Fig. 1.

Workflow of our risk prediction model construction with supervised principal component analysis (PCA) logistic regression method. We calculated z-value in the logistic regression method for each microRNA (miRNA). The cutoff value of the z-value, T and the number of miRNAs, n, was pre-selected (1). The PCA was performed using the pre-selected miRNAs (2). The risk prediction models were constructed based on the combination of the miRNAs and m PC scores (3). This optimal parameter set (T, m) was determined in the discovery cohort using 10-fold cross-validation (4)