Response

To the Editor:

We thank Dr Patel for interest in our article in CHEST.¹ Dr Patel raises important questions regarding the possibility of unmeasured confounding in observational comparative effectiveness research and potential mechanisms for improved outcomes with use of β-blockade during sepsis. We agree that etiological associations cannot generally be inferred from a single observational study, and advocate for care in description of results from observational comparative effectiveness research.² Hence, we explicitly described our findings regarding β-blockers in terms of “associated with outcome” rather than “improving outcome,” we discussed the possibility of unmeasured confounding as a limitation of our findings, we did not make treatment recommendations, we used both propensity score and instrumental variable-based methods to address potential confounding, and called for trials to evaluate the effectiveness of atrial fibrillation treatments during sepsis. As suggested by Dr Patel, APACHE and SOFA severity of illness scores are not a panacea for unmeasured confounding; the moderate discriminative ability of severity of illness scores for predicting mortality is similar to that of models derived from enhanced administrative databases such as the one used in our study.^{3, 4} For many of the reasons raised by Dr Patel and discussed in our article, we respectfully submit that the difference in outcomes observed between patients who received β-blockers and those who received calcium channel blockers may be less susceptible to confounding by unmeasured severity of illness than the results for digoxin or amiodarone. Finally, multiple potentially beneficial mechanisms of β-blockade during sepsis have been proposed, including improved hemodynamics and immunomodulatory effects.⁵ We encourage further investigation of optimal methods to treat atrial fibrillation during sepsis.