Selecting the embryo with the highest implantation potential using a data mining based prediction model.

Area of Science:

• Reproductive medicine
• Embryology
• Biostatistics

Background:

• Embryo selection traditionally relies on subjective morphological assessments.
• Standard scoring systems (SSS) exhibit significant inter- and intra-observer variability.
• Computer-assisted scoring systems (CASS) offer a potential solution to SSS limitations.

Purpose of the Study:

• To develop and compare prediction models for embryo selection using SSS and CASS.
• To evaluate the impact of prediction models on embryo transfer decisions.
• To leverage data mining techniques for enhanced predictive accuracy.

Main Methods:

• Development of prediction models using logistic regression (LR) and multivariate adaptive regression splines (MARS).
• Evaluation of 871 embryos (2008-2013) with both SSS and CASS.
• External validation of models using a separate dataset (109 embryos, 2014).

Main Results:

• SSS-based models showed significant performance decline from training to validation datasets.
• CASS-derived models demonstrated stable predictive performance without significant decline.
• LR and MARS models from CASS achieved moderate discriminative capacity (c-statistic 0.64 and 0.69).

Conclusions:

• CASS enhances the generalizability of embryo selection prediction models.
• Integrating CASS with data mining predictive modeling is a promising strategy.
• This approach aims to improve the selection of embryos with higher implantation potential.