Development of a novel risk prediction and risk stratification score for polycystic ovary syndrome

Objective: The aim of this study was to develop a simple phenotypic algorithm that can capture the underlying clinical and hormonal abnormalities to help in the diagnosis and risk stratification of polycystic ovary syndrome (PCOS).

Methods: The study consisted of 111 women with PCOS fulfilling the Rotterdam diagnostic criteria and 67 women without PCOS. A Firth's penalized logistic regression model was used for independent variable section. Model optimism, discrimination and calibration were assessed using bootstrapping, area under the curve (AUC) and Hosmer-Lemeshow statistics, respectively. The prognostic index (PI) and risk score for developing PCOS were calculated using independent variables from the regression model.

Results: Firth penalized logistic regression model with backward selection identified four independent predictors of PCOS namely free androgen index [β 0.30 (0.12), P = 0.008], 17-OHP [β = 0.20 (0.01), P = 0.026], anti-mullerian hormone [AMH; β = 0.04 (0.01) P < 0.0001] and waist circumference [β = 0.08 (0.02), P < 0.0001]. The model estimates indicated high internal validity (minimal optimism on 1000-fold bootstrapping), good discrimination ability (bias corrected c-statistic = 0.90) and good calibration (Hosmer-Lemeshow χ² = 3.7865). PCOS women with a high-risk score (q1 + q2 + q3 vs q4) presented with a worse metabolic profile characterized by a higher 2-hour glucose (P = 0.01), insulin (P = 0.0003), triglycerides (P = 0.0005), C-reactive protein (P < 0.0001) and low HDL-cholesterol (P = 0.02) as compared to those with lower risk score for PCOS.

Conclusions: We propose a simple four-variable model, which captures the underlying clinical and hormonal abnormalities in PCOS and can be used for diagnosis and metabolic risk stratification in women with PCOS.