Background Early prediction of BPD is important for identifying high risk patients likely to benefit from preventive treatment approaches and for providing prognostic information. Therefore we aimed to develop a risk score for BPD based on early available clinical parameters.

Patients and methods: All infants born at the University Hospital of Lausanne < 32 weeks of gestation (WG) between 1998 and 2007 (n=936) were included. Patients diagnosed with RDS (n=232) were divided in two groups, either developing BPD or not. Independent risk factors for the development of BPD were searched by multivariate logistic regression analysis. The β-coefficients (β=log(OR)) derived from the fitted multivariate model were used to build a scoring system. An internal validation was performed using a two-fold cross-validation technique with two subgroups: two thirds of the patients were used as training set for model calibration and one third as prediction set.

Results BPD-risk score was developed based on five covariates: intubation in the delivery room, early neonatal infection, duration of invasive mechanical ventilation in days, birth weight and gestational age, weighted according their β-coefficients. Area under curve (AUC) was 0.896. Sensitivity and specificity reached 82.7% and 82.6% with a score cut-off of –8 (range –25 to +17). Internal calibration proved a good prediction: AUC for the same cut-off was 0.882 for the training set and 0.927 for the prediction set.

Conclusions A simple scoring system available within the first postnatal week can reliably predict the probability of developing BPD in infants born < 32 WG.