SMCMA Physician

The Promise of AI in Healthcare

In the classic Western The Magnificent Seven, a small village of peace-loving and naïve people hires a crew of seven gunfighters to protect them from the marauding bandits who have been terrorizing them. Now, I haven’t seen this movie in probably forty years, but one scene still sticks with me. In it, one of the gunfighters, played by Charles Bronson, is training one of the villagers to shoot a rifle. He keeps trying, but the villager just doesn’t get it. At one point, Bronson’s character loses his patience, takes the gun away from the villager, and says, “I’ll tell you what: don’t shoot the gun. Take the gun like this, and you use it like a club. All right?” He holds the gun by the barrel and demonstrates hitting someone over the head with the butt of the gun. 

This is a simple moment of comedy in a classic film, but it stuck with me, and I think it illustrates something important. A gun is a sophisticated weapon, but because of hurdles against using it properly and to its fullest effect, the villager is encouraged to use it as a crude instrument. I think this reflects something that often happens with new technologies: because they are unfamiliar, we use them for much simpler tasks than what they are designed for and so fail to use them to their full potential. 

This is exactly the case with information technology (IT) in health care today. While IT can be very sophisticated technology that could be used to great effect in clinical settings, we are only using it as a very blunt instrument, and it is not delivering the results we want. 

AI Impact

Over the past two decades, the technology at the heart of AI and machine learning has progressed rapidly, and these developments have great potential for applications that would be transformative for health care. Many experts in the field are already promoting some of these applications, but their focus tends to be limited to the administrative side of things. The 2017 Accenture list of the top ten AI applications in health care, for example, includes far more administrative tasks than practices that might apply to direct clinical care and practice.  

The clinical setting, however, is where AI can have the most disruptive impact in terms of both health outcomes and lowered costs. AI can not only make predictions but also provide guidance in diagnosis, treatment, and other functions of clinical decision making. In the clinical setting, AI is potentially paradigm changing. More than 95 percent of physicians already gather data on their patients via EHRs, which means there is a treasure trove of data out there waiting to be explored and exploited.

The availability of this data, together with AI technology, can lead to an explosion in the kind of medical knowledge that doctors draw on in clinical decision making. In the past, medical knowledge has been achieved by what we all recognize as traditional scientific method: a research question is formulated, a hypothesis is developed, and data is gathered from a sample study population and compared to controls in order to confirm or disconfirm the hypothesis. In the health care sphere, this data gathering step occurs by means of clinical trials, which are by necessity limited to small portions of the population in very restricted conditions. 

I’ll discuss this at greater length later, but let me put the idea in your head now that AI allows us to reverse the order of this process. In cases where the EHR is well utilized for record keeping, we have already gathered data from a sizable portion of the population. So, we can begin by presenting our questions to this already existing data set. The data manipulation and pattern recognition capabilities of AI can then be employed to generate and confirm answers to real life questions and with real-time application for current patients.

AI’s big advantage over the limitations of clinical trials is that its data set can be drawn from the records of many thousands, potentially millions, of people, as opposed to a clinical trial that has to operate with a much smaller and less random population. Because clinical trials narrow their focus to a small subset of variables in a fixed period of time, they end up taking a snapshot of one moment out of the life of each subject, including only those characteristics researchers identify as potentially relevant. AI, on the other hand, can accommodate a much bigger and richer picture of a patient’s entire life, taking into account any and all data points that have ever been gathered on that subject, including where patients live, what they eat, their socioeconomic status, their preferred mode of transportation, and so on. Health care professionals widely recognize that all of these things, including behavioral and social determinants of health, comprise risk factors, but this recognition is not reflected in the structure of clinical trials and, many times, it’s also not reflected in their specific content. AI allows us to determine the weight of these various risk factors and use that knowledge to mitigate risks in the cases of specific patients. In combination with EHR data, AI can directly inform the actual practice of medicine at the clinical level, making use of the data that is constantly being collected to improve patient outcomes, and it can help to drastically reduce health care costs. AI methodologies have the advantage of studying patients in the real-world setting rather than in a more sterile, academic clinical trial setting.

The rest of this book is devoted to showing the promise of the application of AI in various clinical settings. From the broadest point of view, AI can lead to improvements in care, improved health, economic benefits, and a more effective health care system. More specifically, the application of AI in the clinical setting has potential benefits for three primary stakeholders.

Patients

For patients, it’s simple: better health outcomes will result from getting treatments that work. Getting the right treatment right away has a few advantages. First, patients are likely to get better faster, since they will not have to go through the trial and error process. Second, adverse reactions to treatment will decrease because appropriate treatments are identified before treatment begins. Also, the number of treatment failures will decrease. Because patients won’t be getting treatments that won’t work, the cost of care will be lower. So, the part of the bill that patients must foot, out of pocket or indirectly through their employers or the government, will be lower as well. For patients subscribing to high-deductible health plans, this change would be significant, given that out-of-pocket expenditures keep rising at a much faster rate than individual incomes and faster than the economy as a whole. 

Doctors

AI is not a threat to doctors. Rather, it will help them by lightening their cognitive burdens and their workloads. They will be better equipped to act on accurate risk calculations, instead of having to perform the calculations themselves. Since humans are not as accurate in making these calculations as machines are, doctors can feel relief that the care they’re providing is less risky and more likely immediately beneficial to the patient. They will be able to devote their time to doing things that drive value for their patients and give meaning to their practice. Knowing that whatever they are doing is benefiting their patients will decrease the frustration and burnout that is so prevalent among them now. They will be working smart, not hard.

One quick example: new AI tools are coming on the market that will help physicians spend less time doing documentation and entering data manually and more time talking with the patient, interacting with the patient, understanding the patient’s situation, and developing empathy for the patient’s circumstances. To the extent that these new tools will also help mitigate and decrease the potential for bad outcomes, the guiding worry about malpractice should decrease as well.

Companies and Organizations

Finally, cost savings benefiting the bottom line are a key factor that should appeal to business leaders and employers. For providers, AI offers an accurate risk assessment and cost-benefit analysis tool for treatment options. If companies are able to decrease costs and provide care more efficiently, they will be financially rewarded. Their customer service would also improve insofar as patients would feel heard and find interacting with the system pleasurable rather than painful. Creating greater “stickiness” between patients and providers would benefit both. One note of caution, though: AI works best in a value-based care system rather than a volume-based care system.

 

Uli Chettipally, MD, MPH is an emergency physician, innovator, author, and speaker. He is also currently the Secretary/Treasurer of the Board of Directors and Chair of the Editorial Committee at the San Mateo County Medical Association.