High-flow nasal therapy vs conventional oxygen therapy in mild COVID-19 hypoxaemia: a Bayesian reanalysis of the COVID-HIGH Trial

Background: Clinical effectiveness of high-flow nasal therapy (HFNT) over conventional oxygen therapy (COT) in patients with mild COVID-19-related acute hypoxaemic respiratory failure (AHRF) remains uncertain. The COVID-HIGH trial did not demonstrate statistically significant benefits of HFNT over COT. However, the trial was slightly underpowered, and the event rate lower-than-expected. Bayesian methods provide deeper insight by incorporating prior knowledge and quantifying uncertainty intuitively. This analysis aimed to quantify the probability of benefit or harm associated with HFNT, adopting a Bayesian approach. Methods: We performed a Bayesian reanalysis of the COVID-HIGH trial (NCT, which randomised 364 patients with PaO₂/FiO₂ between 200-300 mmHg to receive HFNT or COT. The primary outcome was escalation of respiratory support (continuous positive airway pressure, noninvasive ventilation or invasive mechanical ventilation) within 28 days. A key secondary outcome was clinical recovery at day 14. Bayesian logistic models with noninformative and informative priors were used to estimate the posterior probability of treatment effects. Results: Escalation of respiratory support occurred in 23.6% (HFNT) versus 30.2% (COT) (risk difference - 6.6%, 95% CI - 15.1 to 2.1; p = 0.14). Across a wide range of priors, the posterior probability mass on the beneficial side remained high, generally > 70%, while the proportion on the harm side remained consistently low at ≤ 6% for all models, underscoring a favourable benefit-risk profile. The acute respiratory failure meta-analysis model (OR 0.76, 95% CrI 0.60-0.97), the COVID-19 randomised evidence model (OR 0.76, 95% CrI 0.60-0.97), the COVID-19 observational evidence model (OR 0.60, 95% CrI 0.45-0.80), and the COVID-19 Bayesian meta-analysis mixed evidence model (OR 0.66, 95% CrI 0.52-0.86) showed posterior probability mass on the beneficial side of 70%-94%. Clinical recovery at day 14 occurred in 61.5% (HFNT) versus 53.3% (COT), with 61-73% of posterior probability mass on the clinical benefit side. Conclusions: This Bayesian re-analysis of the COVID-HIGH trial suggests that HFNT likely reduces escalation of respiratory support and improves clinical recovery in patients with COVID-19 pneumonia and mild hypoxaemia, although the magnitude of benefit remains uncertain and sensitive to prior assumptions. Trial registration: The trial was prospectively registered in ClinicalTrials.gov on December 7, 2020 (NCT04655638).