Résumé : Background: Delayed intubation is associated with high mortality. There is a lack of objective criteria to decide the time of intubation. We assessed a recently described combined oxygenation index (ROX index) to predict intubation in immunocompromised patients. The study is a secondary analysis of randomized trials in immunocompromised patients, including all patients who received high-flow nasal cannula (HFNC). The first objective was to evaluate the accuracy of the ROX index to predict intubation for patients with acute respiratory failure. Results: In the study, 302 patients received HFNC. Acute respiratory failure was mostly related to pneumonia (n = 150, 49.7%). Within 2 (1–3) days, 115 (38.1%) patients were intubated. The ICU mortality rate was 27.4% (n = 83). At 6 h, the ROX index was lower for patients who needed intubation compared with those who did not [4.79 (3.69–7.01) vs. 6.10 (4.48–8.68), p < 0.001]. The accuracy of the ROX index to predict intubation was poor [AUC = 0.623 (0.557–0.689)], with low performance using the threshold previously found (4.88). In multivariate analysis, a higher ROX index was still independently associated with a lower intubation rate (OR = 0.89 [0.82–0.96], p = 0.04). Conclusion: A ROX index greater than 4.88 appears to have a poor ability to predict intubation in immunocompromised patients with acute respiratory failure, although it remains highly associated with the risk of intubation and may be useful to stratify such risk in future studies.