Reply to Aberegg and Wolfe: Aspergillosis in the ICU: Hidden Enemy or Bogeyman?

From the Authors:

In their letter, Aberegg and Wolfe highlight the effect of disease prevalence on the performance of diagnostic tests with reference to our publication in which we provided estimates of the prevalence of aspergillosis in critically ill adults with suspected ventilator-associated pneumonia (VAP) (1). We thank them for their interest in this understudied topic and for articulating the uncertainty that is implicit in prevalence estimates when no perfect method for disease classification is available. They express understandable concern that overestimating the burden of aspergillosis in this population could lead to an epidemic of overdiagnosis and treatment.

In our publication, we emphasized the uncertainty in our prevalence estimate that arises from the definition of aspergillosis that we used; this definition balances the risks of underdiagnosis and overdiagnosis, as we set out (1). To express this uncertainty, we considered the 95% confidence limits in our main analysis, the effect of using higher thresholds for classifying BAL fluid (BALF) galactomannan (GM) as positive, and corroboration of BALF GM with serum GM as well as other Aspergillus biomarkers in both BALF and serum. Aberegg and Wolfe contend that the prevalence of this disease may be substantially lower, based on the posterior probability of aspergillosis with a positive BALF GM, in a low-prevalence population. This is certainly possible, though is not readily incorporated in our estimate because neither the diagnostic accuracy of BALF GM nor true disease prevalence in nonneutropenic patients with suspected VAP is established. They illustrate the point using an assumed disease prevalence of 1%, but this is not a robust prevalence assumption. It is correct that the majority of positive BALF GM results would be falsely positive if the disease prevalence is only 1%; by comparison, the majority would be true positives if the prevalence is greater than 8%, based on a test specificity of 95% (2).

There is no doubt that the prevalence of aspergillosis in the population we describe remains uncertain and we do not purport to have definitively established this. The dependency of prevalence estimates on the accuracy of diagnostic tests used, and vice versa, creates a circular argument that cannot be readily resolved. We acknowledge the superior specificity offered by a tissue diagnosis, which could reduce the uncertainty; however, our experience is that obtaining such material is challenging in both research and clinical practice. This, in itself, increases the risk of sampling bias, leading to error if histology is used as the basis for prevalence measurement.

We certainly do not wish for our publication to drive an epidemic of overdiagnosis. In support of this, our manuscript concluded that use of GM testing on BALF in patients with suspected VAP could highlight those for whom more extensive clinical investigation is warranted. Although overtreatment is not desirable, we are also concerned that the common assumption that aspergillosis is so infrequent as not to justify investigation in this patient group risks underdiagnosis and undertreatment. There is a difficult balance to be struck in the face of uncertainty relating to both the prevalence of aspergillosis and diagnostic test accuracy in the nonneutropenic critically ill population. Well-designed prospective studies to address this would certainly be of high value; however, other efforts to reduce uncertainty—even if imperfect—may help to guide clinical practice.