L26/P-742 Validation of an AI-Driven Predictive Model for Ovulation Trigger Timing in ICSI Cycles

Abstract Study question Can an AI-driven predictive model for ovulation trigger timing standardize follicular response and align oocyte retrieval outcomes with ovarian reserve markers in ICSI cycles? Summary answer The AI-driven ‘Trigger Day Predictive Score’ ensures a predictable physiological response, evidenced by stronger correlations between hormonal markers and oocyte yield, despite similar absolute outcomes. What is known already The timing of the ovulation trigger is a critical decision in assisted reproductive technology (ART), directly influencing the number and quality of oocytes retrieved. For patients with diminished ovarian reserve or discordant follicular growth, this decision is particularly challenging. Current clinical practice relies heavily on clinician experience and generalized protocols, which may lead to inter-operator variability and suboptimal outcomes. An AI-driven predictive model offers a potential solution by providing personalized, data-driven recommendations to standardize trigger timing, thereby aligning the retrieval outcome with the patient’s biological potential. Study design, size, duration This was a prospective observational study conducted between June and December 2025. The study included 101 women undergoing ICSI cycles. Patients with uterine abnormalities or severe endometriosis were excluded to isolate ovarian response variables. Participants/materials, setting, methods Comprehensive data, including demographics (age, BMI), reserve markers (AMH, AFC), and cycle parameters (follicle sizes, E2, LH, Progesterone), were collected. An AI model calculated a (TDPS) score for each patient, forecasting MII yield. All patients received a fixed antagonist protocol. Trigger timing was determined by physicians blinded to the TDPS score. Participants were retrospectively stratified into “Concordance” (physician agreed with AI) and “Not-concordance”. The endpoint was the alignment of (MII) yield with reserve markers. Main results and the role of chance Analysis of 101 ICSI cycles revealed that baseline demographics (age, BMI, AMH) were comparable between the Concordance (n = 79) and Not-concordance (n = 22) groups (p > 0.05). Estradiol (E2) levels at trigger were notably higher in the Concordance group (1950.0 ± 1328.6 pg./mL) compared to the Not-concordance group (1348.5 ± 738.3 pg./mL), showing a strong clinical trend (p = 0.074). Critically, the Concordance group demonstrated a physiological alignment where MII yield was significantly correlated with AMH (r = 0.435, p < 0.001) and E2 levels (r = 0.343, p = 0.003). This significant biological correlation was absent from the Not-Concordance group, suggesting erratic responses when AI recommendations were not met. regarding clinical outcomes, no statistically significant differences were observed in total oocyte yield, embryo quality, or clinical pregnancy rates (57.0% vs. 50.0%, p = 0.561) between the two groups. This indicates that while AI standardization does not artificially inflate yield beyond biological reserve, it ensures the retrieval reaches the patient’s expected potential. Limitations, reasons for caution The study was conducted at a single center with a relatively small sample size (n = 101), which may limit generalizability. Additionally, observational nature precludes establishing causality, necessitating future randomized controlled trials to validate the model’s clinical utility. Wider implications of the findings The validation of the TDPS suggests that AI can revolutionize ovulation trigger management by replacing subjective clinical judgment with a data-driven score. This tool has the potential to standardize care, reducing inter-clinician variability and ensuring that oocyte retrieval outcomes are consistently aligned with the patient’s reserve, particularly in complex cases. Trial registration number Yes