Many rheumatic diseases, including PMR, seronegative RA, sarcoidosis, CNS vasculitis, FM, and Behçet's syndrome, have been described as 'diagnoses of exclusion'. While some may disagree about the particulars—PMR, for instance, may also be diagnosed by pattern recognition and response to treatment—all of us have encountered it before. Like other 'pseudo-probabilistic aphorisms', the phrase 'diagnosis of exclusion' suggests a preceding thoughtful diagnostic evaluation. Simply invoking it conveys the gravitas of Sherlock Holmes to trainee presentations, clinical documentation, and innumerable UpToDate articles. But should it? In our experience, a 'diagnosis of exclusion' is subjective and conditional, which misaligns it with principles of diagnostic reasoning, resulting in unnecessary testing and premature closure.
Invoking the 'diagnosis of exclusion' aphorism immediately begs the question, 'to the exclusion of what?'. In rheumatology, a prespecified list of all possible 'exclusions' does not exist. If asked to provide an exclusionary list before arriving at Behçet's syndrome, for instance, five rheumatologists may opine seven different conditions. If asked 2 weeks later, the same group of rheumatologists may well provide different lists entirely. This is because any list of 'exclusions' will necessarily be subjective and conditional. Such lists are subjective because a list of exclusions depends upon a combination of prior experience, cognitive dispositions, and physician biases, which vary greatly from rheumatologist to rheumatologist. They are conditional because they depend upon the patient's presentation. A patient presenting with oral ulceration and fatigue may invoke a different set of illness scripts for exclusionary diseases as compared with a patient presenting with oral ulcerations and retinal vasculitis.
Digging deeper reveals greater problems for both the diagnostic process and the patients it serves. For the sake of argument, let's imagine that we could all agree upon appropriate lists of 'exclusions' for the aforementioned diseases. What is the sensitivity and specificity of a rheumatologist deciding that the list had been 'excluded?' This question cannot be answered because diagnostic tests have not been designed to function this way. We can say with confidence that a negative test decreases the likelihood of ANCA vasculitis. We cannot say how that same test influences the likelihood of Behçet's syndrome or sarcoidosis, because diagnostic testing has not been designed for a reductionist process. Diagnostic tests increase or decrease the probability of a disease for which the testing has been studied. Outside of this context, we are misapplying basic principles of diagnostic reasoning.
Beyond theoretical concerns, approaching medicine using a 'diagnosis by exclusion' framework results in overtesting. When I was in training, I ordered a RF and ACPA as part of a work-up for a patient who presented with signs of acute Löfgren syndrome. When the RF result came back elevated, we instructed the patient to ignore the finding. Why did we send it, then, other than to satisfy our list of 'exclusions' for a diagnosis of sarcoid arthropathy? RA could not explain the patient's hilar lymphadenopathy, erythema nodosum, or the tattoo-adjacent rash that had recently been biopsied and found to have non-caseating granulomas. More commonly, we have all experienced the consult to evaluate a patient with fatigue and a positive ANA, which was ordered to 'exclude' lupus as part of a FM work-up. The unnecessary testing incurs costs and engenders confusion for patients, who end up 'paying the piper' to satisfy our need to exclude.
Conversely, the 'diagnosis of exclusion' framework may result in diagnostic error. I once inherited a patient with 'CNS vasculitis' that had been diagnosed 'after all other possibilities were excluded' and started on high-dose steroids and CYC. The list of 'exclusions' did not include CNS lymphoma or a biopsy as part of the initial work-up. After many relapses, he ultimately underwent a biopsy and began appropriate treatment for CNS lymphoma. In a more recent example, a patient with 'PMR' because 'alternatives had been excluded' was later found to have TIF1γ DM and oesophageal adenocarcinoma. In both cases, diagnostic momentum and premature closure ensued and the subjective list of 'exclusions' had been inadequately applied, despite plausible alternative diagnoses.
Even in cases where the 'diagnosis of exclusion' framework has resulted in an adequate workup, it closes the imagination to reasonable future possibilities. For cases of FM, for instance, the 'diagnosis of exclusion' suggests that early seronegative RA has been 'excluded' and therefore should not be considered again. On the contrary, we have all seen cases where presumed FM later proved to be an inflammatory arthritis. Similarly, a diagnosis of PMR after 'exclusion' of GCA suggests that GCA has been 'excluded', when a more appropriate approach would be to reassess for GCA at every visit. By its very nature, the word 'exclude' is paralyzing, conveying a degree of negative certainty that invites cognitive biases.
But who could possibly replace Sherlock Holmes? We recommend emulating Thomas Bayes, the patron saint of the probabilistic thinking that underpins diagnostic reasoning (Table 1). Bringing Bayesian reasoning to clinical medicine requires a different—and ultimately more patient-centric—set of skills. These skills include expertise in epidemiology and narrative medicine, which allow the diagnostician to formulate a pre-test probability that the person before them has a disease. They include knowledge of the performance characteristics of physical examination findings and subsequent testing and an ability to use this knowledge to arrive at post-test probabilities. Most importantly, this perspective encourages rheumatologists to become proficient in communicating risk, embracing uncertainty, and considering alternative diagnoses.
It is time to exclude 'diagnosis of exclusion' from the rheumatologic lexicon. The aphorism is subjective and conditional, has resulted in overtesting and diagnostic error, and runs counter to principles of diagnostic reasoning. Embracing the uncertainty of a Bayesian may be difficult, but diagnosing rare and amorphous diseases has never been 'elementary'.
Data availability statement
Data will be made available upon reasonable request.
Rheumatology. 2023;62(1):1-2. © 2023 Oxford University Press