P-377 Artificial Intelligence (AI) for Pre-implantation Genetic Test for Aneuploidy (PGT-A): important epistemic and ethical considerations

Abstract

Abstract Study question What are the epistemic and ethical considerations if it's used AI for PGT-A? Summary answer The AI used to date presents significant epistemic and ethical problems and there are no studies evaluating its clinical efficacy. This topic requires more attention. What is known already The introduction of time-lapse systems offers the possibility of generating large amounts of data to evaluate the state of health of embryos. With the introduction of AI and machine learning this data can be analyzed by algorithms that improve automatically as they are exposed to more data. Almost all AI is based on algorithms developed using neural networks. Many algorithms are difficult to understand or even uninterpretable because they are covered by company secrets. Not having access to the algorithms causes problems of an ethical nature and caution is required in its use for the identification of healthy embryos. Study design, size, duration We performed a study of the articles that mention a possible use of AI as a tool to be used within the PGT-A. The epistemic and ethical implications of current approaches have been considered. We also considered the oral communications presented during the last ESHRE annual meeting and the sensationalist communications associated with them. Participants/materials, setting, methods We used PubMed for article search which includes more than 35 million citations for biomedical literature from MEDLINE, life science journals and online books. To carry out a targeted search we used the following keywords: Artificial Intelligence* OR AI OR Neural Network* OR Machine Learning OR Support Vector Machine OR Automated Classification AND IVF OR IVF OR Embryo OR Pre-implantation Genetic OR PGT*. Main results and the role of chance The results obtained were difficult to interpret. We found articles where they looked at 2 types of results: accuracy for predicting euploidy and agreement with the molecular assay assessment. It has been demonstrated that some algorithms are not always able to totally differentiate a euploid from an aneuploid embryo, especially when speaking of segmental aneuploidies. This data lends many concerns. Almost universally, AI models were opaque in that at least part of the process was not accessible. If not fully accessible, these models are problematic for epistemic and ethical reasons. Epistemic concerns include information asymmetries between algorithm developers and all the operators who routinely perform these analyses. There is another risk of biased predictions caused by known and/or unknown confounders during the algorithm auto-improvement process. Furthermore, we must consider the high difficulty in checking for any errors in real time due to the total non-accessibility. The ethical trap includes: the risk of misrepresenting important values for the health of the embryo and the patient; the concern of incorrect deselection or worse a risk to the health and well-being of future children. There are also possible social implications; and a hole of responsibility, in case of adverse events. Limitations, reasons for caution Our search was limited to a single major medical research database and to what was presented during the last ESHRE annual meeting. Although there are other databases where it might be possible to find different articles. The application of AI in PGT is a very recent topic. Wider implications of the findings It is premature to implement AI for the identification of aneuploidies in embryos. AI for aneuploidy embryo deselection has enormous potential but needs to be done with care and transparency, as epistemic and ethical issues are significant. Currently the only possible use is to identify priority embryos for the transfer. Trial registration number not applicable