Statistical Analyses

The percentages and 95% confidence intervals (CI) are presented for OM/AM and PCOS prevalence. All metabolic hormones except fasting serum glucose were log-10 transformed to obtain a normal distribution before undertaking analyses. For non-normally distributed variables, medians, the standard deviations of the log-10 transformed values, and the range of the untransformed values are shown. Mean differences of normally distributed variables between groups were analyzed using an independent samples two-tailed t-test; this includes variables transformed to normality. A Chi-squared test was used for categorical variables.

Because several socio-demographic factors were associated with OM/AM, we performed a stepwise multivariable logistic regression model with OM/AM as the dependent variable. Predictors that were significant at the level of p<0.10 in bivariate analyses (Table 2) were included. To avoid issues of multicollinearity, abdominal circumference was chosen as the measure of body fat, based on the strength of the bivariate relationship with the OM/AM outcome compared to other indicators of body fat.

Comparison of Normally Menstruating Women and those with Oligomenhorrea or Amenorrhea (OM/AM)

Asterisks denote significant differences between groups based on t-test or Chi-square tests:

General linear models were used to adjust for age effects in multivariable models and to test for differences in cardiometabolic and reproductive biomarker differences by PCOS status. The potential for BMI effects to influence these differences was not modeled due to the likely developmental influence of adiposity on these biomarkers and the cross-sectional design. Pairwise exclusion was used for missing data.

We also explored differences in several socio-demographic, anthropometric, cardiometabolic, hormonal and behavioral characteristics across the four groups of women who met PCOS criteria, but we did not test for significant inter-group differences due to the small sample sizes.