Fig. 6

Stratifying patients by tumour mRNA levels for prognosis. Patients were divided into two groups (low and high) according to their tumour levels of TGFB1, TGFB2, TGFB3, TGFBR1 or TGFBR2 mRNA (a) in the TCGA cohort and its subsets and (Bb) in the GEO cohorts. To avoid arbitrary grouping, the cut point for the percentage of patients in the group “low” was set to 10–90 % of the population with an increment of 10 % (y-axis). The hazard ratios of event (death for the TCGA cohorts and relapse for the GEO cohorts) for the patients in group “high” compared with the patients in group “low” for different periods of follow-up time (from 1 to 10 years, x-axis) were calculated using the Cox proportional hazards regression model. The corresponding Cox regression coefficients (β, β = ln(HR)) with corresponding p-values for Log-rank test <0.05 were shown as heatmap (green: low, red: high). The comparison of Kaplan-Meier estimates of survival for the arrowed data points in panels (a) and (b), which had the lowest Wald test p-values for HR in the corresponding block, were shown in panels (c), (d), (e) and (f) accordingly. The Kaplan-Meier survival curves were shown up to 10 years of follow-up. The statistical results shown in each plot were calculated based on different years of follow-up time as indicated by the grey dotted line. The numbers of patients at risk were listed under each time interval