NEW YORK – A new report from the UK suggests artificial intelligence-powered genomic health predictions (AIGHP) could exacerbate societal issues and lead to privacy and discrimination concerns, though experts spelled out a series of recommendations aimed at mitigating these risks and boosting the potential benefits of the tools.
The "futures" research report looked at the possible effects that AI-powered genomic health predictions might have in the UK's National Health Service (NHS) health system, particularly when it comes to genomic predictions based on polygenic risk scores.
It was penned by Ada Lovelace Institute researcher Harry Farmer with contributions by Maili Raven-Adams and Andrew Strait, and released by the Ada Lovelace Institute and the Nuffield Council on Bioethics on Wednesday.
"The project seeks to anticipate, assess, and navigate the potential impacts of the convergence of AI and genomics over the coming five to 10 years," Farmer explained.
Though he noted that AI-powered genomic health prediction "is not yet the most common approach" for polygenic risk score analyses and conceded that "polygenic scoring can be conducted without AI," Farmer argued that AI-informed approaches are expected to ramp up significantly in the future.
Along with a look at potential benefits of AI-powered genomic health predictions — including actionable insights into disease risk for individuals and populations, improved treatment targeting, more tailored healthcare utilization, and an increased focus on disease prevention approaches — the authors delved into possible risks of the technology such as privacy concerns, genetic discrimination, and the potential adoption of predictive approaches with yet-to-be-defined scientific certainty.
"Large-scale deployment of AIGHP brings financial, ethical, and service-level risks," the report explained, "and the science underlying these techniques is still being developed."
With these considerations in mind, the report highlighted 10 main recommendations for incorporating AI-based genomic health predictions into NHS-based healthcare in the future, which ranged from an emphasis on genomic data as a form of personal data to the need for related data protection laws for genomic data, biometric data, and corresponding research efforts.
Recommendations also addressed approaches for coming up with consistent, nuanced consent models, for example, while flagging organizations that may be tasked with performing public engagement efforts, insurance code development, and work on genomic discrimination legislation.
The team also cautioned against the use of AI-powered genetic health predictions tools at the population level in the UK. Rather, the report suggested that it may be more beneficial to initially limit the approach to specific, targeted settings.
"Our evidence suggests that while it has the potential to improve healthcare outcomes, [AI-powered genomic health prediction] may currently be an ineffective tool for mass disease prevention and reducing healthcare demand at a population level," Farmer wrote, noting that "wide deployment of AIGHP across the population could create greater exposure to the risks associated with the technology — and greater costs — in exchange for uncertain benefit."