PURPOSE: The life-saving role of oxygen therapy in African children with severe pneumonia is not yet established. METHODS: The open-label fractional-factorial COAST trial randomised eligible Ugandan and Kenyan children aged > 28 days with severe pneumonia and severe hypoxaemia stratum (SpO2 < 80%) to high-flow nasal therapy (HFNT) or low-flow oxygen (LFO: standard care) and hypoxaemia stratum (SpO2 80-91%) to HFNT or LFO (liberal strategies) or permissive hypoxaemia (ratio 1:1:2). Children with cyanotic heart disease, chronic lung disease or > 3 h receipt of oxygen were excluded. The primary endpoint was 48 h mortality; secondary endpoints included mortality or neurocognitive sequelae at 28 days. RESULTS: The trial was stopped early after enrolling 1852/4200 children, including 388 in the severe hypoxaemia stratum (median 7 months; median SpO2 75%) randomised to HFNT (n = 194) or LFO (n = 194) and 1454 in the hypoxaemia stratum (median 9 months; median SpO2 88%) randomised to HFNT (n = 363) vs LFO (n = 364) vs permissive hypoxaemia (n = 727). Per-protocol 15% of patients in the permissive hypoxaemia group received oxygen (when SpO2 < 80%). In the severe hypoxaemia stratum, 48-h mortality was 9.3% for HFNT vs. 13.4% for LFO groups. In the hypoxaemia stratum, 48-h mortality was 1.1% for HFNT vs. 2.5% LFO and 1.4% for permissive hypoxaemia. In the hypoxaemia stratum, adjusted odds ratio for 48-h mortality in liberal vs permissive comparison was 1.16 (0.49-2.74; p = 0.73); HFNT vs LFO comparison was 0.60 (0.33-1.06; p = 0.08). Strata-specific 28 day mortality rates were, respectively: 18.6, 23.4 and 3.3, 4.1, 3.9%. Neurocognitive sequelae were rare. CONCLUSIONS: Respiratory support with HFNT showing potential benefit should prompt further trials.