Synthetic Data

Background

Critical care data is routinely collected as part of patient care, but for privacy reasons this data cannot be easily shared with researchers or between hospitals. Across the NHS large volumes of this data exist but individual datasets are small and not always easy to access due to their complexity.

Intended outcomes

This project develops methods to unlock the value in siloed healthcare data to realise benefits to patients and the NHS. Dr McWilliams’ team will develop a synthetic data generator – an artificial intelligence (AI) model – that can be trained across small datasets without any patient data having to leave its hospital of origin. The model will generate data about ‘synthetic’ adult and neonatal patients which can be used for research such as training predictive models, whilst preserving patient privacy.

Neonatal intensive care unit provides the prototypical use-case. At University Hospitals Bristol and Weston NHS Trust there are only 592 significantly premature ventilated infants in the local dataset. The team will demonstrate how their synthetic data can be used to train a model that predicts successful extubation for these patients. This is a key part of de-escalating care which is poorly predicted by existing clinical tests.