Extended Data Fig. 1: Association of increasing thymic health cut-point thresholds with long-term risk of mortality.

a, Risk of death according to increasing thymic-health thresholds in the pooled pan-cancer cohort (Harvard-NSCLC and Harvard-PAN; total, n = 3,476) revealed no clear binarization threshold. This analysis compares the entire population divided into two groups at various percentile thresholds (e.g., above vs. below a given percentile). b, Risk of death in each thymic-health decile versus the lowest decile, as reference, in the pooled pan-cancer cohort (n = 3,476) indicated low, average, and high-performance pan-cancer decile subsets. This analysis provides a more granular analysis by comparing individual deciles (i.e., specific 10% slices) against the lowest decile (bottom 10% thymic health). Panel b demonstrates that no single decile disproportionately drives the association, indicating the absence of an outlier effect and supporting the stability and robustness of the associations. This explains why the binary splits shown in subpanel a consistently produce HRs within a similar range, i.e., the underlying risk reduction is broadly distributed. a,b, Cox proportional hazards regression was used to estimate HRs. In the forest plots, the centre of each box represents the estimated hazard ratio, and the whiskers denote the corresponding 95% CI; arrowheads indicate that the 95% CI extends beyond the visualized limits; shaded box size is for visualization only and does not encode statistical weight. Statistical significance of the binarized thymic-health covariate coefficients at the different cut-points was assessed using two-sided Wald z-tests without adjustments for multiple comparisons.