Prediction of implantation after blastocyst transfer in in vitro fertilization: a machine-learning perspective
Using a machine-learning approach to predict ongoing implantation after blastocyst transfer improves the performance significantly compared with classic multivariate logistic regression models.
Volume 111, Issue 2, Pages 318–326
Celine Blank, M.D., Rogier Rudolf Wildeboer, M.Sc., Ilse DeCroo, M.Sc., Kelly Tilleman, M.Sc., Ph.D., Basiel Weyers, B.Sc., Petra de Sutter, M.D., Ph.D., Massimo Mischi, Ph.D., Benedictus Christiaan Schoot, M.D., Ph.D.
To develop a random forest model (RFM) to predict implantation potential of a transferred embryo and compare it with a multivariate logistic regression model (MvLRM), based on data from a large cohort including in vitro fertilization (IVF) patients treated with the use of single-embryo transfer (SET) of blastocyst-stage embryos.
Retrospective study of a 2-year single-center cohort of women undergoing IVF or intracytoplasmatic sperm injection (ICSI).
Department of assisted reproductive medicine of an academic hospital.
Data from 1,052 women who underwent fresh SET in IVF or ICSI cycles were included.
Main Outcome Measure(s)
The performance of both RFM and MvLRM to predict pregnancy was quantified in terms of the area under the receiver operating characteristic (ROC) curve (AUC), classification accuracy, specificity, and sensitivity.
ROC analysis resulted in an AUC of 0.74 ± 0.03 for the proposed RFM and 0.66 ± 0.05 for the MvLRM for the prediction of ongoing pregnancies of ≥11 weeks. This RFM approach and the MvLRM yielded, respectively, sensitivities of 0.84 ± 0.07 and 0.66 ± 0.08 and specificities of 0.48 ± 0.07 and 0.58 ± 0.08.
The performance to predict ongoing implantation will significantly improve with the use of an RFM approach compared with MvLRM.