Prediction models of bronchopulmonary dysplasia: a systematic review and meta-analysis with validation

Abstract

AbstractIntroductionNeonatal bronchopulmonary dysplasia (BPD) is associated with lifelong respiratory and neurological sequalae. Prediction models could identify infants at greatest risk of BPD and allow targeted preventative strategies. We performed a systematic review and meta-analysis with external validation of identified models.MethodsStudies using predictors available before day 14 of life to predict BPD in very preterm infants were included. Two reviewers assessed 7,628 studies for eligibility. Meta-analysis of externally validated models was followed by validation using 62,864 very preterm infants in England and Wales.Results64 studies using 53 prediction models were included totalling 274,407 infants (range 32–156,587/study). 35 (55%) studies predated 2010; 39 (61%) were single-centre studies. 46 (87%) models were developed for the first week of life. Overall, 97% of studies had a high risk of bias, especially in the analysis domain. Internal (25%) and external (30%) validation were performed infrequently in the 44 model derivation studies. Following meta-analysis of 22 BPD and 11 BPD/death composite models, Laughon’s day one model was the most promising in predicting BPD and death with a fair C-statistic of 0.76 (95% CI 0.70–0.81) and good calibration. Six models were externally validated in our cohort with a C-statistic between 0.70 to 0.90 but with poor calibration.ConclusionFew BPD prediction models were developed with contemporary populations, underwent external validation, or had calibration and impact analyses. To reduce the adverse impact of BPD, we need contemporary, validated, and dynamic prediction models to allow targeted preventative strategies.What is the key question?This review aims to provide a comprehensive assessment of all BPD prediction models developed to address the clinical uncertainty of which predictive model is sufficiently valid and generalisable for use in clinical practice and research.What is the bottom line?Published BPD prediction models are mostly outdated, single centre and lack external validation.Why read on?Laughon’s 2011 model is the most promising but more robust models, using contemporary data with external validation are needed to support better treatments.