Fig. 7: A signature of two serum markers is predictive of lethal prostate cancer in AA patients.

Cross-validated, regularized Cox regression models with different elastic net mixture parameters from ridge (alpha = 0) to lasso (alpha = 1) were implemented to identify a predictive proteomic signature. One-sided P values were obtained empirically by comparing feature frequencies and coefficients against those obtained from random permutations of the response⁷⁴. a Heatmaps of feature frequencies across alpha. Features were ranked by P value for alpha = 1. b Heatmaps of feature coefficients across alpha. Features were ranked by P value for alpha = 1. c Kaplan–Meier plot comparing prostate cancer-specific mortality of AA cases with high levels (>median) of both TNFRSF9/CD137/4-1BB and pleiotrophin (PTN) vs. low levels of either or both proteins. Two-sided log-rank test was used to determine if there were statistically significant survival differences. Adjusted hazard ratio (HR) compares the risk of prostate cancer mortality for those with high levels of both TNFRSF9 and PTN vs. the remaining AA cases. HR estimates were adjusted for potential confounding factors: age, BMI, education, income, smoking history, diabetes status, aspirin use, treatment, and NCCN risk score. In (a and b), P value significance was coded as <0.001 (***), <0.01 (**), <0.05 (*),and <0.1 (.). The exact P values are found in the Source Data file. Source data are provided as a Source Data file. HR  hazard ratio.