Artificial Intelligence: is it good for our health?

Stefano Gortana

6 December 2017

On 21 November, the House of Lords Select Committee on Artificial Intelligence welcomed two panels of experts to discuss how AI is being used in healthcare, its potential future applications, and questions such as whether the NHS has the capacity to take advantage of this innovative technology, and what ethical standards may be required in its development and use. PHG Foundation’s Dr Sobia Raza joined Dr Julian Huppert, Chair of the Independent Review Panel for DeepMind Health and Nicola Perrin, Wellcome Trust’s Head of Understanding Patient Data. They were followed by a panel comprised of National Data Guardian Dame Fiona Caldicott, Dr Hugh Harvey and Professor Martin Severs.

With six experts from diverse backgrounds and specialties, the discussion of AI in healthcare was predictably far-reaching. While the witnesses shared their broad knowledge of issues ranging from data sharing and public opinion to algorithms and commercial value, a number of key themes and common observations regarding the applications of AI in healthcare did emerge. The UK possesses a unique and incredibly valuable resource in NHS patient data, but lacks the infrastructure and governance landscape that is needed to fully capitalise on AI in healthcare.

AI in healthcare: early days, enormous potential 

The panellists highlighted the fantastic impact that AI is already having in image analysis, where algorithms are facilitating rapid and accurate detection of disease. They were also quick to stress that AI is expected to complement the role of clinicians rather than replace it. Nicola Perrin welcomed the development of AI-driven clinical support tools to help clinicians spot patterns more easily or help choose the best care pathway for patients, while Dr Huppert argued that AI will also help combat the ‘hyper-specialisation’ of clinicians by identifying rare interactions that a non-specialist might easily fail to spot. PHG Foundation’s Dr Raza expanded on this, stressing that as datasets become richer over time, our understanding of disease will deepen, with algorithms facilitating the stratification of patients and ultimately the personalisation of healthcare.

Barriers to capitalising on AI in healthcare

As AI and algorithms depend so heavily on the collection, storage and sharing of patient data, there is a clear need to involve the public in the development of this industry. However, there is little public knowledge or understanding of AI, and there is a lack of clarity regarding the ethical standards of its use and even the range of applications. Most crucially, patients must consent to their data being collected, used and shared. Unfortunately, the general public still does not understand how their data is being used in the NHS now, let alone for future AI purposes. This makes data collection and sharing more difficult, with public concerns over privacy and security. There is a great need to improve general knowledge and understanding not just of AI and its applications, but also of the value and uses of patient data more generally.

It was also reiterated that the NHS is not currently configured to capitalise on AI. For example, hospitals are not yet e-enabled; staff not sufficiently knowledgeable; data-sharing arrangements are inconsistent; and no clear governance structures are in place. Furthermore, the NHS also lacks sufficient capacity. The implementation and use of AI technology requires a multi-disciplinary approach, as well as access to the relevant technical expertise, including skilled data scientists. This implies a considerable reorganisation of the NHS that will take time and resources - both of which are in short supply.

It’s all about data

The Committee will have certainly come away convinced of the great value of patient data, as well as the comparative advantage the UK has over other countries due to the wealth of such data currently held by the NHS as the single national healthcare provider. But, as Dr Huppert pointed out, this data is not in effect held in a single location; rather, different organisations within the NHS hold unique datasets in data ‘silos’. The panellists agreed that NHS data would be considerably more valuable if it was held in a modern, secure, shared and digital database.

Dr Raza added that whilst the NHS may possess a wealth of patient data, the development of healthcare algorithms depends on commercial partnerships, and shared access to data. If the UK is to capitalise on its wealth of patient data, the development of AI and the underlying algorithms will need to be a collaborative undertaking between the NHS and industry based on cross-sector data-sharing.

Towards effective and secure data sharing: a question of governance

There was general agreement that anonymised NHS data should not be made publically available, for a variety of reasons including patient privacy, security and the imperative to ensure that the UK benefits economically from this valuable resource. The panellists instead called for a clear regulatory system to monitor access to and use of patient data.

The Committee was particularly surprised to hear that, in the absence of a coherent data-sharing governance structure, NHS Trusts have been negotiating different reward arrangements with different companies and doing so without any communication between or even within individual Trusts. This provided some insight into the causes of data scandals like London’s Royal Free hospital breach of UK data law earlier this year.

When questioned about possible methods of regulating data access and sharing, the panellists were generally supportive of the recommendations made by Dame Wendy Hall and Jerome Pesenti in the government-mandated review Growing the Artificial Intelligence Industry in the UK, which included creating ‘data trusts’. These ‘proven and trusted frameworks and agreements to ensure exchanges of data are secure and mutually beneficial’ are currently being explored by the Government and industry. As Dr Raza observed, the particular style of system is not as important as the details on how data will be stored and used; how to avoid duplication and how to incorporate health organisations already sharing data. With data featuring strongly in the Industrial Strategy White Paper and further details emerging in the Government’s response to John Bell’s Life Sciences Strategy, it will be interesting to see whether the Government understands these challenges and responds effectively.

Read the full PHG Foundation consultation response on AI in healthcare

Have a look at our Healthcare Futures infographic on Medical bots