The authors reply

We thank Ghosh et al (1) for their interest in our study (2), addressing the use of airway pressure release ventilation (APRV) in patients with COVID-19.

As found in a recent secondary analysis of multicenter randomized controlled trials (RCTs) of critically ill adults published in high-impact journals (3), we are aware that an overly optimistic assumption of unrealistic treatment effects at calculation of sample size is one of the main reasons for biased results toward the null (type II error). However, as described in the methods of our study (2), we calculated the sample size with an expected difference of 4 days in ventilator-free days (VFDs), considering a mean of 14 VFD according to the experience in our center; this yields an a priori minimal clinically important difference of 28.5%, which is not far from the 25% reported in the Prospective, Randomized, Multi-Center Trial of 12 mL/kg vs. 6 mL/kg Tidal Volume Positive Pressure Ventilation for Treatment of Acute Lung Injury and Acute Respiratory Distress Syndrome, and is even more realistic than the results observed in the largest RCT of APRV, which reported a significant difference in VFD of 19 and 2 days in APRV and control groups, respectively (an 89% increase in VFD with APRV) (4). More importantly, questioning the calculation of sample size in our study (2) is a moot point, as the sample size was not reached.

We agree with Ghosh et al (1) regarding the limitations of the frequentist approach of data analysis. Bayesian approach combines the background knowledge (the “priors”) into the analysis to obtain a “posterior probability interval,” which facilitates the clinical decision-making process compared with the frequentist paradigm. However, this increasingly popular approach is not free from limitations; the results of the models are prone to the influence from the “prior” specifications, which could be adjusted to better fit the results to a given hypothesis. “Priors” are very important, especially when dealing with low sample sizes (5), and due to the small and few RCTs of APRV performed so far, this could be a relevant issue, because selection of “priors” are based on how much information we have and how accurate it is.

The methodology of APRV has evolved over the last decades. This variability on application leads to important challenges on how to interpret available evidence. When investigating the application of modes of mechanical ventilation, the evolving nature of knowledge on implementation will always lead to some uncertainty, as the question will remain, was it implemented as it should? Pending the publication of more and larger RCTs, which ensure consistent implementation, the question of whether a specific mode is better than others will remain unanswered. However, many of the current studies, including ours, present specific details on how APRV is being applied and as such do contain information that so far highlight its challenges. We share the interpretation of Ghosh et al (1) and suggest that readers do not deem our study (2) as negative; we prefer to consider our results as a step forward in order to increase the awareness of APRV characteristics and pitfalls. As far as we are concerned, the door is still wide open for further research.