O-323 Sex-dependent discrepancies in automatic scoring: implications for the performance of embryo evaluation and selection methods

Abstract Study question Are artificial intelligence-based embryo evaluation and selection algorithms equally effective for male and female embryos? Summary answer Compared to female embryos, male embryos exhibited higher automatic scores, superior performance in KIDScore D5 v3, and poorer performance in iDAScore v2 in predicting implantation. What is known already It is known that the secondary sex ratio favors males in assisted reproductive techniques. The reason could be an unbalanced selection, as there is evidence that male embryos undergo faster development and have better morphology than female embryos. Based on this premise, artificial intelligence (AI) models currently used for embryo selection may not perform equally well for embryos of both sexes. We aimed to analyze the performance of two models for assessing male and female embryos: KIDScore D5 v3 and iDAScore v2. Study design, size, duration This is a retrospective cohort study involving 481 patients included in the preimplantation genetic testing program over a year and a half. The euploid embryos (n = 1,181) were divided into male (n = 577) and female (n = 604) groups and the automatic scores received by two artificial intelligence-based evaluation models (KIDScore D5 v3 and iDAScore v2) were considered with respect to implantation. Participants/materials, setting, methods All embryos underwent assisted hatching on day 3 of embryo development and trophoectoderm biopsy at the blastocyst stage (day 5 or 6). The embryos’ sex chromosomes were revealed using a highly validated method. Routine embryo assessment and selection were performed using conventional morphology according to ASEBIR criteria. Retrospectively, embryos were automatically scored from 1 to 9.9. Means, standard deviations, odds ratios (OR), and areas under the ROC curve (AUC) were calculated using IBMSPSS Statistics software. Main results and the role of chance In general, male embryos exhibited a significantly higher iDAScore than female embryos (6.9±2.2 vs. 6.4±2.5)**; no differences for KIDScore (6.1±1.7 vs. 5.9±1.8). This indicates that if the selection were based on the iDAScore, more male embryos would be transferred. Male embryos: 323 male embryos were transferred with an implantation rate of 53.8%. Implanted male embryos had higher automatic score compared to non-implanted ones (6.7±1.6 vs. 6±1.7 for KIDScore** and 7.3±2 vs. 6.8±2.1 for iDAScore*). For KIDScore: OR = 1.3, 95% CI 1.1-1.5 and AUC 0.612 (0.545-0.678). For iDAScore: OR = 1.1, 95% CI 1-1.3 and AUC 0.575 (0.511-0.638). This suggests a higher performance of KIDScore in predicting implantation in male embryos. Female embryos: 318 female embryos were transferred with an implantation rate of 47.7%. Implanted female embryos had higher iDAScore compared to non-implanted ones (7±2.3 vs. 6.3±2.5)*; no differences for KIDScore (6.2±1.8 vs. 6±1.8). For KIDScore: OR = 1.1, 95% CI 0.9-1.2 and AUC 0.542 (0.472-0.612). For iDAScore: OR = 1.1 95% CI 1-1.2 and AUC 0.584 (0.520-0.648). This suggests a higher performance of iDAScore in predicting implantation in female embryos. *pvalue0.05; **pvalue0.01 Limitations, reasons for caution The primary limitation of our study is the assisted hatching process performed on embryos on day 3 of development, which may impact late-stage development. This could be a possible reason for the considerably lower performance of the models. Wider implications of the findings Our study revealed objective differences between male and female embryos, as well as distinct performance for two AI-based embryo selection models. These findings suggest that the employed embryo selection model may impact the secondary sex ratio. Trial registration number Not applicable