Looking beyond the NICU Long-term outcomes and machine learning prediction

This thesis studies the long-term outcomes after preterm birth, focussing on pulmonary and neurodevelopmental outcome and investigates the value of machine learning in the prediction of these outcomes.
In Chapter 1, a meta-analysis is performed to study the pulmonary outcome of preterm infants. Preterm born children face a three times increased risk of adverse pulmonary outcome. A lower gestational age or birthweight, bronchopulmonary dysplasia and receiving invasive mechanical ventilation were all significantly associated with increased risk of adverse pulmonary outcome.
Chapter 2 identifies studies using machine learning to predict neurodevelopmental outcome. This study finds that models often face risk of inflated predictive performance due to data leakage from testing to training data, and that those with sufficient quality are limited to outcome up to two years and are mainly based on predictors derived from advanced MRI imaging.
In Chapter 3 and 4, machine learning models were developed to predict neurodevelopmental outcome based on readily available predictors from the neonatal period. The models reached moderate overall performances (AUC up to 0.703), yet performed significantly better than the conventional models and reached high negative predictive values (up to 95%). The models performed better predicting outcome at five than outcome at two years of age. Vital signs handled through a basic approach modestly improved prediction of motor outcome but not cognitive outcome.
In Chapter 5 the design and implementation of a sturctured neonatal follow-up with integrated data-pipeline was studied. This structured follow-up of highly vulnerable neonates and their parents facilitates both individual patient care and health care innovation through evaluation and scientific research.