Abstract
BACKGROUND: The effect of different levels of positive end-expiratory pressure in invasively ventilated critically ill patients remains a matter of debate. The REstricted versus Liberal Positive End-Expiratory Pressure in Patients Without ARDS (RELAx) is a multicentric, randomized trial comparing a lower positive end-expiratory pressure strategy versus a higher positive end-expiratory pressure strategy in ventilated patients without acute respiratory distress syndrome, which demonstrated non-inferiority of lower positive end-expiratory pressure compared to higher positive end-expiratory pressure on ventilator-free days. The primary analysis was published in 2020, and a frequentist statistical approach was applied. AIM: To present the protocol of the Bayesian analysis plan that will be used to re-analyse the RELAx trial to provide complementary and additional insight into this clinical trial. METHODS: This re-analysis will focus on the probability of superiority of the intervention. As an ordinal variable, the primary outcome will be ventilator-free days at day 28, and posterior estimates will be obtained by fitting a hierarchical cumulative logistic regression model. Secondary outcomes will be mortality at day 28, as a binary outcome, and ventilation duration, as a continuous outcome. We will adopt neutral, pessimistic, and optimistic priors informed by current literature, and a fourth prior derived from an expert's survey. Probability thresholds will be defined for superiority, severe harm, and a region of practical equivalence. DISCUSSION: The RELAx trial findings raise the hypothesis that a lower positive end-expiratory pressure strategy may be at least as effective, if not superior, in specific patient-centred outcomes. This analysis is designed to augment and contextualize the original frequentist analysis of the largest randomized trial comparing positive end-expiratory pressure strategies in non-acute respiratory distress syndrome patients. Results will be presented with a continuum of credible intervals and probabilities of effects to facilitate a nuanced interpretation. We offer clinically meaningful insights that complement and extend the trial's original analysis by reporting probabilities of benefit, harm, and equivalence.