I read with interest the recent article by Macie and colleagues (September 2006).1I congratulate them on an exhaustive analysis of a well-designed observational study. I disagree, however, with the conclusion on several points. First, it is stated in the abstract that therapy “with ICSs [inhaled corticosteroids] reduced mortality in COPD patients.” Causality cannot be inferred from a case-control observational study no matter how well statistically modeled. Further, could undetected bias explain the association between ICS use and reduced mortality? The effect of ICSs on mortality may have been overestimated because of healthy user and provider selection bias.2–3 Could ICS users be a more robust group in terms of cardiovascular risk factors, or behavioral or lifestyle modifications? Further, providers might be more inclined to prescribe ICSs to less ill, more motivated COPD patients. The article supports this notion on several occasions. In Table 1, ICS users were more likely to receive other medications, visit physicians more frequently, and in older patients had less comorbidity. “In subjects who were > 65 years of age … who died had more comorbidities … received more prescriptions … other than ICSs than did control subjects.” It would have been instructive if the authors could have stratified the data on the basis of the percentage of patients in each group who received a prescription of ICSs but was later not dispensed. This would have provided insight into healthy user and provider selection bias. Examples of confounders not adjusted for in analyses include tobacco usage, body mass index, statin use, severity of COPD, social class, and physical activity. Model adjustments for complex social and behavioral factors are difficult and are outside the scope of observational studies in general.4 The patient out-of-pocket costs of medical therapy in COPD are considerable. It is incumbent on pulmonologists to parcel out the data carefully lest an ineffective therapy for reducing COPD-related mortality becomes the standard of care.