HORMONAL AND GENETIC FACTORS INFLUENCING THE DEVELOPMENT OF POLYCYSTIC OVARY SYNDROME

Abstract

Polycystic Ovary Syndrome (PCOS) is a widespread endocrine disorder with multifaceted aetiology (genetic, hormonal and environmental), but the underlying mechanisms behind the varied clinical presentation are unclear. We integrated 147 studies (38,204 PCOS patients and 52,109 controls) in this study to evaluate the performance of nine prediction models for PCOS using a combination of genetic variants (CYP17, INSR, FSHR), hormone levels (testosterone, LH/FSH ratio, AMH) and metabolic factors (HOMA-IR, BMI). A stacked ensemble machine learning method performed better (area under the curve (AUC) 0.947, Brier score 0.082) compared to logistic regression , random forest , support vector machine and artificial neural network . We detected a significant effect modification of genetic risk and insulin resistance (IR) on PCOS, with the homozygous C/C CYP17 variant carriers having a 72.7% posterior probability of developing PCOS. Time-dependent ROC analysis identified an AUC of 0.894 for onset of metabolic syndrome at 36 months and the random forest classifier trained to distinguish PCOS phenotypes had an accuracy of 91.2%, with the highest feature importance being serum testosterone and insulin. Time series analysis revealed a monthly loss of 0.142 days menstrual cycle duration , and ameliorated with metformin. Our findings confirm that integrated models with multiple parameters perform much better than single feature models, and indicate biomarker-based, personalised diagnostic and therapeutic approaches for PCOS.