We thank Drs Takir and Esquinas for their insightful commentary regarding our findings. In our study, we analyzed data from the electronic medical records of 58 hospitals. We adjusted for patient case mix and several variables indicative of disease severity, including the Laboratory Acute Physiological Score, which integrates 14 laboratory tests, including arterial blood gas, and for prior noninvasive ventilation (NIV) or invasive ventilation (intermittent mandatory ventilation [IMV]). The models predicting NIV and IMV use had strong discriminatory ability with c-statistics > 0.78 using only patient-level variables as fixed effects. Patient characteristics did not differ according to hospital quartiles of adjusted NIV rates. Moreover, we performed a hospital-level rather than patient-level analysis, which greatly reduces the risk of confounding by indication. Nevertheless, we agree that a study limitation is its lack of information about prior pulmonary function test results.