How is the optimal threshold determined in ROC analysis of individual biomarkers?

MtaboAnalyst currently offers two different approaches to determine the optimal threshold based on ROC curves.

  • In Youden approach, the optimal cut-off is the threshold that maximizes the distance to the diagonal line. The optimality criterion is max(sensitivity + specificity) .

  • The other approach is to find the point closest to the top-left corner of the ROC plot with perfect sensitivity or specificity. The optimality criterion is min((1-sensitivity)^2 + (1-specificity)^2)

Please note,

  1. There might be more than one threshold identified using the above criteria;
  2. The optimal cutoff is only calculated for the ROC curve using the actual data points, not the smoothed one.