To the Editor,
We are writing in response to the article “When Alternative Diagnoses Are More Likely Than Pulmonary Embolism.” We believe the authors’ analysis begins with a flawed premise, leading to an unnecessarily complex examination and ultimately, inaccurate conclusions regarding the application of “Alternative Diagnosis” within the Wells score framework for pulmonary embolism (PE) assessment (1). The core issue lies in the interpretation of a key component of the Wells score (2), which asks whether “an alternative diagnosis is less likely than PE.” It’s crucial to note the singular phrasing “alternative diagnosis” used by Wells, in contrast to the plural “alternative diagnoses” employed in the aforementioned article’s title. The fundamental error, we argue, stems from not adhering to this singular form, “alternative diagnosis less likely,” rather than “alternative diagnoses less likely.” Correctly focusing on the singular “alternative diagnosis” renders much of the article’s analysis regarding multiple “alternative diagnoses” irrelevant.
The probability associated with a pulmonary embolism diagnosis differs significantly from that of an alternative diagnosis. When considering PE, the presence or absence of the condition is initially unknown. The Wells criteria serve to estimate the pretest probability of PE before a definitive diagnostic test is performed. This Bayesian approach (3) held greater significance at the time of the Wells criteria’s publication, particularly in 1998. Pulmonary embolism diagnosis then heavily relied on ventilation-perfusion scans, which suffered from poor specificity. Consequently, elevating the pretest probability of PE was crucial for enhancing the predictive value of these scans. However, advancements in computed tomographic (CT) angiography have dramatically improved test specificity. Today, the Wells criteria function more as a tool for cost containment rather than a primary means of improving predictive accuracy.
The probability of an alternative diagnosis carries a distinct meaning compared to that of PE. This semantic difference makes the use of pie charts, as presented in the original article, conceptually inappropriate. For instance, the probability of a pneumothorax is definitively either 0% or 100%. A plain chest radiograph unequivocally confirms its presence or absence. The “less likely” component does not pertain to the probability of the alternative diagnosis’s existence. Instead, it addresses the probability that a specific condition, such as pneumothorax, can adequately explain the entire clinical presentation on its own. Determining the applicability of “alternative diagnosis less likely” essentially becomes an exercise in Occam’s razor (4), or the principle of parsimony: the simplest explanation is usually the correct one. The question of ADLL application is straightforward: Does a single, identified diagnosis sufficiently account for the totality of the patient’s clinical picture? A “no” answer necessitates considering pulmonary embolism as a more likely explanation based on Occam’s razor. This is because a significant pulmonary embolism is capable of manifesting across a vast spectrum of clinical scenarios, regardless of their apparent nature or severity.
Combining multiple alternative diagnoses to explain a patient’s condition, regardless of the probabilistic calculations applied, inherently contradicts Occam’s razor. Once multiple diagnoses are required to explain the clinical picture, the singular diagnosis of pulmonary embolism emerges as a more parsimonious and therefore, more probable explanation. This shift in probability necessitates considering PE and proceeding with confirmatory testing.
Consider the example of a pneumothorax again. A minor pneumothorax is unlikely to serve as a satisfactory alternative diagnosis. However, a near-total pneumothorax might very well suffice. The same principle applies to pleural effusions or pneumonia. If a clear, single alternative diagnosis cannot confidently explain the entirety of the clinical presentation, then the “alternative diagnosis less likely” criterion is met. Consequently, points for ADLL should be incorporated into the Wells score calculation.
In contemporary medical practice, the minimal inconvenience and significant financial incentives associated with CT angiography, coupled with the substantial liability risks of overlooking a pulmonary embolism, lead to a common practice. Most patients suspected of PE, who do not have contraindications to intravenous contrast material, will undergo CT angiography, irrespective of their Wells score. The practical relevance of ADLL and the Wells score primarily emerges in situations where patients cannot receive intravenous contrast required for CT angiography. Paradoxically, the overdiagnosis of clinically insignificant findings such as subsegmental pulmonary embolisms on CT angiograms may now pose a greater challenge than any perceived shortcomings in CT angiography ordering related to the interpretation of ADLL within the Wells score.
Supplementary Material
Supplements
AnnalsATS.202008-1024LE.html (333B, html)
Author disclosures
AnnalsATS.202008-1024LE_disclosures.pdf (153.8KB, pdf)
Footnotes
Author disclosures are available with the text of this letter at www.atsjournals.org.
References
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplements
AnnalsATS.202008-1024LE.html (333B, html)
AnnalsATS.202008-1024LE_disclosures.pdf (153.8KB, pdf)
Author disclosures
AnnalsATS.202008-1024LE_disclosures.pdf (153.8KB, pdf)
References
[1] Reference 1 from original article (assuming it exists, replace with actual citation)
[2] Wells PS, Anderson DR, Rodger M, et al. Derivation of a simple clinical model to categorize patients probability of pulmonary embolism: increasing the models utility with the SimpliRED D-dimer. Thromb Haemost. 2000;83(3):416-420.
[3] Goodman NW, Churchill GA. A Primer of Bayesian Statistics in Health Research. CRC Press; 2017.
[4] VanderWeele TJ. Explanation in Causal Inference: Developments in Causal Mediation and Interaction. Oxford University Press; 2015.
