Robert REED

Predicting the risk of 30-day readmission to hospital amongst preterm infants in France via predictive risk modelling
Etablissement :
Promotion :
Domaine d’étude :
Unité de recherche : Center for Research on Epidemiology and Statistics Sorbonne Paris Cité INSERM, EPOPé
Description de la thèse :

Preventing unplanned rehospitalisation is a key challenge facing health systems. Unplanned rehospitalisations are associated with large costs and inconvenience for the systems and patients. It is recognised that preterm infants face more challenges in the newborn period compared with full-terms. They experience excess morbidity extending beyond the initial birth hospitalisation and readmission rates amongst late preterm infants are between 1.5 to 3 times that of full-term infants.
One way to reduce rehospitalisations is through accurate risk stratification that allows high risk individuals to be targeted with interventions. Predictive risk models can provide such as solution. At present there are no predictive models for the rehospitalisation of preterms. Though explanatory models identifying common risk factors for preterm readmission are common, the divergent properties of predictive compared to explanatory models mean they tell us little about prediction and risk stratification. There is a large body of literature regarding predictive models for 30-day rehospitalisation in a range of settings. However, the overwhelming focus on adult populations and the resultant dearth of models specific to preterm infants is a clear blind-spot. Given the very specific challenges faced by preterm infants none of the current models in the literature can be applied to our population. This study intends to address the current need for a predictive risk model for rehospitalisation specific to preterm infants.
In order to identify common risk factors for rehospitalisation a systematic review of risk factors for the rehospitalisation of preterm infants will be conducted. The next stage will be that of predictive risk modelling. For this stage, we aim to predict and validate predictive models for 30-day unplanned rehospitalisation among a cohort of French pre-term infants. This is a retrospective cohort study using observational data from infants enrolled in the EPIPAGE-2 cohort. EPIPAGE-2 is a French cohort study investigating outcomes of preterm infants born at 22-34 completed weeks’ gestation in all maternity units in 25 French regions. Only infants discharged home alive were deemed eligible. Predictive models will be built using a range of methods including multivariate logistic regression and penalised logistic regression. The predictive performance of the models will be established using 10-fold cross-validation. Model performance will be quantified using area under the receiver operating characteristic curve (AUROC), sensitivity and specificity, Tjur’s coefficient of discrimination and the Hosmer-Lemeshow goodness-of-fit statistic. Missing data in the dataset will be addressed via multiple imputation by chained equations (MICE) when appropriate.