Mapping performance in 114,237 IVF cycles with donor oocytes reveals reproducible inefficiencies and distinct endophenotypes

BackgroundIn vitro fertilization (IVF) outcomes remain variable and difficult to predict, even in individuals without diagnosed infertility. This uncertainty complicates clinical decision-making and impedes the identification of biological factors that underlie IVF success or failure. A major barrier to progress is the absence of standardized criteria to distinguish true biological inefficiencies from random variation or technical noise in IVF procedures. ObjectiveTo develop a robust statistical framework for defining IVF under- and overperformance in a fertile population, enabling precise phenotyping of IVF inefficiencies and laying the foundation for future genomic and mechanistic studies. MethodsWe conducted the largest retrospective cohort analysis to date of IVF outcomes in 114.237 oocyte donation cycles, from 32.492 rigorously screened donors who underwent 84.228 stimulation cycles. IVF performance was evaluated across four developmental stages: oocyte maturation, fertilization, blastocyst formation, and preimplantation development efficiency. Donors were classified as statistical under- or overperformers using a combination of binomial testing and mixed-effects logistic regression models, adjusting for identified confounders. Reproducibility of performance classifications was assessed through multi-cycle consistency analyses and supervised predictive modelling. ResultsDespite stringent fertility screening, 0.9% to 2.1% of donors were identified as statistically underperformers in individual developmental metrics, with similar proportions observed for overperformers. Minimal overlap between donors falling into significant categories across different developmental metrics supported the interpretation of IVF performance as comprising biologically distinct endophenotypes. Underperformance was reproducible across multiple cycles in a subset of donors (P < 0.001), indicating a likely biological basis rather than random or procedural variability. Logistic regression models demonstrated modest predictive ability for future outlier status (AUC > 0.61 and < 0.71), further supporting the presence of stable, donor-specific traits. Interestingly, comparison of donor clinical variables revealed a possibly clinically meaningful difference in the distribution of donors that had a documented live birth prior to donation between performance groups for all developmental metrics, indicating a potential association with proven fertility and better IVF outcomes. Finally, 109 donors across performance groups were surveyed, revealing no identifiable differences in natural fertility outcomes, further reinforcing that IVF inefficiency does not necessarily reflect underlying infertility. ConclusionThis unprecedented and methodologically rigorous analysis demonstrates that statistically significant deviations in IVF outcomes can occur even among presumed fertile donors. These findings challenge the assumption that poor IVF outcomes are inherently indicative of infertility and support a shift toward defining IVF failure through biological, rather than purely clinical, criteria. Our standardized phenotyping framework offers a powerful tool for identifying the molecular and genetic contributors to IVF inefficiency and for advancing personalized reproductive medicine.