Given the pathophysiology of hypoxemia in patients with Covid-19 acute respiratory failure (ARF), it seemed necessary to evaluate whether ROX index (ratio SpO ₂ /FiO ₂ to respiratory rate) could accurately predict intubation or death in these patients initially treated by high-flow nasal oxygenation (HFNO). We aimed, therefore, to assess the accuracy of ROX index to discriminate between HFNO failure (sensitivity) and HFNO success (specificity).

We designed a multicentre retrospective cohort study including consecutive patients with Covid-19 ARF. In addition to its accuracy, we assessed the usefulness of ROX index to predict HFNO failure (intubation or death) via logistic regression.

Among 218 ARF patients screened, 99 were first treated with HFNO, including 49 HFNO failures (46 intubations, 3 deaths before intubation). At HFNO initiation (H0), ROX index sensitivity was 63% (95%CI 48–77%) and specificity 76% (95%CI 62–87%) using Youden’s index. With 4.88 as ROX index cut-off at H12, sensitivity was 29% (95%CI 14–48%) and specificity 90% (95%CI 78–97%). Youden’s index yielded 8.73 as ROX index cut-off at H12, with 87% sensitivity (95%CI 70–96%) and 45% specificity (95%CI 31–60%). ROX index at H0 was associated with HFNO failure ( p  = 0.0005) in univariate analysis. Multivariate analysis showed that SAPS II ( p  = 0.0003) and radiographic extension of pulmonary injuries ( p  = 0.0263), rather than ROX index, were predictive of HFNO failure.

ROX index cut-off values seem population-specific and the ROX index appears to have a technically acceptable but clinically low capability to discriminate between HFNO failures and successes in Covid-19 ARF patients. In addition, SAPS II and pulmonary injuries at ICU admission appear more useful than ROX index to predict the risk of intubation.