Machine learning-driven construction of an immunogenic cell death prognostic model for neuroblastoma

Neuroblastoma (NBL) is the most common extracranial solid tumor in children. Its immunosuppressive microenvironment renders effective treatment with existing immunotherapies challenging. Although ICD plays a key role in reshaping tumor immunity and therapeutic response, the prognostic value of its related genes in neuroblastoma remains undefined. We aimed to establish a prognostic model based on ICD-related genes through machine learning algorithms, to improve outcome prediction for patients with NBL. Gene expression and clinical data of NBL patients were obtained from the Gene Expression Omnibus (GEO) database (GSE62564, N = 498) and the TARGET database (TARGET-NBL, N = 150). Univariate Cox regression and LASSO-penalized regression were used to screen for the ICD-related genes most strongly associated with prognosis, followed by the construction of prognostic models using seven machine learning algorithms. Immune infiltration was assessed using CIBERSORT and ssGSEA algorithms. Drug sensitivity analysis was performed with oncoPredict, leveraging the GDSC2 and CTRP2 databases. Functional enrichment analyses (GO, KEGG, and GSEA) and weighted gene co-expression network analysis (WGCNA) were conducted to elucidate transcriptomic differences between high- and low-risk groups. Univariate Cox and LASSO regression analyses identified 11 core ICD-related genes significantly associated with event-free survival (EFS) (CASP8, PIK3CA, EIF2AK3, CD8A, IL17RA, BAX, NLRP3, IL10, TNF, CALR, HMGB1). Among seven machine learning survival models evaluated, the elastic-net Cox model achieved the highest discriminative performance in the validation cohort (C-index = 0.635) and was selected as the optimal model. High-risk patients exhibited significantly worse survival outcomes ( p < 0.05), and the model yielded area under the ROC curve (AUC) values of 0.685, 0.72, and 0.732 for predicting 1-, 3-, and 5-year EFS, respectively. Immune infiltration analysis revealed that the high-risk group displayed an immunosuppressive “cold tumor” phenotype, characterized by higher infiltration of regulatory T cells and neutrophils, lower natural killer (NK) cell levels, and downregulation of multiple immune-related pathways. Drug sensitivity analysis indicated that the high-risk group had significantly higher predicted IC50 values for most targeted agents (e.g., the MEK inhibitors trametinib and selumetinib, and the TKI neratinib), suggesting a potential trend toward differential drug sensitivity. WGCNA identified the blue module as the most strongly correlated with high risk; genes in this module were predominantly enriched in cell cycle processes such as chromosome segregation and nuclear division. Leveraging 11 ICD-related genes, this study developed a machine learning-based prognostic model with moderate predictive ability for outcomes in neuroblastoma patients and revealed significant differences between risk groups in the immune microenvironment, signaling pathway activity, and drug sensitivity. This model provides a complementary molecular perspective for precise risk stratification in NBL and offers a theoretical rationale for combining immunotherapy with targeted agents to improve outcomes in high-risk patients.