PURPOSE: We investigated if the stress applied to the lung during non-invasive respiratory support may contribute to the coronavirus disease 2019 (COVID-19) progression. METHODS: Single-center, prospective, cohort study of 140 consecutive COVID-19 pneumonia patients treated in high-dependency unit with continuous positive airway pressure (n = 131) or non-invasive ventilation (n = 9). We measured quantitative lung computed tomography, esophageal pressure swings and total lung stress. RESULTS: Patients were divided in five subgroups based on their baseline PaO(2)/FiO(2) (day 1): non-CARDS (median PaO(2)/FiO(2) 361 mmHg, IQR [323-379]), mild (224 mmHg [211-249]), mild-moderate (173 mmHg [164-185]), moderate-severe (126 mmHg [114-138]) and severe (88 mmHg [86-99], p < 0.001). Each subgroup had similar median lung weight: 1215 g [1083-1294], 1153 [888-1321], 968 [858-1253], 1060 [869-1269], and 1127 [937-1193] (p = 0.37). They also had similar non-aerated tissue fraction: 10.4% [5.9-13.7], 9.6 [7.1-15.8], 9.4 [5.8-16.7], 8.4 [6.7-12.3] and 9.4 [5.9-13.8], respectively (p = 0.85). Treatment failure of CPAP/NIV occurred in 34 patients (24.3%). Only three variables, at day one, distinguished patients with negative outcome: PaO(2)/FiO(2) ratio (OR 0.99 [0.98-0.99], p = 0.02), esophageal pressure swing (OR 1.13 [1.01-1.27], p = 0.032) and total stress (OR 1.17 [1.06-1.31], p = 0.004). When these three variables were evaluated together in a multivariate logistic regression analysis, only the total stress was independently associated with negative outcome (OR 1.16 [1.01-1.33], p = 0.032). CONCLUSIONS: In early COVID-19 pneumonia, hypoxemia is not linked to computed tomography (CT) pathoanatomy, differently from typical ARDS. High lung stress was independently associated with the failure of non-invasive respiratory support.