About Piraye Yurttas Beim, PhD
Thank you for this interesting paper! Its great to see further data supporting better outcomes in FET cycle with elevated P in the fresh cycle, this time in freeze all FET cycles.
In our data sets, using ROC curves to define the P threshold, we found a cutoff of 1.04, which identical to yours. In our data set, that would classify 39% of patients as having a positive test. Is that percent similar to yours?
We also find similar low AUC for P in predicting live birth. Which is always interesting to me given the strong effect of P is in regression and GEE models on live birth. Just goes to show how multifactorial live birth as an outcome is. Even very important predictors like age, stage of transfer, quality of embryo, and P have what would be consider poor AUCs by conventional classification.
Thank you! We are prospectively tracking data trends across multiple clinics in the US, and one major shift in protocols that we saw was that more and more clinics are moving to freeze-only cycles. We thought it was important to understand and report the implications for patient outcomes. It’s interesting that you found an identical P threshold in your data analysis as well. Not sure we can give you an apples-to-apples comparison on the question about percentage of patients that have a positive test. The data set that is included in this study was generated using a propensity score matching algorithm to create a matched fresh cohort for comparison to the freeze-only cohort. Accordingly, 66% of the cohort have P greater than 1 ng/mL, which reflects the fact that a freeze-only strategy is utilized more frequently with patients that have elevated P. When we look at the full data set of IVF cycles prior to matching, we see that overall 43% of patients who had P measured during their cycle had P greater than 1 ng/mL. This varies by center with rates ranging from 30% to 60%, possibly due to differences in testing platforms and/or patient selection for testing, as the proportion of patients at a clinic that had P tested at trigger ranged from <10% to nearly all patients. This variability highlights the importance of clinic-based evaluation and calibration of these thresholds.
We completely agree with your comments about low AUC for P in predicting live birth and the multifactorial nature of live birth. It is one of the reasons that we feel that physicians should increasingly become comfortable leveraging multi-variable predictive models in their clinical decision making and patient counseling. So many patients are still counseled based primarily on age or a handful of predictors, like the ones you mention. Thanks again for your comments. If you feel that our Polaris Data Network could be helpful to serve as an extra layer of independent validation for any other trends you are studying at your center, please feel free to reach out to us.