What are the potential selection biases?
Selection bias occurs when there is something inherently different between the two groups being compared that could explain the differences in observed outcomes. This is a potential problem in non-randomized studies. The authors controlled for some potential confounders by including 7 covariates from table 2 in the proportional hazard model (i.e., age, gender, race, smoking status, baseline FVC, baseline DLCO, baseline A-a gradient). However, the authors did not include all the variables in table 2, some of which seem important and were even statistically significant across neutrophil % tertile by univariate analysis (i.e., baseline TLC, baseline FEV1, duration of illness). They mention controlling for a composite physiologic index, but details of index not described in paper. Data on other potentially important confounders was not collected nor included in the hazard model, such as whether on medications or not and which medications (steroids, imuran, cyclophosphamide, NAC), extent an
