Ada Lovelace Institute warns NHS against deploying genomics AI
A report has found that widespread use of artificial intelligence-powered genomic health prediction in the NHS could lead to privacy and ethical issues, discrimination and dependency on the private sector
The government must be careful when looking to use artificial intelligence-powered genomic health prediction (AIGHP), and should avoid deploying it nationally, according to a report.
The report by the Ada Lovelace Institute and the Nuffield Council on Bioethics warned that although the use of artificial intelligence (AI) and genomics is often seen as a holy grail that could transform the NHS, it is currently too risky to plan a widespread roll-out of AIGHP.
Advances in AI and genomics have the potential to be a game changer for the NHS, and are often seen as a key enabler of more preventative and resource-efficient care. The idea of being able to tailor medical care through precision-based medicine and get early insights into disease risks is often met with enthusiasm, but it still comes with huge risks.
“The use of AIGHP to improve understanding of how an individual might respond to a given drug or medication could allow for better prescribing practices, reduce waste, improve outcomes and avoid harmful side effects,” the report said.
“We know that prevention is better than cure: better for people and far less costly to health services. We also know that earlier interventions have better outcomes. By providing people with better insight into their individual genomic health risks, AIGHP could enable people to better protect and promote their health, allowing them to stay healthier for longer, with far less reliance on expensive, curative interventions.
“However, these benefits are not guaranteed. Large-scale deployment of AIGHP brings financial, ethical and service-level risks, and the science underlying these techniques is still being developed. The NHS will need to approach the deployment and cultivation of AIGHP deliberately and carefully if the benefits are to be realised.”
The report added that despite the excitement surrounding AIGHP, there is “substantial disagreement in the scientific community” regarding the accuracy and utility of these types of systems.
AIGHP uses polygenic scoring, which assesses the collective impact of multiple genetic variations on the likelihood of a person developing a certain disease, compared to the rest of the population. However, most of these scoring systems are currently trained on datasets representing people with European ancestry, meaning that if used, they would likely be inaccurate for other members of the population. Another issue is that with many diseases, genomic variations only account for a small proportion of disease risk.
“There is no consensus on whether these difficulties can be overcome, or in what timeframe. Some argue that many of them will be resolved as datasets expand, become more diverse and improve in granularity, and as analytical techniques improve. Others maintain that the difficulties around polygenic scoring are more fundamental and cannot be resolved by improvements to scale, detail or sophistication,” the report said.
Privacy issues
The potential deployment of AIGHP could also lead to a whole host of privacy issues. For predictions to be successful, the system would require large amounts of sensitive data, which also delivers insight into people’s future characteristics, and hence poses several ethical questions concerning privacy, surveillance and new forms of discrimination.
The sensitive data could also inadvertently create information about people the data subject is related to, which brings further ethical challenges, such as consent. The report calls for the government to create a “more granular model of consent” where people can specify exactly which data they want shared.
“This model should be used for patients sharing their genomic data for research or clinical purposes and for research participants. It should provide a new set of standardised options that are structured to enable people to explicitly opt out of particular uses of data, including sharing data with particular entities,” the report said, adding that it would also need reforms of UK data protection law, which “should clarify how to interpret the UK General Data Protection Regulation (GDPR) definition of healthcare data” and “strengthen, rather than weaken, protections around the repurposing of genomic and phenotype data for research purposes”.
It must also make it clear that there is a legal ban on the use of genomic data by UK insurers, rather than the current voluntary option.
AI-powered genomic health prediction has the potential to offer us a lot, but this report clearly highlights that we are not ready to fully embrace it, and nor is it ready to deliver on its promises
Sarah Cunningham-Burley, Nuffield Council on Bioethics
The report said the NHS must be careful when deploying AIGHP as there is the potential to lose control of the delivery or terms of healthcare provision. Deploying systems like these in the NHS would need large amounts of data, compute and AI expertise, which isn’t found in the NHS, meaning it would have to look to the private sector.
“That comes with challenges such as vendor lock-in, poor terms of access in the long run and the difficulty of auditing proprietary systems that will have a material impact on NHS decision-making,” the report said.
“In addition to dependency, our research suggests that some approaches to using AIGHP in the NHS could reduce the resilience and effectiveness of the service. One risk is if the NHS channels money away from conventional, reactive care (or more conventional approaches to disease prevention) to the use of AIGHP systems, believing that such systems will reduce healthcare demand in the long run.
“If AIGHP systems prove unable to reduce demand to the degree expected, the NHS could find itself with a gap between unreduced demand and reduced supply of reactive services.”
The report added that as the NHS currently does not have good evidence, or a clear and credible plan to address the risks associated with a roll-out, it must currently refrain from doing so.
Nuffield Council on Bioethics chair, Sarah Cunningham-Burley, said that while it makes sense to look to technology to help the NHS provide better interventions, “we must ensure that we are not rushing ahead before a full assessment of the benefits and harms has been made”.
“AI-powered genomic health prediction has the potential to offer us a lot, but this report clearly highlights that we are not ready to fully embrace it, and nor is it ready to deliver on its promises,” she said.
“We must also take a step back and engage the public fully so that we can ensure the use of AIGHP in our NHS is not only wanted, but also trusted. Only by embedding ethical considerations from the outset will AIGHP reach its full potential.”
Read more about the NHS and technology
Health secretary Wes Streeting wants the UK to be an international powerhouse for life sciences and medtech, and plans to turn the Department for Health and Social Care into a ‘public growth department’.
The Labour Party wants to create an NHS powered by ‘groundbreaking technology’, with genomics testing for every baby, modernising appointment booking and overhaul incentives for tech adoption.
