Development of a Robust D5 Embryo Ranking System: A Retrospective Multi-Centre Study Involving 8866 Single Fresh Embryo Transfers

Abstract

Background: Inter-operator variation in morphology-based embryo assessment is known to be an ongoing issue in in vitro fertilisation (IVF). This includes inconsistent annotation of embryo morphology and disagreement in the predictive weightings amongst different morphological features. Although currently underexplored, subjectivity reduction using statistical tools is considered clinically valuable, and is more affordable than newer approaches such as time-lapse imaging. Aim: This study aims to develop a robust numerical D5 embryo ranking system by incorporating several patient and embryo characteristics arising from routine static D5 assessment. Method: A total of 8866 autologous-oocyte single fresh D5 transfers performed at 14 Monash IVF clinics during 2012-2018 were retrospectively analyzed, with repeat cycles from the same patients excluded to avoid clustering effects. The dataset was randomly split into two subsets at 60:40 ratio, including one (n=5274) used for regression analysis and model development and the other (n=3592) for validation. Multiple logistic regression was performed to evaluate live birth prediction by several variables, expressed by odds ratio (OR) and 95% confidence interval (CI). A numerical ranking system was developed based on calculated weightings of contributing factors and subsequently tested using receiver operating characteristics (ROC) as expressed by area under the ROC curve (AUC). Results: All included variables were significantly predictive of live birth, including maternal age (OR=1.465, 1.364-1.574, P=0.000), assessment timing (OR=1.096, 1.020-1.177, P=0.012), embryo developmental stage (OR=1.341, 1.244-1.445, P=0.000) and morphological sore (OR=1.520, 1.392-1.661, P=0.000). Increments of resultant numerical ranking scores ([Formula: see text]3, 3-3.99, 4-4.99, and [Formula: see text]5) were associated with increasing live birth rates (9.4%, 27.2%, 43.6%, and 54.9%; respectively). ROC analysis indicated significantly predictive power of the numerical ranking system in both the development (AUC=0.690, 0.675-0.704, P=0.000) and validation (AUC=0.685, 0.667-0.703, P=0.000) subsets. Conclusion: A robust numerical model can be developed for D5 embryo selection, potentially leading to improved inter-operator subjectivity for embryo assessment.