THE AUTHORS REPLY

We thank Dr. Banack et al. (1) for their comments on our study (2), which examined differences in the obesity paradox among persons with smoking-associated and non–smoking-associated chronic obstructive pulmonary disease (COPD). The authors have succinctly summarized many methodological biases that arise in studies describing the paradox.

We are surprised that Banack et al. conjecture that our investigation was premised on a causal relationship between body mass index (BMI) and mortality (1). We gently note that methodological explanations for the paradox, including selection bias, reverse causation, and collider stratification bias, were explicitly considered within our discussion as key elements that may drive this finding (2).

We did not presume causation and took care to avoid causal suggestions of protection or harm. Rather, as is discussed within the article (2), a valid interpretation of our results is an explicit challenge of the obesity paradox within COPD as a manifestation of these aforementioned and other biases, bolstered by parallel findings suggesting methodological biases as explanations for obesity paradoxes within diabetes (3) and heart disease (4). Use of the term “paradox” in the title and text reflects the use of this phrase in common literature.

We agree with the authors that our study did not have sufficient resolution to definitively characterize the BMI-mortality association among persons with non–smoking-associated COPD. We recognize the limitations of our small sample size, a limitation which is also reflective of the broader dearth of research on this COPD phenotype. However, the consistent finding of effect-measure modification establishes that the pattern and magnitude of the BMI-mortality association is different overall between persons with smoking-associated COPD and those with non–smoking-associated COPD. With specific respect to the comment on overlapping confidence intervals (1), we note that within the subanalysis of participants with moderate-to-severe obstruction, which would be more aligned with cohorts with clinically significant disease, there was a clear separation in estimates and intervals between smoking-associated and non–smoking-associated COPD among both overweight and obese individuals. This is illustrated in our Web Table 2 (available at https://academic.oup.com/aje).

More broadly, we acknowledge ardent debate among epidemiologists, clinicians, and physiologists regarding the nature of this finding, and we believe that maintaining continued and dispassionate scientific dialog is warranted.