Reasoning during the COVID-19 pandemic

by Dan Minter

The other day, I had an experience which really shook my confidence.

While on the cardiology service, we admitted a 60-year-old man with new atrial fibrillation and rapid ventricular rate. It started off like any other cardiology admission that we’d had during that month. We walked down to the emergency room and reviewed his chart. No prior history of atrial fibrillation – just hypertension and diabetes. 

Then things started to get a little bit different… His chest x-ray had some peripheral infiltrates, and the emergency department providers ordered a CT scan. This demonstrated patchy, geometric, ground glass opacities bilaterally in the peripheries, raising concern for COVID-19. Despite not having fever, cough, shortness of breath, or any symptoms, his COVID-19 test returned positive within hours. 

Thinking back on this case, I was completely unsettled. Had the ED not ordered a chest CT, would I have tested this patient for COVID-19? Would I’ve been able to reason my way to his underlying diagnosis from the initial data?

I’m a huge fan of the diagnostic reasoning process. I find it awe-inspiring to listen to clinicians pick up on subtle clues and use both intuitive and analytic reasoning to reach a final diagnosis. Yet, the COVID-19 pandemic has really made me question the role of reasoning during this crisis. Decisions have been dichotomized to ”COVID versus not COVID,” and people that have displayed none of the cardinal respiratory symptoms are being diagnosed with the disease. I found myself wondering why it has been so difficult to effectively reason during this pandemic and what we can do differently during these uncertain times.

An essential part of the diagnostic process involves creating a problem representation (PR) and then trying to match that PR to a specific illness script. Largely based on intuitive or associative reasoning, we are able to rapidly match symptoms like fever, stiff neck, and neutrophilic pleocytosis to diagnoses as specific as acute bacterial meningitis. But, this requires an adequate understanding of the syndrome and having well-defined illness scripts. 

Part of the problem is that we are still figuring out this illness. Early case definitions were remarkably narrow and included epidemiologic links to Wuhan as well as features of viral pneumonia, yet almost daily there are reports of new non-respiratory manifestations. The patient I saw with atrial fibrillation and RVR likely wouldn’t even have merited testing in early March, yet he clearly had the disease in mid April. From “COVID toes” to dysregulated coagulation to neurologic manifestations, we are rapidly learning that this is not just a respiratory disease. The illness scripts that we may have had a few weeks ago are now woefully narrow and may miss many cases.

The second major factor complicating our diagnostic reasoning during this time is the dramatic shift in prevalence of the disease in the community. As described by my friend Jack Penner elsewhere in this blog series, a core tenet of analytic reasoning involves interpretation of the pretest and posttest probabilities of a disease. The pretest probability is composed of the overall concordance of the clinical data with a specific illness script as well as the prevalence of that disease within the community. Many parts of the country are currently experiencing dramatic surges in cases, with entire hospitals in New York being dedicated to the care of COVID-19 patients. The sheer prevalence of this condition raises the pretest probability dramatically, thereby diminishing the discriminatory effects of our tests and makes COVID-19 a realistic possibility in almost any presentation (whether it is causal or just incidental).

Persistent questions around who to test and how to test them remain and appear to be updated on virtually a daily basis. Should I be using oropharyngeal or nasopharyngeal swabs? And what about all these reports of patients with abdominal complaints who ultimately are found to have COVID-19 based on incidental CT findings in the lower lung fields. 

Parenthetically, the question around CT for the diagnosis of COVID-19 is interesting. Many of the early studies highlighted the use of chest CT in the diagnosis of COVID-19. However, this too has been called into question, with the American College of Radiology now recommending against using CT scans for aiding with a diagnosis of COVID-19, while The American College of Chest Physicians takes a more favorable position in their recent guidelines.

In the pre-COVID world, we were taught to be judicious with our testing. Simply running somebody through a CT scanner in a non-hypothesis driven fashion raises the likelihood of a false positive or incidental finding which could have been clinically inconsequential to the patient. Yet, this calculus has changed. Sure, I don’t really think this patient has COVID-19, but what if I’m wrong? I could expose dozens of people by not testing – from the dedicated bedside nurse to the person delivering the food. This element of risk is seeping into the way we think about testing

So, is there room to reason during this pandemic? We have established that the sheer prevalence in many areas is confounding our Bayesian calculations, that our current illness scripts are likely inadequate, that many are still confused about the most up-to-date testing algorithms, and that the risk of missing a case is altering how we approach testing and isolation.

Perhaps, we just need more time. Already, our understanding of the disease has rapidly progressed in the few short months since the pandemic began. As we test more people, including those who are asymptomatic, we will slowly begin to move past the tip of the iceberg and truly see how much of the population is infected. Additionally, our illness scripts will evolve and become far more nuanced. Until we have this greater understanding, however, taking a more diagnostically humble and conservative approach is likely an appropriate response. 

Luckily, our patient did well and was able to be discharged to home quarantine. While I still wonder if I would have been able to reason my way to the diagnosis from the initial data, the issues raised by this case are formidable. We’re in a really uncertain time, facing a poorly understood virus the world has never seen before. Yet, managing uncertainty is the “name of the game” in diagnostic reasoning. By recalibrating our risk threshold, testing more broadly, and maintaining a sense of diagnostic humility, we’ll gradually learn more about this disease and, hopefully, become better reasoners in the process.