A robust clinical-laboratory AI model for predicting cumulative live birth per oocyte retrieval as a benchmark for evaluating emerging embryo selection technologies.

OBJECTIVE

To evaluate the performance of a clinical-laboratory artificial intelligence (AI) model for predicting cumulative live birth rate (CLBR) per oocyte retrieval and to assess its utility as a benchmark for emerging embryo selection technologies.

METHODS

This retrospective cohort study included 89 completed ICSI cycles (2023-2024) from a single center. The multivariable AI model, based on hierarchical logis-tic regression, integrates female age with non-linear pe-nalization; serum AMH; number of metaphase II oocytes; fertilization rate; blastocyst formation rate and quality; presence of male factor infertility; and distinct probability coefficients for fresh and frozen embryo transfers. CLBR was calculated per oocyte retrieval, incorporating both fresh and frozen embryo transfers, representing the most clinically meaningful outcome for patients. Model discrim-ination was assessed using receiver operating character-istic (ROC) curves, with area under the curve (AUC) and 95% confidence intervals (DeLong's method). Accuracy, sensitivity, and specificity were determined at a 50% prob-ability threshold.

RESULTS

Mean patient age was 37.2±4.3 years, AMH 2.86±2.1 ng/mL, and number of MII oocytes 7.8±3.2, fer-tilization rate averaged 79±18%, blastocyst formation rate 58±24%, mean blastocyst quality score 2.1±0.9, fresh blastocysts transferred 1.8±0.9, and cryopreserved blasto-cysts 2.1±2.0. Male factor infertility was present in 34% of cycles. The model achieved an AUC of 0.90 (95% CI 0.82-0.95). At the 50% threshold, balanced accuracy was 82%, sensitivity 82.2%, and specificity 81.8%. The confusion ma-trix revealed 7 false positives and 5 false negatives.

CONCLUSION

This clinical-laboratory AI model provides accurate prediction of cumulative live birth per oocyte re-trieval and establishes a validated benchmark for critical-ly evaluating emerging embryo selection technologies. Its use may help ensure that technological innovations are assessed against biologically grounded outcomes before widespread clinical adoption.