PETs in the NHS

Data has become a bedrock of NHS services in the UK. From managing appointments at an out-patient clinic, to using artificial intelligence to read a chest X-ray, data plays a pivotal role in delivering safe, timely, and cost-effective health services. 

What is a PET?

To improve NHS services, and to meet the challenges of the current economic climate, research and innovation is of vital importance. But can data controllers share sensitive, private health data AND support research in the NHS?

The Information Commissioner's Office has published advice on what it calls 'Privacy Enhancing Technologies' or 'PETs'. This can help organisations share data in compliance with data protection laws. PETs support sharing sensitive data for research, without revealing information about a person's identity.

PETs help us show 'data protection by design', and one PET we have been exploring at the West of Scotland Safe Haven in Glasgow is synthetic data.

What was the synthetic data project?

Research Data Scotland funded a collaborative project last year – through the Systems Development Fund - between the Safe Haven in Glasgow, and DataLoch in Edinburgh.

Together, we looked at how we could create synthetic health datasets to support different data science tasks. For example, how can we show cases of data with non-NHS employees when they are learning about data processes and tasks in the NHS?

We do not want to use real patient NHS data for a job like this. Instead, we can use synthetic data. Synthetic data looks like real data. It preserves some of the statistical properties and distributions of a real dataset. However, it is made of entirely fictitious data-points. It looks like real-world data, but it contains zero information about the identities of individuals in the world.

For our project, experienced data scientists tested different ways of creating synthetic health data. We explored different scenarios and methods for checking the quality of the synthetic data. We also developed guidelines for data controllers, describing safe ways to create and share datasets using PETs like synthetic data.

What did we learn?

Our project decided that very simple, 'low-fidelity' synthetic datasets can be really helpful for teaching students, or helping programmers plan computer code for health applications. Low-fidelity means that the data only resembles real-world data in a very basic way. Simple distributions, like the number of people aged over 50 in a population, are preserved. 

But more complicated relationships, like the number of people over 50 that have heart failure AND use a steroid inhaler AND have attended hospital as an outpatient in the last three years, are not represented. Keeping synthetic data simple and ‘lo-fi’ is a good way to guarantee that real patient details are never revealed.

Straightforward sharing licenses, and clear guidance for data controller and users, can also offer reassurances when using synthetic health data.

PETs like synthetic data are a great way to promote data science and research in the health sector. We are planning to create some small synthetic datasets to share with NHS staff and postgraduate students in Glasgow to teach them about health data science. Users will be able to safely test computer code with our synthetic data, without having to access any real-world patient information.

Safe, secure access to data will help drive improvement in the NHS. Synthetic data helps us kick-start projects and learning, without compromising on our duties under data protection regulations.

All views expressed within this blog are those of the author. With thanks to Dr Charlie Mayor of the West of Scotland Safe Haven for authoring this blog and sharing it with Research Data Scotland for publication.