#347 : Factors Related to Mosaicism of Human Embryo: Conventional Statistics and Machine Learning Analysis

Background and Aims: Mosaicism arises from errors in the cell division process and can lead to cells containing a combination of normal and mutated genes. Mosaic embryos can produce healthy babies. However, it may be associated with a lower implantation and a higher miscarriage rate. Despite extensive research, the scientific understanding of the underlying causes of embryonic mosaicism remains limited, and conclusions remain subject to ongoing debate. Machine learning (ML) can capture complex relationships and automatically learn the most informative features, while conventional statistics (CS) are simple to understand. We aimed to apply both ML and CS to determine the critical factors that may be associated with mosaicism.

Method: A retrospective analysis was conducted at the Infertility Department of Hung Vuong Hospital between 2018 and 2022 using medical data from 51 IVF couples. The study utilized both machine learning and conventional statistical methods to identify possible factors associated with mosaicism in human embryos.

Results: Out of the 246 embryos studied, 84 (32%) were found to be mosaic. The performance of the classifiers in the machine learning analysis was relatively poor when using all features, with F1-scores ranging from 0.08 to 0.26. While feature selection helped to improve performance, the F1-scores remained low. On the other hand, the conventional statistical analysis revealed that patients with cryptozoospermia showed a positive association with mosaicism (OR 2.68, 95% CI 1.08 to 6.66, p=0.033).

Conclusion: According to the machine learning algorithms, the current set of clinical features has no relation to the incidence of mosaicism. However, the conventional statistical analysis suggests that patients with cryptozoospermia may contribute to a higher risk of mosaicism. Further research is needed to fully understand the mechanisms underlying mosaicism in human embryos.