P-131 Predicting Embryo Utilisation Rate on Day 5 using an Artificial Neural Network: A Multicentre Retrospective Study

Abstract

Abstract Study question Can artificial intelligence (AI) identify which patient and cycle specific variables are most important for predicting blastocyst utilisation rate (BUR) on day 5? Summary answer Number of mature oocytes (MII) injected is the most significant variable for predicting BUR. Highest BUR association with MII is found when six MII injected. What is known already The day 5 usable blastocyst rate originating from the accumulation of embryos being transferred and cryopreserved is one of the most informative key performance indicators (KPI) in an IVF laboratory. As per the Vienna consensus, the blastocyst development rate competency value is set as ≥ 40% with the benchmark value set as ≥ 60%. Artificial intelligence (AI) has been integrated into clinical settings as a technological advancement aimed to improve clinical success rates. Identifying the predictors that can result in an increased BUR on day 5 can be an additional tool towards improved cycle outcomes. Study design, size, duration In this retrospective, multicentre study, we evaluated six variables for predicting BUR using an artificial neural network (ANN). Study was performed in two fertility centres, between March 2021 and August 2022. A total of 865 cycles were included. Cycle exclusion criteria: Preimplantation genetic testing (PGT), advanced maternal age (AMA) > 38, infertility aetiology of endometrial origin, severe oligospermia and epididymal/testicular sperm extraction. Primary outcome measure was the BUR on day 5 for the different variables. Participants/materials, setting, methods A total of 865 cycles were analysed using the ANN model with six variables: Number of MII oocytes injected (ICSI), use of autologous/donated gametes, maternal age at oocyte pickup (OPU), sperm concentration, progressive sperm motility rate, and fertilisation rate. Cycles were divided into training and test set through stratified random sampling: 73.2% (633) training and 26.8% (232) test. BUR on day 5 was grouped into <60% and ≥60% as per Vienna consensus benchmark values. Main results and the role of chance Number of MII oocytes was found to be the most important variable (100%) followed by the type of gametes used (54.1%), sperm concentration (32.9%), age at OPU (24.7%), progressive sperm rate (23.4%) and fertilisation rate (17.1%).The effect of MII injected on BUR was then investigated and results indicated an inverse correlation, with increasing MII injected resulting in decreased BUR (correlation coefficient of r = 0.344,p<0.001). According to the performed model, 6 injected MII produces the higher rate of utilisation (62.9%). Performance of ANN model was assessed by positive predicted value (PPV), negative predicted value (NPV), false positive rate (FPR), false negative rate (FNR), overall accuracy (OA), sensitivity and specificity. PPV for the training and test sets was 62.2% (194) for the <60% EUR on day 5 group and 79.8% (256) for ≥60% EUR on day 5 group and 64.8% (83) for the < 60% EUR on day 5 and 81.7% (85) for ≥60% EUR on day 5 group, respectively. OA obtained from training and test sets were 71.1% and 72.4% for the <60 and ≥60% EUR on day 5 groups, respectively. The area under the curve (AOC) to predict the UR on day 5 group was found to be 77.6%. Limitations, reasons for caution The results represent the experience gained from current practice and not of a prospective controlled study. Clinical outcomes using this approach were not explored. Several other patient and cycle variables can also be investigated. Wider implications of the findings The number of oocytes retrieved is associated with live birth rate with 15 being the ideal maximum. Increasing oocyte numbers however, can bear greater risks. Predicting embryo utilisation rate on Day 5 can guide the way towards personalised treatment and safety. Trial registration number Not applicable