Key points are not available for this paper at this time.
Abstract Study question What is the respective performance of an AI model, a morphokinetic model and an embryo morphology model for the prediction pregnancy after single blastocyst transfer? Summary answer KIDScore™ D5 v3, NEQsi score, and iDAScore™ v1.0 demonstrated similar performance in predicting biochemical pregnancy, clinical pregnancy, and live birth. What is known already Blastocyst selection for transfer currently mostly relies on Gardner’s morphological classification, which suffers from relatively limited predictive performance and interoperator variability. Both morphokinetic scores and the development of AI tools raised hope for improved embryo quality assessment in IVF. The NEQsi score is a very recently published digital transcription of the Gardner classification, allowing easier data categorization and comparison with other scores / models. The respective performance of these models / scores has hardly been reported up to now. Studying this in a setting with systematic single blastocyst policy allows avoiding biases related to patients’ selection. Study design, size, duration This is a retrospective monocentric study conducted in 2020-2022 on single blastocyst transfers after culture in a time-lapse incubator (Embryoscope©). All embryos with a known clinical outcome after fresh or frozen transfer and with KIDScore™ D5 v3, NEQsi score, and iDAScore™ v1.0 were included, regardless of the ART technique. Participants/materials, setting, methods All patients gave their consent for anonymous analysis of their clinical data. All scores were compared by conducting a One-way ANOVA test followed by a non-parametric Kruskal-Wallis test. A ROC curve was calculated for the prediction of both biochemical, clinical pregnancy and live birth. Main results and the role of chance 1888 single blastocyst transfers were performed in 1223 patients, 907 resulting in a biochemical pregnancy, 626 in a clinical pregnancy, and 495 in a live birth. The mean values were 7.00 (± 1.64) for the KIDscore, 7.99 (± 1.30) for the NEQsi, and 8.50 (± 1.00) for the iDAscore. The mean score values were significantly different between groups according to positive or negative clinical outcome (p 0.0001 for all). The ROC curves showed that all 3 scores are predictive of a biochemical pregnancy (AUC KIDScore = 0.62; AUC NEQsi = 0.63; AUC iDAScore = 0.60), clinical pregnancy (AUC KIDScore = 0.64; AUC NEQsi = 0.65; AUC iDAScore = 0.61), and a live birth (AUC KIDScore = 0.62; AUC NEQsi = 0.63; AUC iDAScore = 0.60). These AUCs were not significantly different, and none of these scores appeared to be superior to the others in predicting live birth after single blastocyst transfer. Limitations, reasons for caution The retrospective and monocentric design calls for caution when generalizing these results. Wider implications of the findings The respective relevance of these scores in shortening time to pregnancy by ranking usable embryos within a cohort should be considered in further studies. Trial registration number not applicable
Reignier et al. (Mon,) studied this question.