Healthcare data is usually stored at the individual level, with diagnoses, symptoms and risk factors associated in the medical record of each patient. However, the use of this data is often strictly regulated because it contains highly sensitive information.

In such cases synthetic data generation may offer a solution, where a generative model is trained on real data but is used to generate synthetic-but-realistic individual level data. It is still important that the synthetic data does not reveal sensitive information about real patients and using geolocation data (such as the location of an individual's GP) may add too much risk. GP locations can be used as proxies for patient locations due to people being registered to their nearest practice from home.

Previously, methods were developed to create realistic individual level synthetic data with geolocation that preserved privacy using aggregated GP data in conjunction with high-fidelity individual-level synthetic data. It was shown that by preserving key local distributions, realistic health prediction models could be learnt that reflected the original ground-truth data. This type of synthetic data may also help in building and monitoring regional simulations, improving the quality and effectiveness of local healthcare services. For example, it could help in simulating the spread, causes and effects of diseases by studying differences across localities.

This talk will discuss the use of AI methods for generating geolocated synthetic data from publicly available GP health statistics from the Quality and Outcomes Framework and high-fidelity individual-level synthetic data from the Clinical Practice Research Datalink as well as demonstrate its potential in health simulation models.

Guest Speakers: Dima Alattal and Allan Tucker, Brunel University of London

CPD Hours = 1 hour

Not a member? Join Now