AN ANALYTICAL MODEL TO DEFINE THE PROBABILITY OF AZOOSPERMIA FROM SERUM FOLLICLE STIMULATING HORMONE

Abstract

To develop and internally validate a model to predict the probability of azoospermia from serum follicle stimulating hormone (FSH) in men with infertility. We reviewed a prospectively maintained database from a male infertility clinic between 01/2016 and 03/2020. Age at semen collection, sperm concentration, FSH, LH, total testosterone, bilateral testis volume (estimated by Prader orchidometer) were extracted from the database. Only pared sperm concentration and clinical data collected within +/-90 days of semen analysis were included. Data from men using medications clomiphene, anastrozole, or human chorionic gonadotropin and from men with diagnoses of Klinefelter Syndrome or Y Chromosome Microdeletion (YCM) AZFc were included. Men with history of vasectomy, solitary testis, recent testosterone or steroid use, and YCM AZFa or AZFb microdeletion were excluded. Probability of azoospermia was determined from the quotient of binned FSH data (Number of Azoospermic Samples / Total Number of Samples). A quadratic model predicting probability of azoospermia from FSH was fit to these data. Logistic regression from continuous gonadotropin, testosterone, and testis volume data was used as a comparator. Accuracy and internal validity of each model were assessed via correlation, discrimination (after exclusion of men with obstructive azoospermia) and calibration. All modeling and statistical analysis was computed in R. A total of 946 pared sperm concentration and hormone data sets from 749 men were analyzed. The quadratic FSH model (Probability of Azoospermia = 0.133[FSH]2 - 0.965[FSH] + 10.1) fit with a high R2 (0.95). The model showed an undetectable FSH (<0.2 IU/mL) confers a 10% chance of azoospermia, the probability of azoospermia is least when FSH is 3.6 IU/L. Most interesting was that all men with FSH >30 IU/L were uniformly azoospermic. On internal validation, the Pearson Correlation Coefficient between the quadratic FSH model and the logistic regression model is 0.93, suggesting excellent agreement. The logistic regression and quadratic FSH models both demonstrated good discrimination with AUCs of 0.86 and 0.84, respectively. Finally, each model is well calibrated with near ideal probabilities throughout the entire range of predictions however the quadratic FSH model’s calibration is closer to ideal at higher, more clinically relevant, predicted probabilities of azoospermia. Based on a single serum FSH level, our model provides an infertility specialist the ability to predict with a high degree of confidence the probability of azoospermia, especially at high levels suggesting production defect. The utility of this study presents itself during the initial infertility encounter with a male who is apprehensive about performing a semen analysis. Future work will be necessary to externally validate the model.