Fig. 3

Clinical-prediction models were established based on IRF family expression levels. A, B Clinical-correlation analysis showed that IRF1 and IRF6 expression is significantly correlated with different clinical stages. C A random-forest model was constructed based on expression levels of IRF family. The weighted values of each member were shown. D, E Survival analysis showed that patients with low IRF score had a good prognosis according to data from both The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) datasets. F The predictive power of multivariate Cox analysis of IRF scores combined with clinicopathological features for OS. G The predictive power of multivariate Cox analysis of IRF scores combined with clinicopathological features for predicting disease-free survival (DFS). H A histogram for predicting OS based on IRF scores and clinicopathological features. I Calibration curve for the OS nomogram; J A histogram for predicting DFS based on IRF scores and clinicopathological features. K Calibration curve for the DFS nomogram. *p < 0.05; **p < 0.01; ***p < 0.001