Background: To determine whether lung ultrasound (LUS) may early predict the failure of non-invasive respiratory support (high-flow nasal cannula-HFNC, continuous positive airway pressure-CPAP, non-invasive ventilation-NIV) in hypoxemic patients.

Methods: In this prospective multicenter international observational study, we enrolled patients undergoing non-invasive treatments for hypoxemia (PaO₂/FiO₂ < 300 mmHg). LUS, PaO₂/FiO₂and ROX index were assessed before (baseline) and 2 h after treatment start. Regional/global LUS aeration scores were computed (4 degrees of loss-of-aeration: 0-normal to 3-severe loss of aeration) in 6 regions per hemithorax (2 anterior, 2 lateral, 2 posterior). Failure was defined as need of respiratory support’s escalation within 48 h (HFNC to CPAP to NIV, any support to intubation/ECMO).

Results: We studied 100 patients (age 70 [57-76] years; female sex 39%; supports: 13 HFNC, 68 CPAP, 19 NIV); the overall rate of treatment failure was 22%. At the baseline, clinical and ultrasound parameters were similar in failing and non-failing patients; after 2 h, failing patients had lower PaO₂/FiO₂. (149 mmHg [124-201] vs. 200 [171-243]; p = 0.001), lower ROX index (7.8 [4.9-9.2] vs. 10.9 [7.9-13.8]; p = 0.003) and higher lateral (3.0 [1.0-6.0] vs. 1.5 [0.0-3.0]; p = 0.047), antero-lateral (4.0 [1.0-9.0] vs. 2.0 [0.0-4.0]; p = 0.027) and global (13.0 [8.0-17.0] vs. 10.0 [7.0-13.0]; p = 0.036) LUS aeration scores. No improvement in lung aeration was observed in failing patients within the initial 2 h of treatment (global LUS score variations 0.0 [-2.0-1.0] vs. -3.0 [-5.0 – -2.0]; p < 0.001). ROX index and antero-lateral/global LUS scores’ variations were independent predictors of failure. AUCs for treatment failure were: 2-hour ROX index 0.71 [0.58-0.84], 2-hour PaO₂/FiO₂ 0.73 [0.60-0.85], global LUS score variations 0.73 [0.62-0.89]. A combined clinical-ultrasound score (ROX-US) showed AUC of 0.82 [0.73-0.91]. A ROX-US≥1 identified the success of the treatment with sensitivity 95% and specificity 50%; a ROX-US≥2 identified the success of the treatment with sensitivity 45% and specificity 96%.

Conclusions: Changes in LUS aeration scores induced by 2 h of non-invasive respiratory support help early predict the risk of treatment failure. LUS score improved only in responders and was an independent predictor of failure.