Radiology Fast and Slow
By J. Raymond Geis, MD
Nobel Prize in Economics laureate David Kahneman, in his best-selling book Thinking Fast and Slow, describes two different thought systems we use daily. System 1, as he calls it, is the fast, instinctual decision making approach we use most of the time. System 2 is what we use when we stop and evaluate things in a methodical, logical manner.
After reading perhaps 250,000 cases over the years, I suspect I’m like most radiologists who read the majority of exams in System 1. As Dr. Kahneman’s research dramatically shows, however, this approach is regularly affected by cognitive bias, which distorts my perception and interpretations in ways that vary from accepted best practice.
Even when I know this, it’s hard to avoid bias. I’m prejudiced by the most recent unique case I dealt with, like the “exception” case of a 4 mm lung nodule in a young person that turned out to be cancer, despite statistics demonstrating the chance of that occurring as less than 1%. For a while after that I’m inclined to recommend short term CT follow up for any 4mm nodule, and to heck with standard Fleischner Society recommendations for what to do with such a nodule, which in this instance recommends do nothing.
I’m not alone. A 2010 RSNA article by Eisenberg, Bankier and Boiselle showed only 77% of radiologists even knew about the Fleischner criteria, and between 34% and 60% of radiologists followed them correctly. A 2011 AJR article by Feely and Hartman showed in one practice the criteria were followed correctly 52% of the time.
Even if I want to follow them, I don’t remember them. Each of our reading rooms has a bulletin board, and somewhere on it is a photocopy of the Fleischner Society recommendations. On a busy day on the assembly line of reading exams as fast as possible, finding that piece of paper among all the other recommendations tacked to the board takes time and cuts into my productivity. The urge is strong to assume I remember, give a recommendation, and move on to the next case.
A big advantage of computers is they don’t operate in System 1. They are less susceptible to emotion and cognitive bias. Give a computer Fleischner Society criteria and it will apply them consistently and reliably forever.
Here’s a thought. How about for my chest CT report, I put in all the findings, then turn off the mic and let a computer do the rest? Using natural language processing (NLP), the computer takes the patient’s age, relevant history and smoking history from the EMR, combines that with the findings I dictate, and provides interpretation and recommendations based on accepted criteria.
That scenario is a win all around. I will be more efficient, get through cases faster, and either get home for dinner with the family or make more money. Report recommendations would be more consistent and more accurate, so quality improves.
Dr. Eliot Siegel, a leader in imaging informatics, gets knowing laughs each time he shows the film clip of Lucille Ball on the chocolate factory assembly line, and her struggles to keep up with the increasing load. Radiologists are high-paid assembly line workers, just like Lucy. The future of our radiology assembly line will probably be similar to the evolution in other industries, like cars. Photos from Henry Ford’s original assembly lines show hundreds of people on the line. Modern automobile assembly lines instead have hundreds of robots, with a few highly trained people managing the process.
Efficiency, standardization, and quality assurance are goals of any industrial process, including medical imaging. Except in the rarest of cases we need to eliminate cognitive bias in radiology exam interpretation, to get away from the pitfalls common in Kahneman’s System 1 decision making style we use most of the time. Perhaps it is time for imaging informatics to provide tools to do that.