 24 issue of Newsweek asserts that ML even has the capacity to "cure America's sick health care system" by driving down the costs of care while increasing effectiveness. For example, investors predict that medical startups using ML to enhance the management of diabetes could reduce the annual costs associated with diabetes treatment by $100 billion annually, the article reports. 

"I think machine learning will become integrated into anesthesia, but will never replace the physician," said Jody Locke, ABC vice president for anesthesia and pain practice management. "The big issue in anesthesia—and the big potential for machine learning in anesthesia—is risk stratification and identifying the potential for never events." 

Healthy Boundaries 

While remaining open to ML's potential, clinicians and providers would also be wise to remain vigilant for side effects and unintended consequences, cautions consultant Anne Bruce in an article for staffing solutions provider CEP America. "AI technology, while powerful, has inherent limitations when compared with human understanding," Bruce writes. 

She offers the following strategies to help clinicians and healthcare organizations use AI to improve care while minimizing the risk of division or disruption within the organization and care teams: 

Frame AI as a tool to augment (not replace) training, experience and clinical judgment.
Emphasize the ways AI platforms can be a solution for patients, providers and the hospital, such as by providing more provider/patient face time or more personalized treatment plans.
Advocate with EMR vendors and end-users for increased interoperability between systems. Lack of EMR integration is still a major barrier to the effective use of AI, she says.
Engage colleagues, collaborative partners, medical students and even patients in discussions about the possibilities and limitations of AI.
Welcome the benefits of AI while acknowledging the need for vigilance in implementation and use. AI should never serve as substitute for medical knowledge and judgment. 

In an article in Becker's Health IT & CIO Review, Fatima Paruk, MD, of Allscripts advises health systems and clinicians to "prepare now by establishing data governance, infrastructure and strategy; and by gaining an understanding of where the limitations are today, and what the data has the power to do for us. As AI systems consume large amounts of data, the best preparation is to make the effort to accurately collect, annotate and curate existing data. Data has extreme value, and knowledge of the process of making the data available for analytics should be a key objective of any institution." 

Similarly, Lisa Suennen of GE Ventures encourages clinicians to learn about AI's inherent strengths and weaknesses, and to "become comfortable with technology that can be a tool, not a threat. For those who aren't technically conversant, it's an excellent time to invest in education on the topic to know how best to maximize the human side of medicine through targeted technological application." 

As anesthesiologist Arthur Atchabahian, MD, of New York University wisely puts it in an article on robotics in Anesthesiology and Pain Medicine, "We cannot resist technological advances. Our role is to manage those advances to best benefit patients, but also to avoid disappearing, like travel agents and bank tellers, who were displaced by the Internet. Taking ownership of the technology is paramount: we need to be the drivers of progress rather than those who resist it out of inertia." For information on ways that decision support tools incorporating predictive modeling and machine learning capabilities might improve productivity and efficiency in your practice, please contact info@anesthesiallc.com. 

We want to hear from you. Do you have a topic you would like to see covered in an ABC eAlert? Please send your suggestions to info@anesthesiallc.com. 

With best wishes, 

Tony Mira

President and CEO