2.3. Construction and verification of the CuAL signature

ESCA samples derived from the TCGA functioned as a training set to develop a prognostic signature. Conversely, samples from the {"type":"entrez-geo","attrs":{"text":"GSE53624","term_id":"53624"}}GSE53624 dataset were employed as a validation set. Using univariate Cox regression analysis, four significant CuALs were identified. A signature for predicting prognosis was then developed using LASSO and multivariate Cox regression analysis, resulting in the determination of risk scores for ESCA patients. These scores were then used to classify patients as high- or low-risk groups. Survival differences between these groups were analyzed via the “survminer” package. Time-dependent ROC curves, generated using the “timeROC” R package, assessed the signature's efficacy in forecasting prognosis through AUC values [31]. Multivariate Cox regression further confirmed the independence of the signature. R package “rms” was used to create the nomogram, and the calibration curves validated its clinical applicability.

Patients were stratified using the “ConsensusClusterPlus” R package based on the expression of total genes, cuproptosis-associated genes, 442 CuALs, and 3 CuALs [32]. Principal component analysis (PCA) was conducted to visualize patient group distributions, and the “pheatmap” R package created a heatmap of risk scores and clinicopathological features [33].