Statistical evaluation
Statistical analyses were performed via S-PLUS Professional and R statistics v.1 (MathSoft, Seattle, WA, USA). The descriptive characteristics of the variables are summarized by mean, median, frequency and percentage. Kaplan-Meier survival analyses and log rank test were done to estimate BCR-free rates of all classifications and compare them. The accuracy of the four prognostic models for prediction of BCR was quantified according to Harrell’s concordance index11,12. Harrell’s concordance index provides the probability that, in a randomly selected pair of patients in which one patient experiences the event (BCR) before the other, the patient who experiences the event had the worse predicted outcome according to the predictive model. In accuracy analyses, a value of 100% indicates perfect prediction versus 50% being equivalent to a toss of a coin. In our study, the accuracy value of four prognostic models were calculated according to Harrell’s concordance index. 0.05 was considered as the significance threshold for p-value.