Novel machine learning model developed for embryo selection for frozen embryo transfer (FET) and to calculate appropriate number of embryos recommended for embryo banking.

Abstract

Embryo selection for FET is routinely based on subjective parameters, and in most modern IVF labs quantitative approach for embryo selection is not available. In order to determine which factors affect FET pregnancy rates, a de-identified data set was acquired, including 1254 patients and 8595 total embryos. 1233 embryos were transferred, 802 (65%) resulted in pregnancy, 665 (54%) in clinical pregnancy, and 641 (52%) had positive FHT’s.

The data set was analyzed to identify correlations between variables and pregnancy success rates, and analysis of variance (ANOVA) confirmed their significance. Age of the female partner, embryo grade, day of blastocyst development, biopsy technique, and PGTa were identified as factors to affect pregnancy (P<0.05). Using the factors with the greatest impact on pregnancy rate, the data set was modeled using machine learning, Resulting models demonstrated AUCs of 0.626, 0.630, and 0.584 for logistic regression, Least Absolute Shrinkage and Selection Operator (LASSO), and random forest, respectively, which is consistent with existing IVF literature. The relatively low values of AUC are due to the fact that most implanted embryos in the data set have been used after PGTa, which greatly increases chances of FET success. These models were then used to create a graphical app where inputting various information (age, embryo grade, day of development, PGTa) can calculate implantation potential, along with a confidence interval for each embryo. Another graphical app was created to calculate potential cumulative chances of pregnancy for the specific set of embryos, which can be used to optimize embryo banking for families that pursue fertility preservation due to cancer diagnosis, genetic diseases, and the need for PGT-A, PGT-SR and PGT-M.

These new models can be used clinically to select the embryo with the highest implantation potential, which can significantly improve the chances of implantation and decrease the time to pregnancy.