Objective: To estimate the risk of death or tracheostomy placement (D/T) in infants with severe bronchopulmonary dysplasia (sBPD) born < 32 weeks' gestation referred to regional neonatal intensive care units. Study design: We conducted a retrospective cohort study in infants born < 32 weeks' gestation with sBPD in 2010-2011, using the Children's Hospital Neonatal Database. sBPD was defined as the need for FiO2 ⩾ 0.3, nasal cannula support >2 l min(-1) or positive pressure at 36 weeks' post menstrual age. The primary outcome was D/T before discharge. Predictors associated with D/T in bivariable analyses (P < 0.2) were used to develop a multivariable logistic regression equation using 80% of the cohort. This equation was validated in the remaining 20% of infants. Result: Of 793 eligible patients, the mean gestational age was 26 weeks' and the median age at referral was 6.4 weeks. D/T occurred in 20% of infants. Multivariable analysis showed that later gestational age at birth, later age at referral along with pulmonary management as the primary reason for referral, mechanical ventilation at the time of referral, clinically diagnosed pulmonary hypertension, systemic corticosteroids after referral and occurrence of a bloodstream infection after referral were each associated with D/T. The model performed well with validation (area under curve 0.86, goodness-of-fit χ(2), P = 0.66). Conclusion: Seven clinical variables predicted D/T in this large, contemporary cohort with sBPD. These results can be used to inform clinicians who counsel families of affected infants and to assist in the design of future prospective trials.