P-298 Artificial intelligence (AI) score dynamics from day 3 to day 5: stability vs. progression in predicting clinical outcomes in IVF embryos

Abstract Study question Does the stability or progression of AI scores between Day 3 and Day 5 better predict pregnancy outcomes and aneuploidy in IVF embryos? Summary answer IVF preimplantation embryos with AI score progression, rather than stability, from Day 3 to Day 5 were more strongly associated with clinical pregnancy and euploidy. What is known already The predictive value of AI-driven models for pregnancy outcomes following both cleavage-stage and blastocyst-stage transfers has been demonstrated. However, the impact of score dynamics—specifically stability versus progression—between Day-3 and Day-5 of development remains largely unexplored. Understanding these dynamics could provide insights into clinical outcomes, particularly if different patterns of score changes can be quantified and tested for predictive performance. This study aims to evaluate the dynamics of Day-3 and Day-5 AI models in relation to pregnancy outcomes, examining how score differences influence pregnancy probability. Furthermore, we investigate the association between these score dynamics and euploidy. Study design, size, duration A single-center cohort study conducted at Hygeia IVF Embryogenesis between 2022-2024. 794 embryos with known implantation outcome (quantified via beta-hCG) and known AIVF AI scores calculated on Day-3 and Day-5 were used to assess associations of score differences with implantation. 481 embryos with known ploidy outcome (aneuploidy/euploid) and known AI scores were used to assess associations of score differences with euploidy likelihood. Participants/materials, setting, methods To account for the differing ranges of AI scores on Day 3 (1-5) and Day 5 (1-99), score ranges were first normalized to a 0-1 scale, with Day 3 scores divided by 5 and Day 5 scores divided by 10. Logistic regression was then performed on the normalized score difference between Day 3 and Day 5 to evaluate the association between score changes and the tested outcomes (implantation and euploidy, respectively). Main results and the role of chance Logistic regression of the normalized score difference between Day 3 and Day 5 revealed a significant positive association with pregnancy outcomes. The intercept of -1.3438 represented the log odds of pregnancy when Day 3 and Day 5 scores were identical (i.e., score difference = 0). The coefficient for the normalized score difference was 1.468 (p = 0.000567), which was both positive and significant, indicating that as the normalized score difference increased between Day 3 and Day 5, the likelihood of pregnancy also increased. The odds ratio (OR) of 4.34 suggested that for every 1-unit increase in the normalized score difference, the odds of pregnancy increased by a factor of 4.34 (95% CI: 1.91–10.19). A significant positive association with euploidy outcome was also observed. Coefficient for the normalized score difference was 1.2970 (p = 0.0121); OR was 3.65 (95% CI: 1.31-10.10). Given the influence of age on ploidy outcome, we further stratified the data by age (<37, 38-40 yrs old, >40 yrs old) to show that OR increased significantly (with appropriate P < 0.01) in the >40 age group (OR = 10.13), indicating that score progression between Day 3 and Day 5 may be especially relevant for this subset of older (>40) patients. Limitations, reasons for caution This single-center study limits generalizability to other populations. The analysis focused solely on Day 3 to Day 5 score dynamics, excluding other developmental stages and live-birth outcomes. Additionally, potential confounding factors such as culture conditions and demographics were not analyzed. Wider implications of the findings These findings suggest that AI-driven assessment of score differences and progression between Day 3 and Day 5 of in vitro embryo development could serve as a valuable predictor of pregnancy success. Broader validation across diverse populations is essential to confirm clinical applicability. Trial registration number No