Fig. 7: Survival prediction and risk scores in LUSC.

Results for Cologne patients for model trained on Berlin cohort. A Bootstrap distribution of c-index comparing UICC8 staging versus niche enhanced risk-scores (n = 109 patients; p < 0.001, two-sided Mann–Whitney U test). Box plots show the median (center line), interquartile range Q1–Q3 (box), and whiskers to the most extreme points within 1.5× IQR; points beyond are shown as fliers. B Kaplan–Meier analysis, patients (n = 109) stratified by the risk scores as assigned via the cell-niche survival model. Patients are split into tertiles for better comparison to UICC8 (see Supplementary Fig. 13 for a corresponding Kaplan–Meier analysis based on the UICC8 model). Separation tested significantly via log-rank test. C Sankey plot showing the redistribution of patients (n = 109) from the UICC8 to the niche-based risk stratification. A large number of the UICC8 stage 1 patients was reclassified as risk score 2 (RS2) indicating potentially undertreated high-risk patients. D Kaplan–Meier analysis for UICC8 stage 1 patients (n = 163; black curve) and the restratification by risk score (RS1 = blue curve, n = 66 patients; RS2 = orange curve, n = 209 patients). Separation tested significantly via log-rank test. E Example tumors (n = 12) with the characteristic niche patterns of risk score 1. Colored dots with niche numbers in legend (left) correspond to dots in tumor samples representing cells colored by their niche membership (right). Bar plots indicate relative abundance of niches in tumors: Hot niches are indicated by red, cold niches by blue bars. Niches are sorted by the abundance within the risk-score group relative to the global average, e.g., TLS^High the niche was observed more often than in the whole cohort whereas Immu^Null was observed less frequently. F Same as E but for risk score 2. G Same as (E) but for risk score 3. Scale bars: (E–G) 500 µm. Source data are provided as a Source data file.