AI and Assisted Reproduction: A Public Policy Issue

Here is a briefing document summarizing the main themes and important ideas from the research conducted by the Othering and Belonging Institute on "How AI is Changing Assisted Reproduction":

Briefing Document: AI in Assisted Reproduction

I. Executive Summary:

This research brief examines the increasing use of Artificial Intelligence (AI) in Assisted Reproductive Technology (ART), specifically In Vitro Fertilization (IVF). While AI offers the potential to improve IVF success rates and reduce the need for invasive procedures, the lack of robust regulatory frameworks raises significant ethical, financial, and equity concerns. The document highlights the historical context of IVF, the current state of AI applications in embryo selection and preimplantation genetic testing (PGT), and the existing policy landscape in the United States and internationally. It argues for proactive regulatory solutions to prevent the adoption of ineffective technologies, mitigate algorithmic bias, and ensure ethical oversight of AI in fertility medicine. The brief also raises concerns about the potential for a resurgence of eugenics.

II. Key Themes and Ideas:

• The Rise of AI in IVF:
• AI, particularly machine learning and deep learning, is being used to analyze embryo images and predict viability, potentially replacing or augmenting traditional methods of embryo selection and PGT. "New AI-enabled technologies claim to utilize images alone to select the embryo most likely to result in a live birth, and some companies claim to go further, stating that their image-based algorithms can conduct a type of preim-plantation genetic testing without the need for biop-sies."
• Time-lapse imaging has been used to visualize embryonic development, as early as 1929.
• Declining IVF Success Rates and the Appeal of AI:
• Despite advancements, IVF success rates have declined in the United States and peer countries since 2010.
• The high cost of IVF ($12,400 per cycle in the US in 2020) creates pressure for successful outcomes, making AI-enabled technologies attractive to both doctors and patients.
• Ethical Concerns and Algorithmic Bias:
• AI algorithms are susceptible to bias, potentially exacerbating existing inequalities in access to care and perpetuating discriminatory practices.
• There is a risk that algorithms could "learn" to select for or against certain traits without explicit consent or knowledge from the parents, raising eugenic concerns. "an algorithm learning from itself could ultimately deduce that a certain genetic trait is not favorable and select against this trait without consent or knowledge from the parents, without ever assessing the underlying genetic material."
• The disability rights community is particularly concerned about the potential for genetic testing to be biased against individuals with disabilities.
• Preimplantation Genetic Testing (PGT) and its Evolution:
• PGT has expanded from testing for single-gene mutations (PGT-M) and aneuploidy (PGT-A) to screening for chromosomal rearrangements (PGT-SR) and polygenic disease risk (PGT-P).
• PGT-P, which tests for polygenic traits and the likelihood of diseases like diabetes and depression, raises significant ethical questions about the selection of embryos based on predicted disease states and traits. "PGT-P has questionable analytic and clinical validity, as an embryo’s score is based off the genetics of the original study population of those who have lived long lives in specific environments."
• Concerns arise that the use of PGT-P, particularly among predominantly white and upper-middle-class IVF patients, could lead to a concentration of disability and disease in low-income and minority communities.
• Lack of Regulation in the United States:
• The US federal government does not directly regulate IVF. Regulation primarily occurs through the FDA (for drugs and medical devices) and CLIA (for clinical laboratories).
• The Fertility Clinic Success Rate and Certification Act (FCSRCA) of 1992 mandates that fertility clinics report success rates to the CDC, but enforcement is weak.
• The American Society for Reproductive Medicine (ASRM) and the Society for Assisted Reproductive Technology (SART) provide voluntary guidelines, but their self-regulation model is criticized as ineffective. "Despite criticisms that suggest SART’s self-regulation model may be ineffective, a study of SART’s methods showed that of forty-four clinics found to have at least one violation in 2019, thirty-four had resolved the violations by 2020."
• There is no standardized procedure to assist clinics in improving their success rates, and HHS has no ability to sanction or revoke the license of clinics with low success rates.
• Regulation of AI in Medical Devices:
• The FDA has two broad categories that AI in ART could fall into: Software in a Medical Device (SiMD) and Software as a Medical Device (SaMD).
• Most SaMD are Class II devices, but could be classified as Class III devices if they “support or sustain human life, are of substantial importance in preventing impairment of human health, or which present a potential, unreasonable risk of illness or injury.”
• The FDA's current regulatory framework is ill-equipped to handle the iterative nature of AI algorithms, which can change and adapt as they learn from data.
• The article mentions the Digital Health Innovation Action Plan geared toward resolving problems that relate to the faster iteration inherent in software-based technologies.
• International Comparisons:
• Other countries, such as the United Kingdom and European Union, have more comprehensive regulations governing embryo research and AI in medical devices.
• The UK has the Human Fertilisation and Embryology Authority (HFEA) to oversee clinics and provide guidance.
• The EU's proposed AI Act classifies AI algorithms based on risk levels and imposes stricter requirements for high-risk systems.
• The Influence of Private Equity:
• The acquisition of fertility clinics by private equity firms may lead to the "consolidation and industrialization" of the IVF field, potentially incentivizing the overutilization of adjunct treatments like PGT.
• Patients at private equity-affiliated clinics were more likely to use preimplantation genetic testing.
• Recommendations for Regulatory Solutions:
• The US should adopt regulations similar to those in peer countries, such as the UK's Human Fertilisation and Embryology Act, requiring licensing, inspection, and enforcement of clinic standards.
• The FDA should create policies that allow for software updates without requiring companies to undergo re-review, while still ensuring patient safety.
• The NIH should develop clear guidelines for embryo research, even in the absence of federal funding.
• HHS and the FDA should propose federal regulations that mandate embryonic stem cell research oversight (ESCRO) committees.
• Transparency and Eugenics:
• Lack of transparency in AI-driven embryo selection raises concerns about the potential for algorithms to select against certain traits without informed consent.
• The historical context of eugenics and its connection to early IVF research underscores the need for caution and ethical oversight. "Early interest in IVF stemmed from the eugenics movement’s desire to “improve the human race” through selective genetics."
• Appropriate AI Frameworks:
• The Biden administration introduced the “Blueprint for an AI Bill of Rights,” which recognized that “systems supposed to help with patient care have proven unsafe, ineffective, or biased,” and that disparities in AI frameworks should be assessed, mitigated, and monitored.

III. Key Quotes:

• "The scientist’s freedom to inquire is not immutable; society might again force scientists to consult institutions that were not developed for judging the motives of humanitarian biologists and physicians... Probably the worst consequence imaginable to scientists working in a political democracy would be the pre-emption by the state of a branch of science such as human embryology."
• "The majority of the currently FDA-approved AI algorithms have proceeded through 510(k) premarket notification or de novo pathway approval, but it is unclear how many of these algorithms have been resubmitted via 510(k)s for each change made to improve the functionality or performance of their devices."
• "THE PHENOMENON where policies in one country are influenced by existing policies in other countries is called policy diffusion and is a well-documented feature of modern global politics. "
• "None of the algorithms presented a “clinically relevant means” to aid in embryo selection, due to the “heterogeneity in the origin and culture of the embryos used for the development of [the algorithms]."
• "To some people with disabilities, the very field of genetic counseling and genetic testing is “inherently directive in a way that is biased against individuals with a disability,” as it causes parents to harbor concerns about embryos and their future disease state rather than an embryo’s potential."
• "[P]ublic health is committed to ending such inequities and reducing the incidence of disease that impacts people’s well-being."

IV. Policy Implications:

• The United States needs to develop a comprehensive regulatory framework for AI in ART to address ethical concerns, prevent ineffective technologies from being adopted, and ensure equitable access to care.
• Federal agencies, including the FDA and NIH, should establish clear guidelines and oversight mechanisms for embryo research and the use of AI in fertility medicine.
• Greater transparency and data sharing are needed to assess the efficacy and safety of AI-enabled ART technologies.
• The potential for algorithmic bias and eugenic practices must be carefully considered and mitigated.
• Policymakers should consider the role of private equity in the fertility industry and its potential impact on patient care and access to treatment.

V. Conclusion:

AI holds promise for improving IVF outcomes, but its rapid development and deployment require careful ethical and regulatory oversight. By proactively addressing the challenges outlined in this brief, policymakers can ensure that AI in ART benefits all individuals seeking to build families, while safeguarding against potential harms. Without proper regulation, there is a risk of exacerbating existing inequalities and perpetuating discriminatory practices.

Comment below to let us know your views on this article and issue within public sector policy making.