Using the (0, 1) criterion, the point on the curve that minimises the trade-off between sensitivity and specificity, thus having the shortest distance to the coordinate (0, 1), is chosen and patients are classified accordingly as positive or negative around this corresponding protein expression value. It should Tipifarnib 192185-72-1 be noted that although prognosis is generally considered a time-to-event outcome, time-dependent ROC methods have only recently been established and to date can not be implemented with ease for the cutoff point determination over time (Heagerty et al, 2000). For this reason, standard ROC curve analysis was used in this study and is expected to yield the best cut off value for markers to discriminate between patients who have died from disease vs those alive or censored after 5 years.

Statistical analysis Univariate survival analysis using Cox proportional hazards regression was performed for each protein marker. The assumption of proportional hazards was verified before each analysis. Hazard ratios (HRs), 95% CI and P-values were used to determine the effect of each protein marker on survival time. In the case of protein markers, the baseline hazard of 1.0 was systematically attributed to negative protein expression. HR >1.0 indicate an adverse prognosis with positive expression, whereas HR <1.0 indicate improved prognosis with marker positivity. To determine the reliability of the prognostic effects of markers significant in univariate analysis, bootstrapped replications of the data were analysed.

This approach allows one to sample the data with replacement, for example, 1000 times resulting in 1000 different ��resamples’ of the original dataset. For each of these resamples, multiple Cox regression analysis was performed using a forward selection procedure. The number of times a particular marker was selected as an independent factor, after adjustment for the remaining variables, was determined. The most reliable independent markers were combined into multimarker phenotypes with different combinations of their negative or positive expressions. Kaplan�CMeier survivals curves were analysed using the log-rank test. ��2-Tests were performed to determine the association of marker Dacomitinib expression on the absence or presence of local recurrence. P-values were two-sided and considered statistically significant if <0.05. Analyses were performed using SAS (9.1, The SAS Institute, NC, USA). Results Survival analysis Univariate analysis The expression of four markers was associated with survival time including negative expression of Ki67 (P=0.033; HR=0.72 (0.53�C0.97)), positivity for RHAMM (P<0.001; HR=2.19 (1.65�C2.91)), absence of RKIP (P=0.015; HR=0.69 (0.51�C0.93)) and loss of CD8+ TILs (P<0.001; HR=0.55 (0.41�C0.74); Table 2).