P-264 Correlations between the artificial intelligence scoring system (iDAScore v1.0) and live birth outcomes in preimplantation genetic testing for aneuploidy cycles

Abstract

Abstract Study question Is the iDAScore v1.0 score correlated with pregnancy outcomes in single embryo transfer (SET) cycles following preimplantation genetic testing for aneuploidy (PGT-A)? Summary answer We demonstrated that elevated iDAScore v1.0 scores are positively correlated with the probabilities of pregnancy and live birth (LB) in SETs following PGT-A. What is known already To select the blastocysts with high development potential in advance, several groups have proposed the implementation of artificial intelligence (AI)-based software is capable of predicting implantation in women undergoing in vitro fertilization (IVF). It has also been known that using high resolution next generation sequencing preimplantation genetic screening (hr-NGS) for PGT-A can enable the exclusion of aneuploid embryos before embryo transfers and thus improve pregnancy outcomes of SETs. However, the capacity of AI-based assessments has remained unclear in IVF cycles with PGT-A. Study design, size, duration This retrospective study, approved by the Institutional Review Board of Chung Sun Medical University, was performed to assess the dataset of 317 SET cycles after PGT-A from January 2017 to September 2019. A single euploid or mosaic blastocyst with an AI score calculated using the iDAScore software (v1.0) was selected for frozen embryo transfer according to the given priority of blastocyst morphology. The iDAScore groups were categorized by quartiles of AI scores. Participants/materials, setting, methods Embryos were cultured in a time-lapse incubator and qualified blastocysts underwent next generation sequencing (NGS)-based PGT-A. The confounding factors associated with LB were evaluated using logistic regression analysis in generalized estimating equations (GEEs). The differences between iDAScore groups were assessed using the Mann–Whitney U test or Fisher’s exact test, as applicable. The receiver operating characteristic (ROC) curve analysis was used to estimate the predictive powers. A P value < 0.05 was considered statistically significant. Main results and the role of chance The results revealed the patient age, anti-Müllerian hormone, body mass index, oocyte sources (autologous or donor), PGT-A results (euploidy, low-level mosaicism, or high-level mosaicism), aberrant chromosome types (none, whole chromosome, segmental chromosome, or whole with segmental chromosome), and aberrant chromosome sites (0, 1, 2, or > 2) were not correlated with LB. However, embryo biopsy days (day 5 vs. day 6, OR = 2.271, 95% CI = 1.534–14.824), blastocyst morphology scores (OR = 1.185, 95% CI = 1.034–1.359), and AI scores (OR = 1.640, 95% CI = 1.231–2.183) were significantly correlated with LB. The AI scores were then divided into quartiles (group 1: 5.0–7.9; group 2: 8.0–8.5; group 3: 8.6–8.9; and group 4: 9.0–9.5). The ongoing pregnancy (29.1% vs. 54.2%–57.5%) and live birth (27.8% vs. 52.8%–56.3%) rates of group 1 were significantly lower than other groups. The blastocyst morphology score (5.3 ± 1.2), and KIDScore D5 scores (v1 = 3.0 ± 1.7; v3 = 4.6 ± 1.5) were also lowest in group 1. The ROC curve analysis confirmed a significant but limited LB prediction capability for iDAScore v1.0 (AUC = 0.6), which was similar to the KIDScore D5 (AUC = 0.61–0.62). Limitations, reasons for caution Because of the retrospective nature, the major limitation was the lack of randomization, which may present a risk of selection bias. According to the criteria of blastocyst selection for SETs, the embryos with low (< 5.0) or high (> 9.5) AI scores were not present in this dataset. Wider implications of the findings Although prediction capability of iDAScore remains perfectible, the AI score is still significantly associated with LB probability. Euploid or mosaic blastocysts with low AI scores (<8.0) processed a decreased LB rate, indicating the potential of annotation-free iDAScore system as a decision support tool for deselecting embryos with poor post-implantation development. Trial registration number not applicable