It’s impossible to read about the future of healthcare without encountering two pixelated vowels that, together, represent the hopes and fears of an industry seeking more intelligent solutions.
Though the field of artificial intelligence (AI) has been around since 1956, it has made precious few contributions to medical practice. Only recently has the hype of machine-based learning begun to merge with reality.
What Is Artificial Intelligence, Really?
Confusion surrounding AI—its applications in healthcare and even its definition—remains widespread in popular media. Today, AI is shorthand for any task a computer can perform just as well as, if not better than, humans.
But there are different forms of computer intelligence to consider when thinking about its role in medicine.
Most of the computer-generated solutions now emerging in healthcare do not rely on independent computer intelligence. Rather, they use human-created algorithms as the basis for analyzing data and recommending treatments.
By contrast, “machine learning” relies on neural networks (a computer system modeled on the human brain). Such applications involve multilevel probabilistic analysis, allowing computers to simulate and even expand on the way the human mind processes data. As a result, not even the programmers can be sure how their computer programs will derive solutions.
There’s yet another AI variant, known as “deep learning,” wherein software learns to recognize patterns in distinct layers. In healthcare, this mechanism is becoming increasingly useful. Because each neural-network layer operates both independently and in concert—separating aspects such as color, size and shape before integrating the outcomes—these newer visual tools hold the promise of transforming diagnostic medicine and can even search for cancer at the individual cell level.
AI can be sliced and diced many different ways, but the best way to understand its potential use in healthcare is to break down its applications into three separate categories: algorithmic solutions, visual tools and medical practice.
In healthcare today, the most commonly used “AI” applications are algorithmic: evidence-based approaches programmed by researchers and clinicians.
When humans embed known data into algorithms, computers can extract information and apply it to a problem. Take cancer treatment, for example. Using consensus algorithms from experts in the field, along with the data that oncologists enter into a medical record (i.e., a patient’s age, genetics, cancer staging and associated medical problems), a computer can review dozens, sometimes hundreds, of established treatment alternatives and recommend the most appropriate combination of chemotherapy drugs for a patient.
Perhaps my favorite algorithmic solution comes by way of Dr. Gabriel Escobar and his colleagues in The Permanente Medical Group’s division of research.
The team’s research centered on one of the most important populations in any hospital: patients in a medical or surgical unit who will experience a deterioration in clinical status and be transferred to the ICU.
Though these patients receive intensive care for an acute event, and seemingly return to their prior health status, they are three to four times more likely to die than if a physician had intervened and prevented the deterioration in the first place.
Dr. Escobar, along with division chief Dr. Tracy Lieu and associate executive director Dr. Philip Madvig, compiled data from 650,000 hospitalized patients, 20,000 of whom required this type of ICU transfer.
The team then created a predictive analytic model to identify which hospitalized patients today are most likely to end up in the ICU tomorrow. They then embedded the algorithm into a computer system, which continuously monitors the health status of all hospitalized patients. Finally, they designed alerts to notify physicians whenever a patient is deemed “at risk.” With this information, the doctors can intervene in advance of a major complication and save hundreds more lives each year.
To appreciate the potential of visual pattern recognition in medical care, one must understand how often the human eye fails even the best clinicians.
A pair of independent studies found that 50% to 63% of U.S. women who get regular mammograms over 10 years will receive at least one “false-positive” (a test result that wrongly indicates the possibility of cancer, thus requiring additional testing and, sometimes, unnecessary procedures). As much as one-third of the time, two or more radiologists looking at the same mammography will disagree on their interpretation of the results.
Visual pattern recognition software, which can store and compare tens of thousands of images while using the same heuristic techniques as humans, is estimated to be 5% to 10% more accurate than the average physician.
The accuracy gap between the human and digital eye is expected to widen further, and soon. As machines become more powerful and deep-learning approaches gain traction, they will continue to advance such diagnostic fields as radiology (CT, MRI and mammography interpretation), pathology (microscopic and cytological diagnoses), dermatology (rash identification and pigmented lesion evaluation for potential melanoma), and ophthalmology (retinal vessel examination to predict the risk for diabetic retinopathy and cardiovascular disease).
Uses In Everyday Medical Practice
On the TV show House, one doctor’s genius trumps the expertise of his colleagues, implying that if all physicians were as smart as Dr. Gregory House, diagnostic enigmas would all but disappear along with unnecessary deaths in hospitals.
In reality, the biggest difference between physicians is not their level of intelligence, but (a) how they approach patient problems and (b) the health systems that support them. And because “a” and “b” combine to create wide variations in clinical outcomes nationwide, machine learning offers great hope for the future.
Two AI approaches, both currently available, could radically improve physician performance.
The first is natural-language processing, a branch of AI that that helps computers understand and interpret human speech and writing. This software can review thousands of comprehensive electronic medical records and elucidate the best steps for evaluating and managing patients with multiple illnesses. The second approach involves using computers to watch (and learn from) doctors at work.
In San Francisco, Adrian Aoun is using his background in artificial intelligence (AI) to explore how machines can learn from skilled clinicians in real-time.
Rather than extracting and analyzing data retrospectively (after doctors populate their medical records), Aoun’s primary care startup Forward is using AI to follow what doctors do, step-by-step. With touch-screen data entry and voice recognition, Forward’s computers record and analyze how the best physicians achieve superior outcomes. The results benefit their colleagues and their patients.
If all physicians matched the performance of the top 20% nationwide, patient deaths from cancer, infection and cardiovascular disease would decrease by the hundreds of thousands each year.
Unfortunately, the biggest barrier to artificial intelligence in medicine isn’t mathematics. Rather, it’s a medical culture that values doctor intuition over evidence-based solutions. Physicians cling to their independence and hate being told what to do. Getting them comfortable with the idea of a machine looking over their shoulder as they practice will prove very difficult in years to come.
Understanding The Hype And Fear Of AI
Startups and tech firms have hopped all aboard the AI hype wagon, promising a host of sophisticated new solutions from nurse-bots to “AInsurance” (insurance powered by AI) to AI wearables for the elderly, to name a few. Most are interesting but not transformative. In general, they are algorithmic and not true machine-learning approaches. Nearly all have failed to move the needle on quality outcomes or life expectancy.
For every entrepreneur hyping AI as the next big thing in medicine, there are many who fear machines will replace (or even turn on) humans. I believe these fears are grounded more in science-fiction than reality. It’s true that computer intelligence is advancing faster than human intelligence. But this development offers far more opportunities than dangers.
If we see computer speeds double another five times over the next 10 years, machine-learning tools and inexpensive diagnostic software could soon become as essential to physicians as the stethoscope was in the past.
At the same time, we need to accept a difficult truth: If technology is going to improve quality and lower costs in healthcare, some healthcare jobs will disappear. According to one study, Artificial Intelligence is set to take over 47% of the U.S. employment market within 20 years. Though blue-collar jobs have long been in technology’s cross hairs, doctors and other health professionals are starting to feel the pressure, too. Unfortunately, that’s the nature of progress. What improves lives and lowers prices for many will negatively affect those who benefited from the old model of success. Uber and Lyft’s impact on the taxi industry is one obvious example. Robotics in manufacturing is another.
Without question, the role of the physician will change in the future. Fortunately for doctors, however, computers have yet to demonstrate the kind of empathy and compassion that millions of patients rely on in their medical care.
The Promise And Potential Of AI In Medicine
I expect entrepreneurs and businesses will continue to invest in AI applications, and hype them more and more. Indeed, machine learning has the potential to take medicine far beyond what it’s capable of today.
Evidence of this fact can be found in an ancient Chinese game invented more than 2,500 years ago.
“Go,” a two-player board game in which opponents try to claim the most territory, is incredibly complex and abstract, with a seemingly infinite set of possible moves. Its degree of difficulty left few observers believing that a computer could ever best a competent human. That myth was shattered in 2015 when AlphaGo, a program created by the Google Deepmind division, bested Lee Se-dol, one of the world’s top players.
What’s most interesting, though, is how AlphaGo went about it. Unlike IBM’s Deep Blue, which defeated chess champion Garry Kasparov in 1997, AlphaGo didn’t “learn” by studying humans and replaying prior matches. According to an article in Nature, humans may have taught AlphaGo the rules, but the program mastered the game by playing against itself.
This type of “deep learning” could be the very thing that catapults American healthcare into the future—helping to clarify the best care approaches, creating new approaches for diagnosing and treating hundreds of medical problems, and measuring doctor adherence without the faulty biases of the human mind.
These kinds of advances will come sooner to medical organizations that are integrated, capitated and technology enabled. I predict these organizations will embrace algorithmic solutions on smartphones or tablets first, followed by pattern recognition software and, finally, machine-generated best practices for individual patients.
Over time, patients will be able to use a variety of AI tools to care for themselves, just as they manage so many other aspects of their lives today. It may not happen soon. After all, efforts to produce self-driving vehicles date back to the 1950s. But sometime in the future—more years than entrepreneurs would like and fewer years than most doctors hope—AI will disrupt healthcare as we know it. Of that we can be sure.