External validation of a time-lapse prediction model

Objective To study the performance of a previously published implantation prediction model based on morphokinetics in a different setting, in an unselected population and with various embryo transfer strategies. Design Retrospective monocentric study. Setting University-based assisted reproduction technology (ART) center. Patient(s) 450 unselected couples undergoing intracytoplasmic sperm injection (ICSI) cycle with embryo culture in the EmbryoScope (Unisense Fertilitech), corresponding to 528 embryos with known implantation. Intervention(s) None. Main Outcome Measure(s) Implantation rates (IR) in embryo categories defined by the model in the overall population and in subgroups according to the day of embryo transfer. Result(s) The distribution of IR among detailed morphokinetic categories in the overall population and in subgroups according to the day of embryo transfer was more heterogeneous than expected according to the published model. The distribution corresponded better to the original when a simplified version of the model was used, although it worked better in the cleavage-stage group than in the blastocyst-stage group. Conclusion(s) This study was unsuccessful in replicating the sensitivity of the previously published model for predicting implantation rate of embryos ranked according to morphokinetic categories. Further work is required to assess the utility of the model for embryo selection. Each team using time-lapse technology should build a center-specific prediction model based on its own data and transfer policy.