Medical informatics & machine learning for patient stratification & personalised prediction of outcome and clinical event in paediatric critical care

Routine clinical practices generate vast amounts of data, particularly in paediatric intensive care units (PICU), where high-resolution bedside monitoring data is continuously collected. Despite its potential, much of this data is discarded upon patient discharge, missing opportunities to extract valuable information that could be used for advancing research and improving the quality of patient care.

Recent advancements in artificial intelligence (AI) and machine learning offer the potential to explore this rich data and develop patient-centric approaches for paediatric critical care, ultimately enhancing clinical decision support systems and supporting the development of hypotheses in clinical research.

Previous studies have applied machine learning to preprocess multi-parameter physiological data and detect artefacts. Statistical approaches have also been used to identify the clinical and physiological phenotypes associated with improved neurological outcomes for critically brain-injured children. However, many efforts in predictive modelling rely on single-parameter vital signs, limiting the depth of insights that can be gained from the data.

This project seeks to overcome this limitation using medical informatics and machine learning algorithms to analyse multi-parameter physiological data. The aim is to predict key clinical events and the length of patient stay in the PICU, along with the outcomes for critically brain-injured children. The explainable nature of this analysis will allow for deeper exploration into the factors influencing these predictions, helping to better understand how clinical and physiological phenotypes affect patient outcomes.