Fig. 5.

Time-dependent Cox regression analysis for associations between tumour mRNA levels and patients’ clinical outcomes. a Time-dependent Cox regression analysis for the associations between primary breast tumour mRNA levels of TGFB1, TGFB2, TGFB3, TGFBR1 and TGFBR2 and patients’ overall survival in the TCGA cohort and its subsets stratified by tumour diameter (T), regional lymph node status (N) and ER status. b Associations between primary breast tumour mRNA levels and patients’ relapse-free survival in the GEO cohorts. A Heaviside function approach were applied to calculate the hazard ratios of death (y-axis, logarithmic scale) for a 1-unit increase in the mRNA expression levels (log2-transformed and standardized) of the five genes for the period for 0 to less than 3 years and the open-ended period from 3 years and beyond, for each group of patients (title on each panel). The proportional hazards assumption was met (P ≥0.1) for each of the separate time intervals <3 years and ≥3 years. The shapes were filled black if the corresponding p-values for likelihood-ratio test were <0.05. Missing data in the panel “ER neg.” were due to exceeding the y-axis limit and were non-significant