We appreciate the letter by Petroianni and Esquinas (1), who proposed three important points for deliberation (2). First, they raised concerns about the need for a control arm to adequately assess the effect of noninvasive ventilation (NIV) compared to invasive mechanical ventilation (IMV) for use in patients with pneumonia. We agree that despite our use of an instrumental variable approach to account for selection bias, our analysis cannot substitute for a rigorously performed randomized controlled trial. They also expressed concern about imbalance in the proportion of individuals with a primary diagnosis of pneumonia between the NIV and IMV arms and suggest that this could lead to residual confounding. Our use of an instrumental variable approach effectively balanced the primary diagnosis across the two arms (details available in Supplementary Table 3 in the original publication), lessening the risk of confounding due to differences related to primary diagnosis. Finally, they noted that our results were limited by our inability to account for relevant clinical variables that clinicians use when deciding which mode of mechanical ventilation to provide. We agree that this is an important limitation to our work that will hopefully be addressed by future clinical trials.
Well-powered or enriched randomized trials to evaluate the appropriate mode of ventilatory support for patients with pneumonia and respiratory failure are urgently needed. Our results highlight the lack of existing evidence to support such a critical clinical decision and the ongoing uncertainty as to what drives clinicians to make these decisions. Future work must take into account the difficulty that bedside clinicians face when attempting to decide: 1) whether to initiate mechanical ventilation for a patient with pneumonia and 2) whether it is best to “try and fail” with a noninvasive therapy or “succeed without trying” by using IMV.