Estimating the number of oocytes to expose to sperm in assisted reproductive technology treatment.

This study was approved by the Amherst College institutional review board (#19–012). Patients who initiated their first stimulation cycle between January 1, 2014, and December 31, 2020, at a single, large, university affiliated ART practice (Boston IVF) were eligible for inclusion in the study. Cycles with planned or implemented pre-implantation genetic testing (PGT), day 3 transfer, use of donor oocytes, and/or a gestational carrier were excluded. Patients were randomly separated into equal-sized training and testing sets. The algorithm was developed as described earlier ( 5 ) with three main differences. First, we used a more precise definition of useable blastocyst rate: the proportion of mature oocytes (ICSI cycles) or inseminated oocytes (conventional IVF cycles) yielding blastocysts that were transferred fresh or cryopreserved. Second, we included a larger set of covariates, including predictors not available in the national database ( Supplemental Table 1 ). Third, instead of predicting day of transfer/freeze, we only included women with blastocyst transfers/freezes since this decision would be known at the time of using the tool at this particular clinic. Operating characteristics were compared between the SART CORS models fit to the BIVF cohort and the BIVF models fit to the BIVF test set.

The median patient age at stimulation was 34.9 years (IQR: 31.9, 38.0) among the 8,279 stimulation cycles with a blastocyst transfer or cryopreservation ( Table 1 ). The median blastocyst rate was 0.26 (IQR: 0.13, 0.41). Across multiple day 5 transfers, 74.9% of women were observed to have a live birth. The clinic-specific model was similar at predicting blastocyst rate as compared to the national model ( Table 2 ). Like the SART CORS algorithm, among women with intended blastocyst transfer or freeze, the BIVF algorithm recommended exposing fewer than all oocytes to sperm most often to women 42 years (5%) ( Table 2 ). Among cycles recommended to expose fewer than all oocytes to sperm, the median patient-specific number of oocytes to expose increased with increasing age, from 5 oocytes among women 42 years. Performance was similar among IVF and ICSI cycles ( Supplementary Table 2 ).

Increased efficiency in embryo cryopreservation in assisted reproductive technology (ART), with the advent of vitrification, has seen a rise in frozen embryo transfers along with a concurrent increase in surplus embryos ( 1 ). These can become an emotional and financial burden for patients and a logistical and financial burden for clinics ( 2 – 4 ). Recently, a prediction tool was developed to estimate the optimal number of oocytes to fertilize during ART, such that an adequate number of embryos would be created to have the desired number of live births while minimizing the number of surplus embryos ( 5 ). The tool was developed using national data from the Society for Assisted Reproduction Technology Clinical Outcome Reporting System (SART CORS). In large national databases, specific details are often lacking; for example, total number of oocytes was available but not the number of mature oocytes. Furthermore, there is concern regarding ART practice heterogeneity across the nation, possibly limiting the utility of a prediction tool developed using pooled national data which may not be useful at the level of an individual clinic. The current study investigates the validity of those concerns. We developed a similar tool using data from a single high-volume ART center which had information on additional parameters that were not available in the national database. We then compare the accuracy of this tool using data from this single clinic with the tool developed using the national registry.

This single-center study corroborates the findings of the tool derived from SART CORS data. We developed the same tool using data from a homogenous practice and with additional predictors and did not find notably different accuracy in the predictions or recommendations. Our study is limited by the relatively small number of women over 40 (n=794). Nonetheless, these findings support the widespread applicability of the online tool developed using the national database in contexts where patients wish to limit the creation of surplus embryos.