Fig. 2.

Predicted depression sum scores and predicted probabilities of specific symptoms of depression across different levels of testosterone in women (n ​= ​5102). The predictions were obtained from cubic spline regression models with sampling weights, adjusting for age, BMI, alcohol use, smoking, physical activity, and pregnancy. The predicted values reflect a situation where the covariates are set at their mean or mode. The dashed lines indicate the predicted mean of depression sum score and the predicted probability (i.e., prevalence) for each depressive symptom. The shaded ​light red regions represent ​a ​95% ​confidence ​interval ​band. ∗p ​< ​.05 for the association between total testosterone and depression or individual depressive symptom (Wald test).