AI in healthcare: don’t bite off more than you can chew

By:

Sep 20, 2017

Oncologists are tasked with making medical decisions that require intensive medical training and experience, while caring for patients suffering from some of the most costly and least structured and understood problems in medicine. The job description of these specialist physicians lands easily within the realm of intuitive medicine—diagnosing conditions solely by their symptoms and treating them with therapies whose efficacy is uncertain. Practicing on the frontiers of modern medicine as such is not a simple or routine job by any stretch of the imagination.

Enter IBM’s Watson Health Oncology tool. The supercomputer leverages artificial intelligence in attempt to make our best medical minds even better in the area of cancer care by providing decision support and treatment recommendations for conditions which are not well understood. Yet Watson’s technology, striving to cater to the most demanding of decision-makers in medicine, is still in its infancy. A recent exposé published in STAT has highlighted shortcomings, thus far, of the AI tool in meeting the demands of oncologists.

Artificial intelligence is a potentially Disruptive Innovation relative to reliance on human intuition in analyzing and making decisions—a task that currently commands a steep price in the case of specialist physicians. So, it should not come as a surprise that Watson has not met the needs of oncologists right away. Successful Disruptive Innovations begin as technologies not “good enough” to meet the needs of a market’s most demanding users. Instead, they seek applications in which the innovation is already “good enough” to help people and companies make progress. Then, as the technology improves, they gain market share amongst more demanding people and companies, solving more complex and costly problems.

AI technologies may be under-serving oncologists in some respects, since they do not yet seem able to adequately contribute to the complex and unstructured problems of the oncology world. However, they are likely “good enough” for other applications in healthcare, when they are employed to solve more structured problems that abide by known and understood sets of rules. If AI entrepreneurs deploy them in this way, they’ll have the opportunity to build upon their capabilities and iron out the wrinkles until they are prepared to take on more complex and less structured decisions.

Taking a page out of their own playbook

The evolution of the personal computer can be an example of how IBM should look to structure their efforts. Prior to the personal computer, minicomputers and mainframes dominated the market for those seeking to take advantage of computing capabilities. Then, in the 1980s, IBM began putting Intel’s microprocessor chips into personal computers (PCs), joining Apple in delivering computing capabilities to individuals and families as opposed to more demanding and profitable customers like businesses and corporations. These new users were quite happy to have a PC despite its limited capability or performance, as the alternative was no help at all in completing tasks like typing documents and building spreadsheets. PCs then grew to dominate the computing market over the coming decades, as they became more capable in performing complex tasks, eventually upending mainframes and minicomputers among even the most highly skilled and demanding customers.

The IBM of the 1980s did not attempt to sell their PC to the most demanding of businesses and corporations for which their product was not yet “good enough,” so why are they attempting to do so in 2017? IBM and other entrepreneurs should instead construct AI technologies for more structured and understood problems that people today still struggle with, starting with the following decision-makers:

  1. Individuals and families

Artificial intelligence technologies hold endless potential in aiding the everyday health decision-making of the typical person. They could manifest themselves in the form of helping patients determine whether a condition warrants a visit to an emergency department, urgent care, or triage elsewhere—holding potential to spare patients time, fret, and unnecessary cost by improving facility utilization. On an even more day-to-day level, artificial intelligence tools could even help people with dietary restrictions avoid ordering food they may have an adverse reaction to, potentially playing a role in preventing threatening health complications.

  1. Those responsible for clinical documentation

Clinical documentation is commonly performed by physicians and nurses, and in some cases by medical assistants and medical scribes. Artificial intelligence technologies can help whoever may be documenting a patient encounter to focus on documenting only the more nuanced aspects of a visit by auto-filling more routine and rules-based sections—leaving more room for empathy in patient encounters.

Documentation of clinical encounters is integral to providing quality patient care. As patient records accumulate over time, they provide a longitudinal perspective of patient health and inform current and future health-related decisions. Unfortunately, according to a study performed by the American Medical Association, among some high-performing practices, for every hour physicians provide direct clinical face time to patients, nearly two additional hours are spent on electronic health record and desk work within the clinic day. These types of often routine clerical tasks take a toll on physicians (and nurses) and prevent them from delivering personable care at the top of their license.

Determining “good enough”

Artificial intelligence will help and empower these decision-makers, but there will always be instances in which patient cases do not align with rules. These anomalous events require human intuition and creativity to solve.

Establishing a foothold market among these decision-makers will provide AI entrepreneurs a launchpad upon which they can improve their capabilities and move upmarket to help more and more technical practitioners. Eventually, even physicians with the amount of medical training and specialized expertise as an oncologist will be able to make use of advanced AI tools, but we cannot expect to be able to reach such demanding consumers from the get-go.

For more, see:
How Disruption Can Finally Revolutionize Healthcare

As a Research Associate, Ryan investigates potentially disruptive healthcare delivery models and the technologies that will enable their success. He is particularly interested in health information technology and is currently researching Disruptive Innovation in the space of electronic health records.