Objective: To determine the association of arterial partial pressure of carbon dioxide PaCO2with severe intraventricular haemorrhage (sIVH), bronchopulmonary dysplasia (BPD), and neurodevelopmental impairment (NDI) at 18-22 months in premature infants. Design: Secondary exploratory data analysis of Surfactant, Positive Pressure, and Oxygenation Randomised Trial (SUPPORT). Setting Patients: Multiple referral neonatal intensive care units. 1316 infants 24 0/7 to 27 6/7 weeks gestation randomised to different oxygenation (SpO2target 85-89% vs 91-95%) and ventilation strategies. Main outcome measures: Blood gases from postnatal day 0 to day14 were analysed. Five PaCO2variables were defined: minimum (Min), maximum (Max), SD, average (time-weighted), and a four level categorical variable (hypercapnic (highest quartile of Max PaCO2), hypocapnic (lowest quartile of Min PaCO2), fluctuators (hypercapnia and hypocapnia), and normocapnic (middle two quartiles of Max and Min PaCO2)). PaCO2variables were compared for infants with and without sIVH, BPD and NDI (±death). Multivariable logistic regression models were developed for adjusted results. Results: sIVH, BPD and NDI (±death) were associated with hypercapnic infants and fluctuators. Association of Max PaCO2and outcomes persisted after adjustment (per 10 mm Hg increase: sIVH/death: OR 1.27 (1.13 to 1.41); BPD/death: OR 1.27 (1.12 to 1.44); NDI/death: OR 1.23 (1.10 to 1.38), death: OR 1.27 (1.12 to 1.44), all p<0.001). No interaction was found between PaCO2category and SpO2treatment group for sIVH/death, NDI/death or death. Max PaCO2was positively correlated with maximum FiO2(rs0.55, p<0.0001) and ventilator days (rs0.61, p<0.0001). Conclusions: Higher PaCO2was an independent predictor of sIVH/death, BPD/death and NDI/death. Further trials are needed to evaluate optimal PaCO2targets for high-risk infants.