Bayesian and heterogeneity of treatment effect analyses of the HOT-ICU trial - a secondary analysis protocol

Background: The Handling Oxygenation Targets in the Intensive Care Unit (HOT-ICU) trial is an ongoing randomised clinical trial exploring the benefits and harms of targeting a lower (8 kPa) versus a higher (12 kPa) arterial oxygenation target in adult patients acutely admitted to the intensive care unit (ICU) with hypoxaemic respiratory failure. Methods: This protocol describes a secondary analysis of the primary trial outcome, 90-day all-cause mortality. We will analyse the primary outcome using Bayesian methods, which allows quantification of probabilities of all effect sizes. We will explore the presence of heterogeneity of treatment effects (HTE) using Bayesian hierarchical models in subgroups based on baseline parameters: (a) severity of illness (Sequential Organ Failure Assessment (SOFA) score), (b) severity of hypoxaemic respiratory failure (partial pressure of arterial oxygen (PaO ₂)/fraction of inspired oxygen (FiO ₂) ratio), (c) vasopressor requirement (highest noradrenaline dose in the 24 hours prior to randomisation), and (d) plasma lactate concentration (latest prior to randomisation). Additionally, we will perform separate assessments of the treatment effect interaction with each of the baseline parameters above on the continuous scale and present these using conditional effects plots. Conclusions: This secondary analysis will aid the interpretation of the HOT-ICU trial by evaluating probabilities of all effect sizes. In addition, we will evaluate whether HTE is present, thus, further evaluating benefits and harms of a lower versus a higher oxygenation target in adult ICU patients with acute hypoxaemic respiratory failure.