Predicting live birth by combining cleavage and blastocyst-stage time-lapse variables using a hierarchical and a data mining-based statistical model

Abstract

Prolonged embryo culture is increasingly used as a way of improving pregnancy rates, especially in the context of single embryo transfer. So far, only a handful of studies examined the relation between implantation potential and time-lapse parameters extracted from later stages (morula and blastocyst) of embryo development. For this retrospective study all 285 single vitrified-thawed blastocyst transfers (SVBT) from all consecutive unselected patients whose fertilized oocytes were submitted to time-lapse monitoring (TLM) from a two-year cohort were analysed. Two different statistical models were created; a hierarchical one including the two strongest live birth (LB) predictors (t2 and texpB₂) and a more complex model based on principal component analysis (PCA) and logistic regression methods. The first, four-category, hierarchical model effectively distinguished between blastocysts of increasing LB rates (8, 30, 40, 53%). For the second data-mining model quartiles of the created Sc parameter had increasing LB rates (12, 19, 40, 49%). AUC values were comparable for both models (0.723, 95CI%:0.66-0.79 versus 0.717, 95CI%:0.65-0.78). The combination of cleavage- and blastocyst-stage variables through hierarchical or data mining-based algorithms was used successfully to predict live birth. However, due to the lack of internal / external validation the predictive capacities of this model could differ largely in different datasets.

Keywords: Blastocyst culture; Data mining; In-vitro fertilization; Predictive models; Time-lapse monitoring.