My shot...at revising ILD histopathology criteria
I had to do it. My Yale ancestors created UIP/DIP and the like (although they got those largely wrong, see my previous blog posts). So it is now my turn.
While Tom Colby long ago relegated the wedge biopsy from gold to a silver standard, such biopsies are still considered potentially helpful. But how do we assess even a silver standard? A few years ago, I felt frustrated that the biopsies I was looking at seemed difficult to classify and wanted additional input. I recruited an international group of pathologists with interest/expertise in ILD to review a series of cases using both fixed and whole slide images to see if we could improve our criteria. We have recently published this work (no paywall).
For the truly curious, all the images used in this study are available online.
The approach we took paralleled that taken by Carol Farver (a member of this group) in the lung transplant field (and years before by Juan Rosai in breast biopsies). I generated specific fixed images of ILD and asked questions based on various ATS pathology criteria. Even with simple fixed fields, there were notable differences of opinion. Two areas that seems particularly difficult were the borderline between scattered giant cells and poorly formed granuloma and presence of dense inflammation away from scar.
This was also reflected in a concurrent review of WSI of actual cases. I chose both cases which I felt were straightforward (organizing pneumonia, diffuse alveolar damage, eosinophilic pneumonia, smoking related ILD, a few classic UIP cases) and cases I felt were harder (other cases of fibrotic ILD). Gratifyingly, cases I thought were straightforward had high concurrence. The rest were more problematic, as predicted. Pathologist confidence in diagnosis is a real thing. So far, we were just replicating well known and unsurprising results.
We then attempted to generate more robust criteria via a Delphi like process using rounds of internal discussion. This was not data driven but based on experience of the participants. Note that the ATS criteria are not data driven either, so this process was not fundamentally different. Some areas of consensus emerged that had not been previously commented on in the literature.
As shown by the image below, sometimes you see bronchiolectasis in a fibrotic ILD but without conventional honeycombing. We agreed with Burke and colleagues who had previously reported that this most likely represented honeycombing but with fibrosis not seen in the section as this correlated with radiologic honeycombing.
The granuloma shown below elicited a variety of opinions since some considered this a response to cholesterol rather than a “true” granuloma. We finally agreed that granulomas were defined by their context as well as their content which meant that we agreed that this would count. Isolated giant cells within alveoli with cholesterol clefts were still thought to be non-specific.
In the image below although there is fibrosis around an airway which raises the issue of bronchocentric fibrosis, it is connected to the interlobular septa. We considered such cases to represent bronchocentric fibrosis only when it was an isolated finding, unlike in this case.
Other areas remained contentious, especially the extent of inflammation required to suggest an alternative diagnosis. We then tested the refined criteria using ten cases from first round plus an additional ten cases. Disappointingly interobserver variation remained stubbornly high. Even cases reviewed in first round compared to second improved only marginally. It is possible that we could have shown improved outcomes if we had tested more cases or were able to focus on cases for which the improved criteria were most applicable. However, we did not have the bandwidth to do more extensive testing.
What can we conclude? First, least surprisingly, some cases are just not readily classified using current histopathologic criteria. This has become an increasing problem since with advances in radiology, the cases that are biopsied consist increasingly of “hard cases” (as shown by a subset of our authors and my experience leading to this paper).
What can we do? We can try to improve conventional histopathology as much as possible. For example, we could establish benchmarks relative to radiologic categories to ensure that cases are diagnosed at similar rates in different centers (assuming the radiologic categories are reasonably reproducible too!). We could create larger databases of WSI as reference data sets. We could use AI to evaluate the histopathologic images.
Alternatively, we can propose that conventional histopathologic categories may simply be insufficient to adequately classify at least some ILD whether due to sampling issues (e.g. IPF can have regions of near normal lung and end stage lung) or more subtle biologic issues. At that point, we have to look beyond conventional histopathology whether it is a simple tissue-based phenotype (presence of basal cells (https://www.atsjournals.org/doi/10.1164/rccm.201712-2551OC), aberrant or not (https://advances.sciencemag.org/content/6/28/eaba1983) or a more comprehensive genomic classifier (https://www.atsjournals.org/doi/pdf/10.1164/rccm.202003-0877OC )(recognizing that any tissue based system is potentially subject to sampling issues).
Alternatively, we can focus on non-invasive testing e.g. advanced radiology, peripheral blood or BAL classifier as has been discussed for some time (https://thorax.bmj.com/content/72/5/391.long). In addition, excluding therapeutic toxicity needs further investigation especially with telomer length and immunosuppression. Finally, we should also acknowledge that improved diagnostics may not improve patient outcomes. After all, the initial trials of our current anti-fibrotics had only a minority of patients undergo biopsy.
Finally, I need to acknowledge my contributors and especially Bob Camp, the brilliant pathologist who made the web portal possible.
THE END
Comments
Post a Comment