Fig. 5

Stratification of the validation cohort critically depends upon tumour sampling. a Values of NEAT logHR (top) and p values (log-rank test, bottom) are shown for subsampled datasets generated by taking a maximum of one (dashed line), two (dotted line) or three (dot-dash line) samples per tumour. The vertical line in each graph indicates NEAT performance using all available samples. Stratification performance improves significantly as the number of samples taken increases. b Variation in per-patient NEAT HR driven by tumour sampling. Each plot corresponds to a patient and shows the distribution of logHR from NEAT across the available tumour samples. Vertical bars indicate logHR range for every possible combination of the specified number of samples. Therefore, logHR calculated using all samples is shown on the right of each plot as a single point. For many patients (14/22, 64%) the logHR distribution encompasses the classification threshold (logHR = 0); hence risk group assignment is critically influenced by the tumour sample(s) analysed