Estradiol

The demographic data analyzed included age and body mass index. Cycle data included protocol type (divided into antagonist based, long luteal, Lupron stop, flare, or mini-IVF/ natural cycle), the number of follicles bucketed into groups of 5-mm diameter measured at monitoring visits during ovarian stimulation, estradiol level (Roche, Indianapolis, Indiana), and endometrial thickness (measured by transvaginal ultrasound in millimeters). Missing values were replaced by zero, and no normalization was performed. To improve the performance of our algorithm, we additionally calculated the following features from existing data: estradiol levels per follicle >16 mm in diameter and the total number of follicles, since these were found to be additive to the model's performance. All data were anonymized and stripped of personally identifying information.
The primary outcome measures were average outcome improvement in 2PNs and total usable blastocysts. The model was built and optimized to maximize the total number of 2PNs, and then tested on its ability to maximize both 2PNs and total usable blastocysts.

Steroidal hormones were extracted from serum before the assay. Briefly, serum was thawed on ice; diethyl ether 5 v/v was added and thoroughly mixed. This mixture was centrifuged under 1000 g for 3 min to separate phases. Upper layer (ether) was then stocked on a clean tube and was repeated twice for optimal extraction. A precision curve was run for a coefficient value of r² > 0.96. Estradiol and testosterone assays (Roche Diagnostics GmbH) were performed on the same day. Sensitivity was 5 pg/mL and 0.25 ng/mL respectively. CV was 2.73% and <1% respectively. Luteinizing hormone (LH) serum levels were not extracted before the assay and were measured by enzyme-linked immunosorbent assay (ELISA) kit as per manufacturer's instructions (Elabscience Biotechnology Co. Ltd, Wuhan, China). Sensitivity was 0.938 mIU/mL and CV <10%.