P-301 Assessment of ongoing clinical outcomes prediction of an AI system on retrospective SET data

Abstract

Abstract Study question Would patients with fresh and frozen embryo transfer had achieved pregnancy before if the embryo was chosen by AI? Summary answer CHLOE (AI) can predict pregnancy, ongoing clinical pregnancy, and miscarriage following a single embryo transfer (SET). What is known already The use of time-lapse incubators has provided embryologists with more information to evaluate embryo development, resulting in varying clinical practices among clinics in prioritizing this information. However, manual annotation of time-lapse videos is time-consuming and prone to interoperator inconsistencies. To overcome these challenges, AI tools like CHLOE (Fairtility) can be utilized. CHLOE uses AI-based predictors to predict implantation and provides clarity on the biological factors driving these predictions. However, before incorporating AI tools into clinical practice, it is important to validate their effectiveness. Study design, size, duration Single study center that took place between July of 2021 and December 2022 at a private clinic in Spain. This was a retrospective cohort analysis that reviewed 118 time-lapse videos from single fresh embryo transfers and 92 time-lapse videos from single frozen embryo transfers with known ongoing clinical pregnancy outcome. CHLOE EQ score and CHLOE Rank efficacy of prediction of clinical outcomes and miscarriage was quantified using the metric AUC. Participants/materials, setting, methods Time-lapse videos were evaluated using CHLOE (Fairtility), an AI tool, to determine CHLOE EQ score and rank related to clinical outcomes (biochemical pregnancy, clinical pregnancy, and miscarriage) following fresh and frozen SET. CHLOE rank and embryology were compared with chi-square and AUC was calculated with logistic regression to measure prediction accuracy. T-test was used to check differences in CHLOE EQ score in different outcomes. Main results and the role of chance Embryologist vs CHLOE Ranking weren’t significant (p > 0.05). In fresh SET the mean EQ score was 7.76, and in frozen SET was 7.07. Following fresh SET, CHLOE EQ score was not-significantly predictive of biochemical pregnancy (AUC = 0.53, n = 104, p = 0.462), clinical pregnancy (AUC = 0.51, n = 79, p = 0.949), and miscarriage rate (AUC = 0.50, n = 68, p = 0.949); CHLOE Ranking was more predictive than embryologist rank for biochemical pregnancy (embryologist vs CHLOE rank: AUC = 0.51, p > 0.05 vs AUC = 0.61, p > 0.05), clinical pregnancy (embryologist vs CHLOE rank: AUC = 0.51, p > 0.05 vs AUC = 0.70, p > 0.05) and miscarriage rate (embryologist vs CHLOE rank: AUC = 0.51, p > 0.05 vs AUC = 0.75, p > 0.025). Following frozen SET, only top 1 embryos ranked by embryologists were transferred, and CHLOE EQ score was predictive of biochemical pregnancy (AUC = 0.60, n = 85, p = 0.213), clinical pregnancy (AUC = 0.64, n = 60, p = 0.919), and miscarriage rate (AUC = 0.87, n = 52, p = 0.437); CHLOE Ranking was predictive of biochemical pregnancy (AUC = 0.59, p > 0.05), clinical pregnancy (AUC = 0.64, p > 0.05) and miscarriage rate (AUC = 0.90, p > 0.05). Limitations, reasons for caution This study is a single-center retrospective analysis where embryos were chosen for transfer by human embryologists and is part of a broader effort to validate the responsible integration of AI into clinical practice. Wider implications of the findings The use of AI-based tools has the possibility to enhance the consistency, efficiency, and effectiveness of embryo selection. The information from quantitative and qualitative morphokinetics provided by AI tools like CHLOE brings greater clarity to predictions, enabling more personalized care for each individual embryo. Trial registration number Not Applicable