When the opportunities data present to the UK are discussed, healthcare generally tops the list. Whether it is for research or sharing between services, opportunities exist to make the sector more efficient and prepare it for future challenges.

But at the same time, there are threats posed by using data irresponsibly. Health data is highly personal and sensitive, and its misuse, whether stolen or deployed without oversight, can have serious consequences for victims.

Balancing these two sides is a major concern for legislators. In recent months, the UK Government has revealed the Goldacre Review along with its health data strategy, Data Saves Lives: Reshaping Health and Social Care with Data. These aim to promote innovation and improve the country’s healthcare system while keeping people’s data protected.

To better understand the issues, DIGIT spoke with Chief Executive of Lenus Health Paul McGinness. Lenus Health is an Edinburgh-based digital health company. Its technology is used across multiple NHS Scotland health boards to support virtual wards and clinical decision making in both the diagnosis and management of long-term conditions.

The UK’s Health Data Potential

Among the Goldacre Review’s observations is the “unfathomable depth and potential” of the data the NHS has access to. But this comes with the challenge of managing it to ensure it is useable, stored safely, and properly communicated.

The report warns, however, that the current system relies too much on fragmented projects using large volumes of duplicated data. This increases costs, reduces innovation, and weakens security.

While this system was adequate as a means of getting by in a pre-digital age, it is preventing the healthcare sector from scaling its approach to using data.

“Where some of the problems lie in relation to best use of data in operational practice relate to how data is shared between different healthcare organisations,” McGinness notes.

“How patient data is shared between community and secondary care is a big issue. As a result, an encounter a patient has with parts of the health system can often be an unconnected episode of care. It’s not on the continuum of your care as an individual because the data can be fragmented and siloed.”

Safe Havens

While there are drawbacks, there is good reason behind this arrangement – organisations have their own legal entities and governance arrangements, making safe data sharing difficult.

One solution for this is the use of Trusted Research Environments (TREs), or Safe Havens, which provide places to aggregate data from across the health system for research.

Among the Goldacre Review’s suggestions was the concept of the data pact between the public and the healthcare system. This gives patients the right to access their data and to either provide or deny consent for that data to be used for research.

“Using technology to allow staff to spend more time with patients, while giving patients greater access to their data are laudable commitments to the new strategies around TREs,” McGinness says.

He added: “There’s still tension between using TREs to provide safe access to data for researchers and how consent for using patient data for research and model training is gathered and stored. Furthermore, the point around how patients can revoke this consent must also be addressed.

“There remains an issue around how researchers and data scientists can exploit the latest cloud technologies and data science tooling to develop their models in TREs. TREs can often have limitations and are restrictive on what software can be deployed.

“It means that there’s tension between access to data, and also exploiting the use of the more modern tooling available that allows researchers to do the data analysis work quicker and more effectively, that still remains unresolved.”

Trust and AI

Both the UK and Scottish Governments have created strategies around using AI to ensure the technology benefits society while avoiding potential pitfalls.

But meeting these aims requires trust in the technology. There are, often reasonable, concerns around the use of AI – bias, opacity, and the how data is used.

“Healthcare is a very high risk setting for deploying AI,” McGinness notes. “It’s important that any data analysis and AI work is fair and ethical.”

One advantage the UK has is that public trust in the NHS is relatively high. This, in turn, translates into people generally being comfortable with sharing their data with the organisation.

A UK Government report from December last year found that 84% of people who trust the NHS to act in their best interests were happy to provide personal information to the NHS to develop new healthcare treatments.

But that there are still holes in public trust. The same report found that over half of its respondents said they knew little or nothing about how their data was used and collected.

This is why transparency and accountability are vital. If people are to trust their data is being used appropriately, they need to be able to see how and why it’s being used.

“We need to be as transparent as possible,” McGinness says. “That means having interpretable, explainable models much more appealing than the black box approach you often get with neural networks.

“A lot of the features that Lenus Health develops in our models are clinician-driven, and that’s really important for ensuring a high degree of explainability. When we present risk scores to a clinician, we give as much context as possible. We present the global explainability for the model, outline what the key features are driving the model performance, and then we provide a local explainability for the individual patient.”

Empowering People

It is vital to ensure that AI is used as a tool to support human decision-making, not as a substitute. The 2020 English A-Level scandal, where an algorithm downgraded millions of grades, showed the dangers of allowing Ais to make life-affecting decisions for people.

Transparency and explainability help clinicians understand why the system has made its recommendations, ensuring it is used to support staff.

“That allows the clinicians to interrogate the model score to ensure it is bio-plausible,” McGinness says. “They’re not just looking at scores, they’re looking at what data has changed to influence the model score.

“That enables them to understand the prediction, so the clinician still makes the final decision.”

Safe havens are another part of fighting bias in the use of AI. Too often, bias slips into algorithms not through malice, but due to a simple lack of data or oversight. It’s easy to find anecdotal examples of this, like Amazon’s hiring algorithm overly favouring men because it was using previous, biased, data to inform its decisions.

Recommended

• DIGIT Movers and Shakers | July 2022
• Scots fintech BR-DGE announces partnership with travel counsellors
• AMTD: Is this really the latest meme stock?

As such, providing algorithms with a variety of information is essential to preventing bias.

“It’s extremely important to take your models out of where you’ve trained them and train them on other datasets with more diverse populations to evaluate model performance. Breaking down model performance across gender, age and ethnicity is also worthwhile in combatting bias,” McGinness says.

With data-sharing, models can be trained in one safe haven with its data and then validated in another safe haven. Ultimately, the best way to defeat bias in AI is to use more, and higher quality, data.

For the UK’s health services, ensuring strong data sharing is key to preparing for future challenges.

“For us,” McGinness concludes, “that will enhance clinical practice, improve health outcomes, and make available rich structured data to support development of machine learning models and algorithms that can be used for clinical decision support freeing up clinicians to deal with more complex cases.”

Get the latest news from DIGIT direct to your inbox

Our newsletter covers the latest technology and IT news from Scotland and beyond, as well as in-depth features and exclusive interviews with leading figures and rising stars.

To subscribe, click here.