Blastulation and ploidy prediction using morphology assessment in 33,999 day-3 embryos

Although contemporary practice in in vitro fertilization (IVF) favors embryo transfer at blastocyst stage, several centres worldwide employ cleavage-stage Day-3 embryo transfers. The advantage of cultures extended to Day-5 and beyond, is to ensure that the embryo to be transferred will not arrest between Day-3 and Day-5, and that it provides additional morphological quality markers that can be used for selection. To bridge that gap for centres that practice Day-3 transfers, we intended to model the association between Day-3 morphology and blastulation/ploidy tested with modern sequencing technologies, and to develop a validated predictive model for these outcomes. We conducted a retrospective cohort study including 33,999 Day-3 embryos from 5,702 cycles between March 2017 and December 2021 at ART Fertility center, Abu Dhabi. Day-3 embryos were evaluated for cell number, and degree of fragmentation. Expanded blastocysts with existent inner cell mass (ICM) and trophectoderm (TE) cells underwent TE biopsy for PGT-A by next generation sequencing (NGS). The primary objective of the study was to develop prediction models for blastocyst biopsy and euploidy by using Day-3 embryo morphology and patient characteristics. The final models for euploidy and blastulation prediction included Day-3 cell count and fragmentation, and female age. Both models were well-calibrated in validation samples to differentiate blastulation (intercept: 0.087 ± 0.034, slope: 0.827 ± 0.022) and euploidy (intercept: 0.02 ± 0.015, slope: 0.913 ± 0.024) potential of embryos. The machine learning (ML) model obtained higher correct call scores in all performance targets (blastulation, euploidy) when compared to selecting the embryo with highest cell count and lowest fragmentation (P < 0.0001). In conclusion, the developed prediction models ( https://artfertilityclinics.shinyapps.io/GEMMA-D3B/ ) showed its potential in prioritizing Day-3 embryos likely to develop into high-quality and euploid blastocysts. The models can be used by any laboratory set-up without needing specialized equipment or software and is available for further performance validation.