ROX vs. HACOR: early prediction of HFNC failure in ICU patients admitted from the emergency department with lung cancer-related hypoxemia

Delayed intubation in lung cancer patients receiving high-flow nasal cannula (HFNC) therapy is associated with increased mortality. Reliable bedside tools are essential for early risk stratification. We aimed to compare the predictive performance of the ROX index and HACOR score for the early identification of HFNC failure in this high-risk population. This retrospective observational study included 176 ICU patients with lung cancer and acute hypoxemic respiratory failure who received HFNC. Time-point analyses of the ROX index and HACOR score at 1 and 3 h were restricted to patients with complete physiological data at both time points ( n = 88). The primary outcome was clinician-determined HFNC failure, defined as endotracheal intubation within 48 h after HFNC initiation. HFNC failure occurred in 50 of 176 patients (28.4%) in the overall cohort and in 39 of 88 patients (44.3%) in the complete-case analytical cohort. Patients who required endotracheal intubation exhibited higher baseline illness severity (SOFA and APACHE II) and persistent tachypnea. At the 3-hour assessment, the ROX index showed higher observed discrimination than the HACOR score (AUROC 0.860 [95% CI 0.760–0.960] vs. 0.740 [95% CI 0.590–0.890]). Only the prespecified 1- and 3-hour time points were evaluated; therefore, these findings do not establish 3 h as the optimal assessment time. In an exploratory analysis in which the threshold was derived and evaluated in the same analytical cohort, a 3-hour ROX value ≤ 5.1 identified patients at higher risk of HFNC failure; this threshold has not been independently validated. Respiratory rate alone also showed high prognostic value at 3 h (AUROC 0.850 [95% CI 0.730–0.970]). Within the complete-case cohort, which represented half of the overall population, the 3-hour ROX index showed higher observed discrimination for clinician-determined HFNC failure than HACOR. However, selection and incorporation bias may have affected this comparison. The data-derived threshold should not be used as a stand-alone intubation criterion and requires prospective validation using standardized serial measurements and independently adjudicated outcomes. Not applicable; this was a retrospective observational study.