Abstract Study question What are the clinical parameters that can predict a premature LH rise or LH surge in patients undergoing ovarian stimulation for IVF? Summary answer The parameters predicting an early LH rise are: Age, BMI, day 3 LH levels, LH on day of antagonist administration, and number of stimulation days. What is known already Normal reproductive function and fertility require precise regulation of FSH and LH secretion. In IVF stimulation cycles some patients will have an early LH rise even while being treated with a GnRH antagonist. Previous studies have shown a possible link of LH rise with ovarian response to gonadotropins. Nevertheless, little is known of which clinical parameters can predict this LH rise; an occurrence that may necessitate cycle cancellation. Early prediction of premature LH rise will enable us to foresee this unwanted hypophyseal event, and in turn may lead to early antagonist administration and/or early triggering for oocyte retrieval. Study design, size, duration A retrospective study enrolling 382 patients who underwent IVF treatment at Rambam Medical Center (2015 - 2021): age 34.1±6.7 years; BMI 25.0±5.1 kg/m²; AFC 11.2±7.1; day 3 FSH 7.2±2.9U/L; day 3 LH 6.4±4.3U/L; eggs retrieved 8.0±6.1; fertilizations 4.7±4.2. Patients were stimulated with rFSH (Gonal F)/rFSH+LH (Pergoveris)/ highly purified urinary FSH+hCG(Menopur); GnRH antagonist (Cetrotide/ Orgalutran) was administered using a flexible protocol; triggering of ovulation was performed using GnRHagonist and/or hCG, based on estradiol levels. Participants/materials, setting, methods Clinical parameters included from chart data were: age; BMI; estradiol, progesterone, LH and FSH levels at day3/ day of antagonist administration/ and last day before triggering; total stimulation days and total FSH dose. In order to establish a model to predict premature LH rise we performed multiple linear regression using ANOVA with a confidence interval of 0.95 considered significant. A machine learning based model was used to test the accuracy of our model using Python. Main results and the role of chance An LH rise was calculated as the difference between pre-trigger- and basal LH levels. For the total group of patients this LH rise was significantly predicted by patient age; BMI; LH levels at day 3 and LH at the day of antagonist administration; and total stimulation days (R = 0.742, R²=0.55, p < 0.001). Importantly, when analyzing the data of specific age groups, the prediction of the model was strongest in the young patients (age 25-30: R = 0.94, R²=0.88, p < 0.001) and weakest in the older patients (age > 41: R = 0.48,R²=0.23, p = 0.003). Accordingly, using a machine learning model in which a Python based algorithm is trained by analyzing 80% of the data, we were able to successfully predict the remaining 20%: mean squared error of 4.32 and 82.3 in the young- and old patient groups respectively. Interestingly, 21 patients demonstrated a “full LH surge” as defined by a more than two fold increase in LH levels. These patients were also predicted by a similar model (LH levels at day 3 and at the day of antagonist administration, BMI and total FSH dose; R = 0.89, R²=0.80, p = 0.015). Limitations, reasons for caution This is a retrospective study and its conclusions need to be validated by a prospective randomized trial. Wider implications of the findings Using machine-learning based analysis of patient data from IVF cycles, we were able to predict patients at risk for premature LH-rise and/or LH-surge. Interestingly, the prediction of this model was strongest in young patients. Utilizing this model will help prevent IVF cycle cancellation and better timing of ovulation triggering. Trial registration number NA
Read full abstract