P-262 Comparative ability of iDAScore and embryo morphology grade to predict clinical pregnancy: Retrospective cohort study of 1510 fresh and frozen elective single embryo transfers

Abstract

Abstract Study question Can iDAScore predict ongoing clinical pregnancy (CP) with a sensitivity and specificity equivalent to that associated with manual morphology assessment and grading? Summary answer The fully automated iDAScore was able to predict CP with equivalent performance to manual morphology assessment and grading in this retrospective cohort study. What is known already iDAScore is an artificial intelligence (AI) based algorithm developed by applying machine learning to morphokinetic time laps (TL) image data of embryos with a known treatment outcome. Embryos are automatically assessed on day 5 of culture and ranked according iDAScore, ranging from 1 to 9.9. Embryos may be prioritised for transfer based on highest sore. Along with other published AI algorithms, iDAScore has been proposed to optimise the chance of CP following ET by improving objectivity of embryo assessment compared with manual scoring systems. The fully autonomous assessment of embryos by AI algorithms has beneficial implications for laboratory workload. Study design, size, duration Retrospective audit of 787 fresh and 723 frozen single embryo transfer cycles which took place from April 2019 to September 2022. All recipient, surrogacy, warmed oocyte, embryo biopsy, cleavage stage and slow thaw frozen ET cycles were excluded. Participants/materials, setting, methods Selection for transfer was based on blastocyst morphology grade. iDAScores were obtained retrospectively. The area under the receiver operator characteristic (AUROC) curve and sensitivity and specificity for CP prediction of both iDAScore and blastocyst morphology grade was compared overall and in two stratified analyses. The first assessed fresh and frozen cycles separately, the second compared their performance in female age groups of ≤ 35 and >35 years. CP was defined by ultrasound detection of foetal heartbeat. Main results and the role of chance The mean grading of blastocysts by iDA was 8.34 ± 1.4, with a strong correlation to classic morphological scoring (r = 0.69, P < 0.001). The clinical pregnancy rate for fresh and frozen embryo transfer was 31.0% (95%CI 27.9-34.3) and 44.0% (95%CI 40.8-48.0) respectively. iDA score was positively associated with clinical pregnancy rates in both fresh (adjOR 1.69, 95%CI 1.44-1.99) and frozen embryo transfers (adjOR 1.45, 95%CI 1.26-1.67), independent of maternal age. There was no difference in the AUROC for iDA (AUC 0.64, 95%CI 0.62-0.67) versus conventional morphology (AUROC 0.63, 95CI 0.61-0.66) when all eSETs were considered, or when fresh eSETS (AUROC for fresh iDA 0.66 95%CI 0.62-0.70 vs morphology 0.65 95%CI 0.62-0.69) or frozen iDA 0.63 95%CI 0.59-0.67 vs morphology 0.61 95%CI 0.57-0.64) eSETS were considered separately. The iDA score exhibited slightly better performance in the age stratified analyses, with a higher AUROC in women >35 years; iDA 0.68 95%CI 0.64-0.71 vs morphology 0.64 95%CI 0.60-0.67, p = 0.021) but no difference was observed in younger women. This age difference in iDA performance was primarily driven by frozen embryo transfers (p = 0.002). For women >35years an iDA score of 8.75 was associated with an AUROC of 0.63 and 67% sensitivity and 60% specificity for prediction of clinical pregnancy. Limitations, reasons for caution This was a retrospective single centre study and the performance of iDA needs to be confirmed prospectively in a multi-centre trial. That selection of embryos for transfer was based on morphology, may have contributed to overestimation of model performance. Wider implications of the findings The application of AI based embryo ranking algorithms to embryo selection for transfer, has the potential to maintain clinical pregnancy rates while reducing the time burden associated with conventional morphology / morphokinetic assessments. Trial registration number Nil