Reply to Wang et al.: Causal inference considerations in proteomic biomarker studies for interstitial lung disease

From the Authors: We thank Dr Wang and colleagues for their comments about our study1 and their own experiences at their medical center. We agree with the authors that we need to be cautious when drawing causal inferences from biomarker studies and the pathobiology driving these early processes, which result in recurrent lung injury and fibrosis. While higher plasma levels of proteins correlate with radiological phenotypes of interstitial lung injury and fibrosis,2-4 we do not entirely know whether it is a consequence of damage to the lungs or indicative of a mechanism that is driving this damage. A blood–lung tissue proteome mapping resource would partially address this issue and further guide preclinical and translational ­experiments to determine the pathobiology of subclinical pulmonary fibrosis. Related, as we pointed out as a limitation in our study, although we had repeated protein measurements over time, we do not know the rate of changes in these plasma proteins that ­correlate most strongly with a higher probability of developing disease. As Wang et al suggest, pulmonary fibrosis likely exists on a continuous spectrum, with years of microinjuries and abnormal remodeling ­occurring in the lungs of a patient prior to being clinically diagnosed. Even in the clinical setting, when we see a patient in the hospital or office, that itself is a snapshot of what is happening to that person, as we collect as much data to inform what we think is happening and how we can help improve their health. As indicated in the Discussion of our article and a recent perspectives piece on designing a preventative clinical trial for pulmonary fibrosis,5 we echo the authors’ recommendation to implement a model that integrates different modalities to identify those adults at highest risk of progressing to pulmonary fibrosis. We agree with the authors that implementation of such models and assays in the clinical setting is crucial to improve the trajectories of these at-risk individuals and help identify therapeutic targets. A major limitation of this approach has been determining which model is the simplest, most accessible, and most effective, which necessitates pooling resources and data and achieving consensus among investigators on how best to accomplish this goal. Recent studies suggest that much can be gleaned from electronic health records,6 and integrating such data with more well-phenotyped and curated research cohorts to construct such prediction models is a potential future research area. Also, the generalizability of models still needs to be tested in heterogeneous population samples, which is a significant undertaking. The authors suggest incorporating genomic risk scores, and we agree that, while promising, much of the genetic variation identified has been identified in overwhelmingly non-Hispanic White individuals of European ancestry and does not correlate as well with interstitial lung abnormalities in other groups.7 Wang and colleagues should be applauded for implementing such a workflow at their medical center, and we and other clinical investigators aspire to accomplish similar approaches as appropriate in working with our own local medical centers. We hope our study contributes to building on prior work and ongoing research for improving efforts to identify individuals at greatest risk of developing pulmonary fibrosis. Supplementary material is available at American Journal of Respiratory and Critical Care Medicine online. Please see the ICMJE disclosure forms, which have been provided as Supplementary material.