This study also had several important limitations. First, although we performed rigorous statistical methods to correct for potential confounders, including adjusting for patient sociodemographic factors, comorbidities, and behaviors that might influence treatment outcomes, it is possible that we were unable to completely correct for confounding factors, including severity of comorbid illness and hypertension. This population may be at risk for important comorbidities that are potential causes of heart failure in the presence of COPD, such as obstructive sleep apnea. Furthermore, factors such as alcohol abuse and use of other medications, including nonsteroidal antiinflammatory drugs, may have influenced the risk of COPD and CHF. We did not have information on these factors available to us within the data set and, therefore, could not include them in our models. To control for severity of hypertension, we restricted the cohort to patients on two treatments. Nonetheless, patients may be receiving dual therapy for other comorbidities, and we may have been unable to adequately adjust for confounding by indication for other conditions with similar indications for the selected medications, such as the prescription of ACE-Is to treat proteinuria among patients with diabetes. Second, although we were able to confirm that patients filled their prescriptions from VA pharmacies, we did not assess whether they actually consumed their medications. Not taking prescribed medications would not invalidate the results unless medication consumption occurred disproportionally across medication class combination. Moreover, it seems unlikely that patients would consistently fill medications without taking them regularly. It is also possible that patients obtained additional medications from other pharmacies. Third, the cohort was predominantly male, and although it has been shown that heart failure and hypertension occur less commonly in male patients than in female patients,32 we may not be able to generalize conclusions from the results to female patients. Furthermore, data on race/ethnicity other than white were unavailable, precluding an analysis of differences in antihypertensive efficacy among race/ethnic groups. Fourth, we cannot make inferences about whether the use of thiazides were associated with reductions in disease-specific mortality, although we believe that this information would be valuable in future studies. Finally, the outcome measure was based on the primary ICD-9 discharge diagnosis code. Although this approach is common and valid in observational studies,33 we did not have echocardiography assessment of left ventricular function, which may or may not have been otherwise clinically defined by the attending physician or by cardiologist consultation. Without echocardiographic studies, we were unable to distinguish whether heart failure diagnosis was attributable to systolic or diastolic heart failure or the possible influence associated with the use or selection of β-blockers or ACE-Is/ARBs.