BLOG: Red light, yellow light, green light medicine
For all of the hype around artificial intelligence in health care, it has actually had little impact on our day-to-day practice of medicine so far.
While most applications of this promising technology have not yet surfaced in commercial products, we can certainly ponder some ways AI will influence us as we undergo some megatrends that are beginning in medicine.
The first megatrend that will welcome AI is the need to reduce the cost of medical care. The complexity of taking care of an older, sicker population is right now colliding with the need to reduce costs. In the days of fee for service, doctors and hospitals were motivated to run toward the patient, delivering more diagnostic tests and more ancillary services with the assumption that more is better. As we move toward at-risk, capitated contracts, we will be motivated to do the opposite — to step away from the patient and deliver the same result with fewer services at lower cost. To do this, we need AI to scour our records and identify variables that require less intensity of care but can accurately predict a good or poor outcome. According to industry veteran Jim Mazzo, whom I interviewed recently for a Healio video, “For each patient, we need a status of red light, yellow light or green light to guide our future care.”
Take, for example, a patient with macular degeneration. AI studies might find that patients who have three consecutive visits with a stable macular thickness on OCT and a uniform pattern of injection treatments qualify for a “green light” and do not need to return for 6 months, cutting the frequency of visits in half and eliminating an extra OCT. A “yellow light” might occur if that same patient did not have a stable macular thickness, requiring maintenance of more frequent visits. And a green light could turn to a red light if an otherwise stable patient had the sudden onset of a change in vision like metamorphopsia, necessitating a physician’s prompt and direct intervention.
AI systems could be built into our EHR, tying together data from diagnostic instruments and physical exams to come up with a recommendation for next steps in patients with common diagnoses.
In other words, AI would codify patterns of treatment that we already practice but embolden us to stretch out the frequency and intensity of services based on studies of similar, large populations. And while we physicians sometimes bristle at unofficial standards of care being imposed upon us, we would also benefit, both legally and personally, from the virtual second opinion that an AI system could provide.
I believe an important part of AI’s future in medicine — and a megatrend of its own — will be the increased use of structured psychometrics. That is, accurate history collection using standardized symptom measures. These are a far cry from our current, imprecise methods of collecting history, and they could give us a completely new view of the diseases we treat. We will explore that topic in the next installment of this blog.
Meanwhile, we can wonder at the sophistication of Alexa, the AI device that helps us create grocery lists, set timers and play guessing games, and hope that a similar power will someday help us take better care of patients.
Collapse