Optimizing risk predictive strategies in febrile neutropenic episodes in children and young people undergoing treatment for malignant disease

The aim of this thesis was to investigate the clinical problem of the initial management of febrile neutropenia (FN) in children and young people undergoing treatment for malignant disease, to thoroughly evaluate the existing research, and to collect and synthesise this to quantify the risk of adverse clinical outcomes, through development of develop a new risk prediction model, using individual participant data (IPD). A further aim was to develop methodological approaches to IPD analysis in the development of predictive models, including the graphical display and communication of such information. The research produced five systematic reveiws of the existing medical literature in this area, and helped create a global collaboration of 19 research groups (PICNICC) which has shared data on over 5000 episodes of FN. This individual patient data was synthesised using hierarchical logistic regression meta-analysis to develop a new predictive model for MDI, which is robust to internal validation techniques (bootstrapping and leave-one-out cross-validation). The multivariable predictive model derived had six components: Tumour type, temperature, clinical description of being “severely unwell”, and the results of measurements of three elements of the full blood count: haemoglobin concentration, total white cell count and absolute monocyte count. It showed good overall fit (Brier[scaled] 4.5% discordancy), moderate discrimination (AU-ROC 0.736) and good calibration between predicted and actual estimates of the risk of MDI (calibration slope 0.95). We have demonstrated that such a data sharing project is feasible across many different jurisdictions and eras of study, and we now need to undertake a series of further projects to evaluate the model and go on to improve the management of paediatric FN across the world.