Fig. 5: The analysis of the top-ten genes identified by Cox PH and stable Cox on the breast cancer transcriptome dataset.

a, Kaplan–Meier plot of two subgroups (above and below the median of β^TX) separated by the top-ten genes identified by the Cox PH and stable Cox for three test cohorts. The P value is calculated using a two-sided log-rank test. HR is calculated by a univariate Cox regression of the subgroup assignment variable on outcomes. b, Univariate Cox regression analyses among clinical subgroups of patients. The subgroups are defined according to the clinical variables of patients. Box and error bars represent the HR value and its 95% CI, respectively. The subgroups for HR calculation are separated based on optimal cut-off value defined as the minimal proportion of observations per subgroup of at least 30% to further separate the clinical subgroup into more refined subgroups. The detailed sample size of each subgroup can be found in Supplementary Table A2. c, The favourable/unfavourable consistency of individual genes across training and test cohorts.