Making Radiology Antifragile

Introduction

Physicist Richard Gott correctly predicted that the pyramids of Egypt would last longer than the Berlin wall. His insight, called the Lindy effect, was that the longevity of nonperishables, such as stories, ideas, and innovations, depended on how long they have been around. Simply put, the longer something has been around, the longer it will be around. This tautology belies the fact that time filters bad ideas. Though not everything new will be discarded, as all ideas that stood the test of time were once novel, the novel is more perishable than the antediluvian [1].

Examples of Lindy abound. Although tires are becoming stronger, the central idea of the wheel remains. E-readers entice us by simulating natural reading, such as turning pages. The classics are timeless because their central themes keep repeating. The Great War was, in fact, a prototypical Greek tragedy. Even thoroughly discredited practices such as bloodletting are not dead. Bloodletting’s central idea—the removal of “bad” blood—is present in therapies such as dialysis and plasmapheresis.

Central Ideas of Radiology

Newer specialists, such as radiologists, must understand Lindy so that they do not abandon their central themes. Before imaging, physicians diagnosed diseases such as tuberculosis by listening. Imaging showed the disease in action, or active disease, imposing the dictum “seeing is believing” in medical practice. With time, radiology has changed, but the demand for “seeing” remains, though the methods of rendition have evolved. Today, 3-D printing and advances in postprocessing give surgeons maps for complex surgeries.

The original radiologists, who predated the appellation “radiologist,” were clinicians with interest in imaging. Radiology’s birth fused biology and physics conceiving bioengineering. Even as radiology became its own specialty with its own training program, radiologists sat at the table of patient decision making, thus becoming the “doctor’s doctor.” Over time, distances between doctors and the doctor’s doctors increased. CT images can now be acquired in El Paso, moved to servers in Chicago, and interpreted in Bangalore. Teleradiology has increased imaging’s reach, but in-house radiologists are not condiments. They are central to the idea of radiology.

Wisdom by Subtraction

Knowledge grows by confirmation and refutation. We favor confirmation, but refutation is also important. To know what something is, it is valuable saying what it is not. Radiologists’ undervalued role is in discarding diagnoses. To discover what the patient has, it is important knowing what the patient does not have. Radiologists often discard diagnoses, but few congratulate radiologists who correctly say what patients do not have. All effort is directed toward nailing the diagnosis. The result is that radiologists overfit at scale or resort to lengthy differentials in attempts to confirm. Refutation, which is akin to chiseling marbles to make statues, should be accorded the same intellectual status as confirmation.

Another consequence of our fixation with confirmation is that imaging information has become the gateway to therapy. We laud CT angiograms that detect occlusions in patients with stroke, selecting them for thrombectomy. The imaging that leads to therapeutic inaction is less celebrated but is just as valuable. As therapies become more complicated, therapeutic index narrows, and our role in deselecting patients for treatment becomes more valuable. To give refutation primacy in decision making, refuting must actively be encouraged. Radiologists instinctively learn, for instance, the features that make lung nodules more likely malignant, as we are wired not to miss cancer. A similar pedagogic importance should be given to learning features that make lung nodules more likely benign.

Suitcases on Wheels

Futurists may view technological progress as an ever-rising apotheosis of sophistication. This has certainly been the case with CT scanners. Competitors have built scanners with increasing detectors leading to “detector arms race.” The progress has been incremental, the benefits real, but the returns are now diminishing. However, revolutionary progress need not always be predicated on state-of-the-art engineering. Take suitcases with wheels. When they arrived, they revolutionized travel by dramatically increasing comfort. Yet, this contraption was an application of common sense and hardly an innovation worthy of science fiction writings. What is astounding is that the suitcase on wheels arrived in the market after the lunar landings!

The wheel is a metaphor for simplicity, and in imaging simplicity, at least in terms of size, is essential for portability, which is essential for access. The tendency to obtain the highest signal-to-noise ratio, regardless of hardware size, should be replaced by a tendency to obtain the highest signal-to-noise ratio for the hardware size. We are already seeing these changes with ultrasound and x-ray technologies, which are now so small that they are literally handheld. Similar trends need to be seen with CT and MRI.

Variability—Mistaking Signal for Noise

Inspired by the automobile industry, health care standardized processes, which sprouted tomes of protocols to make medicine emulate structural engineering. In radiology, reporting has been standardized so that radiologists use a common language to describe abnormalities. Guidelines have been issued on how certain abnormalities such as lung nodules should be followed. Ostensibly, the goal is that whether in rural Texas or academic medical centers in Boston, radiologists issue the same quality reports.

Standardization’s allure is understandable, but in fighting variability we miss the point of variability. Diagnostic medicine, unlike building cars, possesses a dollop of uncertainty. Standardization tries domesticating uncertainty, but far from taming uncertainty, it gives the impression that uncertainty does not exist and merely creates faux precision. The much-maligned interobserver variability often signals that the abnormality is at the limits of visual perception and says as much about the “abnormality” as it does about the observer.

Variability contains information. Consider the detection of subsegmental pulmonary embolus (SSPE) on CT, with which interobserver disagreement is as high as 45% [2]. The engineer would consider variability a wart to be excised and would create protocols to ensure radiologists agree with each another. But uniformity does not address variability’s cause. Variability is a symptom—a symptom of deeper epistemic limitation. Emboli in small arteries resemble mixing artifacts, and forcing consensus does not change the fact that the presence and absence of small clots in small vessels look similar. Furthermore, SSPEs are often an overdiagnosis, and seeking and treating them can cause net harm. Far from being noise, variability signals that SSPEs are clinically insignificant [3].

Efficiency

Optimized systems, in which radiologists are working at full capacity and machines are cranking studies at all hours and patient throughput is maximized, are administrators’ dream. However, efficient systems are fragile because they fly too close to the wind. Systems should have built-in redundancies, so that they do not collapse during crunch times.

The fee-for-service payment system, which rewards volume of imaging and encourages efficiency to maximize productivity to maximize reimbursement, has been radiology’s strength, but also its weakness. The field, susceptible to reimbursement cuts, is in an unenviable rut. When reimbursements drop, we increase efficiency to maintain revenue, which begets more reimbursement cuts, which lead to more efficiency. Aside from the fact that efficiency has limits, an overoptimized system is more prone to external economic shocks, such as the decline in imaging demand during the recent pandemic.

Fragility in Action

The recent shortage of CT contrast medium is a case study in fragility. General Electric manufactured contrast solely in China. The single country setup reduced information costs and enabled economies of scale. The cost savings were partly passed on to the buyers. Many hospitals are encouraged, presumably by being offered lower rates, to sign exclusive vendor contracts. A series of entities had all their eggs in one basket, so to speak, for reasons that made financial sense. However, the world is interdependent, and our economic fortunes are influenced by other countries. When the Chinese government closed production plants to achieve zero COVID-19, contrast’s supply chain was suddenly interrupted and panic ensued in US hospitals [4].

The contrast fiasco also testified to radiology’s resilience. It turned out that contrast was not vital for many clinical situations it had been used for, such as in patients with acute abdomen. When pushed, radiologists selected situations in which contrast was not necessary. The information provided by contrast was often redundant and often could safely be foregone.

Conclusion

We must identify our field’s vulnerabilities. Often, system’s strength becomes its nemesis. For radiology to become antifragile (ie, grow with stress), it should strive to be less efficient and should tolerate variability. Importantly, it should have a framework to understand fragility.