A machine learning-based score to predict live birth after mosaic embryo transfer

This study aimed to develop an artificial intelligence-based scoring system to prioritize mosaic embryos according to live birth outcomes.

This multicentre, observational, retrospective study included 264 transferred mosaic embryos from 2583 PGT-A (Preimplantation Genetic Testing for Aneuploidies) cycles performed between January 2017 and January 2023. Trophectoderm (TE) biopsies from day-5 (D5) or day-6 (D6) blastocysts were analysed using Next-Generation Sequencing (NGS) (VeriSeq, Illumina®, San Diego, CA, USA). Biopsied embryos were vitrified and subsequently transferred. Clinical, embryological, and laboratory variables were collected to build predictive machine learning models for live birth. Models excluding cohort-invariant variables were refined to derive the final scoring system.

Among mosaic embryos, biochemical, clinical pregnancy, and live birth rates were 50.75%, 41.66%, and 36.36%, respectively. The best-performing model, validated through tenfold cross-validation, identified embryo quality and biopsy day as the most influential predictors (42% and 34% weights, respectively), while mosaicism-related factors such as monosomy/trisomy (18%) and mosaicism degree (6%) had lower influence. The resulting score suggests prioritizing high-quality embryos biopsied on D5, as the type and level of mosaicism play a minor role in gestational potential, except when comparing embryos of similar quality where lower mosaicism levels and absence of monosomy are advantageous.

The AI-derived score highlights embryo quality as the primary determinant of success in mosaic embryo transfer, supporting prioritization of high-quality, D5-biopsied embryos to improve live birth outcomes in ART.