Aims The survival of preterm infants is a matter of wide public interest. Survival prediction is important for clinicians when advising parents and for risk adjustment when comparing providers. Prediction models are generally based on historical data, often from hospital rather than population-based cohorts. Here we demonstrate the use of near-contemporaneous electronic National Health Service (NHS) clinical data to provide a practical, up-to-date, web-based survival prediction tool for preterm infants admitted to neonatal units in England. We compared this with existing UK and US predictors and evaluated the change in survival over recent years for extremely preterm babies from 22+0–25+6 week gestational age (GA) admitted to neonatal care, in comparison with previous published data.
Results There were 17,491 infants included in the cohort, of whom 16,164 (92%) survived. Birth weight, GA, sex, antenatal steroids, and multiple birth were factors included in the final model. The interactive tool is available online for open access. The model showed good discrimination internally (area under ROC curve (AUC) = 0.89, 95% CI 0.88 to 0.90) and on independent data (AUC = 0.87, 95% CI 0.82 to 0.91). Predictive performance was similar to previous UK models and improved over a US model. There has been no statistically significant increase in survival to discharge of admitted infants born at 22+0–25+6 weeks gestation in England since 2006 (Relative Risk 1.10, 95% CI 0.98 to 1.22, p = 0.11).
Conclusions We have established the feasibility of employing contemporaneous, population-based, routinely collected, electronic NHS data for survival modelling for preterm babies admitted to neonatal care. The model is available in a web tool readily accessible to clinicians, parents, and healthcare managers and can be regularly updated to readily assess population changes in neonatal survival.