Int J Surg. 2025 Dec 23.
BACKGROUND: Immunotherapy has significantly improved survival outcomes for various cancer types, despite suboptimal efficacy in over 60% of patients. However, identifying reliable predictive biomarkers for patient response is still a focus of ongoing research, as current indicators often lack consistency. This network meta-analysis (NMA) systematically compared the predictive performance of 13 biomarkers across multiple cancer types to identify optimal predictors of immunotherapy efficacy.
METHODS: We systematically searched PubMed, OVID, Embase, Cochrane Trials, Web of Science, and trial registries (ClinicalTrials.gov, WHO ICTRP) from inception to September 1st, 2025, and conducted a comprehensive NMA evaluating sensitivity, specificity, diagnostic odds ratio (DOR), superiority, and area under the curve (AUC) for 13 biomarkers. These included circulating tumor DNA (ctDNA), programmed cell death ligand 1 (PD-L1; at varying thresholds), tumor mutational burden (TMB), CD8 + tumor-infiltrating lymphocytes (CD8 + TILs), microsatellite instability (MSI), and inflammatory markers such as neutrophil-lymphocyte ratio (NLR), platelet-lymphocyte ratio (PLR), lactate dehydrogenase (LDH), Lung Immune Prognostic Index (LIPI), and immune-related adverse effects (irAEs). Subgroup analyses were performed for non-small cell lung cancer (NSCLC), melanoma, gastrointestinal (GI) cancer, urothelial carcinoma, and head and neck squamous cell carcinoma (HNSCC). Heterogeneity and publication bias were assessed using I2 statistics and Deeks' funnel plots.
RESULTS: This analysis included 54,634 patients from 194 clinical studies worldwide, evaluating various predictive biomarkers. PD-L1 expression was the most frequently assessed (n = 212), with different cutoffs (≥1% [n = 110], ≥ 5% [n = 45], ≥ 10% [n = 32], ≥ 50% [n = 25]), followed by TMB (n = 68). Other markers included NLR (n = 20), irAEs (n = 17), MSI (n = 14), CD8 + TILs (n = 14), ctDNA (n = 14), PLR (n = 12), LIPI (n = 11), and LDH (n = 10). Among these, ctDNA demonstrated the highest sensitivity (0.82, 95% CI: 0.72-0.89) and overall discriminative power (DOR = 9.75, 95% CI: 5.20-16.73; AUC = 0.769). PD-L1 demonstrated threshold-dependent performance: ≥ 50% cutoff demonstrated the highest specificity among PD-L1 thresholds (0.78, 95% CI: 0.73-0.81) and diagnostic accuracy (DOR = 2.60, 95% CI: 1.86-3.52; AUC = 0.661) but the lowest sensitivity (0.42, 95% CI: 0.36-0.49). ≥ 10% threshold showed sensitivity of 0.44 (95% CI: 0.38-0.51) with specificity of 0.74 (95% CI: 0.70-0.78; AUC = 0.656). ≥ 5% cutoff demonstrated sensitivity of 0.54 (95% CI: 0.48-0.60) and specificity of 0.66 (95% CI: 0.62-0.70; AUC = 0.631). Conversely, ≥ 1% cutoff achieved the highest sensitivity among PD-L1 thresholds (0.68, 95% CI: 0.65-0.71) at the cost of the lowest specificity (0.48, 95% CI: 0.45-0.51; AUC = 0.601). TMB balanced sensitivity (0.56, 95% CI: 0.50-0.60) and specificity (0.69, 95% CI: 0.65-0.73; AUC = 0.637). MSI status had the highest specificity (0.89, 95% CI: 0.85-0.93; AUC = 0.727) but low sensitivity (0.36, 95% CI: 0.27-0.46), supporting its role in confirmatory testing. CD8 + TILs showed good sensitivity (0.69, 95% CI: 0.58-0.79) but lower specificity (0.59, 95% CI: 0.49-0.67; AUC = 0.632). irAEs displayed relatively higher sensitivity (0.69, 95% CI: 0.60-0.77) with moderate specificity (0.59, 95% CI: 0.50-0.67; AUC = 0.674). Among inflammatory markers, PLR (AUC = 0.623) showed slightly better predictive power than NLR (AUC = 0.613), while LIPI and LDH exhibited the least overall effectiveness (AUC = 0.585 and 0.544, respectively).
CONCLUSION: Biomarker performance varies by cancer type and clinical context, underscoring the potential for individualized immunotherapy strategies. ctDNA, PD-L1 (high thresholds, as ≥50%), and TMB emerge as leading predictors, while combinations may optimize sensitivity and specificity. Future research should focus on overcoming heterogeneity and standardization challenges to further refine and individualize immunotherapy approaches and target patients who may benefit from immunotherapy.
Keywords: PD-L1; ctDNA; immunotherapy; network meta-analysis; precision oncology; predictive biomarkers; tumor mutational burden