From the farm to the Oval Office to the clinic, AI must account for context.
What do a scientific algorithm, rat poison, and a president all have in common? They played a role in setting the stage for the artificial intelligence (AI)-driven algorithms of today.
Today’s AI-fueled healthcare algorithms face challenges not found in any other industry. In retail, AI can reach across multiple channels to aggregate and integrate data for predictive analytics. But in healthcare, channels are often siloed in disparate, proprietary systems. AI holds great promise for stronger patient outcomes, higher efficiency, and cost reduction, but it still struggles to find its footing through an often-ill-defined mission and purpose.
Today’s data models need to drive the algorithms behind AI, dealing with the current chaos while still trying to navigate to a brighter future of care. For an algorithm to be effective in such a crucible, the first thing to consider is that it needs to contain much more than simple mathematical computability.
What really makes an algorithm special is an understanding of the context in which it will operate: where, when, and how. Comprehension of context is the key to understanding what needs to be accomplished with an algorithm and ultimately will guide its effective design and development.
To present a clear picture of context, it’s helpful to examine how a natural algorithm has played out in a healthcare setting. This can show us how the same core algorithm evolved and manifested itself in different contexts and provided value differently at different times and in different systems, until it ultimately saved the life of President Dwight D. Eisenhower.
During the 1920s, farmers in North Dakota and elsewhere found their livelihood at risk from inexplicable cattle deaths. Cows undergoing minor procedures like castration were hemorrhaging to death, and miscarriages in cattle were on the rise. Two veterinarians named Lee Roderick and Frank Schofield identified moldy hay as the cause and found that negative health effects could be reduced by avoiding it and providing fresh blood transfusions. As such, farmers steered clear of sweet-smelling clover hay, a tradition that continues to this day. In 1932, Roderick concluded that the pathology induced by this hay was related to a reduction in “prothrombin” activity in the blood.
But that did not go far enough. In 1933, a farmer frustrated with cattle deaths in his farm doubted the diagnosis of Sweet Clover Disease and brought the problem to the biochemistry lab of Karl Paul Link, PhD, who identified the blood-thinning compound known as dicoumarol. Dicoumarol arises when fungus reacts with a substance called coumarin in the hay. Dicoumarol-like compounds were shared with clinicians treating patients susceptible to blood clots, but there was a problem. Patients needed an antidote to limit the blood-thinning characteristics of dicoumarol, otherwise it could prove fatal, just as in the farm animals.
Link believed that this compound could be part of a “geometric” algorithm that would be the key to anticoagulant treatment in human patients. He explains it beautifully: “It seems appropriate to me to visualize successful anticoagulant therapy with the dicoumarol-type drugs as being shaped like a triangle with accurate prothrombin assays at one corner, Vitamin K at another, and sound clinical judgment at the third. Each corner is linked to the other by way of the connecting sides. There should be no separation. Each is vitally dependent on the other two.”
However, much to the frustration of Link, Vitamin K was not administered adequately in most clinics, and the result was that dicoumarol gained the notoriety of being a dangerous drug.
It was during a prolonged illness that Link decided to study laboratory records and read the history of rodent control from ancient to modern times. On his return to the lab, Link proposed that a candidate compound be put to rodenticidal use. Compound No. 42 was promoted as rodent poison under the auspices of the Wisconsin Alumnus Research Foundation (WARF) from which the commonly used name Warfarin originated. This revolutionized rodent control and became a market leader, with these first-generation anticoagulants still available as rodenticides today.
Link understood that clinicians were hesitant to consider the use of rat poisons in their patients. In 1951, an army inductee tried to commit suicide with Warfarin using it as directed for killing rats. It turned out that while it did induce Sweet Clover Disease, it was not fatal. This led Link to make it available for clinical use in smaller quantities. And in September 1955, Warfarin had its big break. President Eisenhower had a heart attack and was given Warfarin as a treatment. The improvement in his condition popularized Warfarin use, and it has since become a standard element of anticoagulation therapy.
None of this evolution is a surprise to modern, sophisticated physicians. They know that an algorithm is not a destination on its own. They have no illusions that simply building an algorithm will transform care. Inherently intuitive clinicians are more concerned about how an algorithm is implemented than the mere fact of its existence. When the medical community understands the context of why AI-based algorithms are being inserted into the process of care, then acceptance, adoption—and ultimately clinical results—become much easier to achieve.
Sawad Thotathil, MD, clinician and senior director, Persistent Systems, directs product strategy and management for the company’s Digital Healthcare group, helping health systems achieve their digital transformation goals.
Get the best insights in healthcare analytics directly to your inbox.