P-122 Evaluation of an updated artificial intelligence embryo viability model on implantation and miscarriage

Abstract Study question Is the implantation and miscarriage prediction performance of iDAScore model v2.0 increased as compared to original v1.0 model? Summary answer The predictive is better for than on both miscarriage, although differences did not reach statistical significance. What known already In recent years, several artificial intelligence (AI) models embryo based image or time-lapse data have been published validated in various ways. Some these are commercial products while others more an academic nature. long-term benefit AI clinical practice relies their use routine that continuously monitored updated. has available from Vitrolife since 2021 external validations published. An updated version (iDAScore v2.0) recently released. design, size, duration This was observational retrospective study with collected during years 2019 2022 one Danish clinic fresh single transfers performed day 5. Blastocysts were selected transfer by experienced embryologists morphology (Gardner score). Within period, we included all 849 embryos which had information hCG test results, (fetal heartbeat (FH) detected ultrasound) live birth (LB). Participants/materials, setting, methods Embryos cultured EmbryoScope+ incubator (Vitrolife A/S). Time-lapse images automatically captured throughout incubation period. We used calculate scores retrospectively. For each version, (early pregnancy loss miscarriage), FH LB expressed terms area under ROC curve (AUC). Differences between versions assessed paired comparisons AUCs. Main results role chance median age recipients 34 years. One third blastocysts derived ICSI remaining standard IVF cycles. Among transferred blastocysts, rate 58.0% (492/849), 46.5% (395/849), 42.4% (360/849), overall 26.8% (132/492).The AUC 0.652 (0.616 0.689) 0.632 (0.595 0.669) v1.0. Similarly, 0.644 (95% CI: 0.607 0.681) 0.626 0.588 0.663) However, statistically significant (P-values: 0.12 0.17 LB) comparisons.For found increase (0.574; 95% 0.514 0.633) (0.592; 0.534 0.649), P-value: 0.40. Limitations, reasons caution only includes a clinic. Patient characteristics procedures followed might be representative other clinics. limited chosen trained embryologists. Furthermore, sample size does provide performance. Wider implications findings It important constantly improved. Updated should take into account larger database well potential new methodologies evolve rapidly developing field. shows update seems represent actual increase. Trial registration number 1-16-02-277-22