Human leukocyte antigen class I (HLA-I) molecules are essential for neoantigen presentation and T cell recognition, yet the clinical significance of allelic imbalance within HLA-I genes (HLA-AI) in immune checkpoint inhibitor (ICI) therapy remains undefined. Here, we established haplotype-specific, coverage-based (cHLA-AI) and plasma-derived (bHLA-AI) models compatible with routine clinical sequencing, based on 292 fully heterozygous patients with paired tumor and blood samples from the phase III CHOICE-01 trial. Tumor mutational burden (TMB)-low tumors with HLA-AI derived no benefit from first-line immunochemotherapy, whereas all other patients achieved significant survival gains (mOS 16.53 vs. 29.57 months, HR = 2.29, 95% CI 1.59-3.30, p < 0.001,interaction P = 0.019; mPFS 5.59 vs. 9.92 months, HR = 2.02, 95% CI 1.43-2.90, p < 0.001, interaction P = 0.016). These findings were validated in the RATIONALE-304 and RATIONALE-307 trials and independent real-world cohorts, and extended to cfDNA (bHLA-AI), where integrated tissue-plasma assessment delineated four therapeutic trajectories. Incorporating cHLA-AI with pathology, PD-L1 and TMB significantly improved 2-year OS prediction (DeLong's P = 0.003). Multi-omic profiling linked cHLA-AI to active DNA damage response signaling, high TMB, elevated intratumor heterogeneity (ITH-high), pronounced chromosomal instability (CIN-high), immune-cold microenvironments, and failure of on-treatment TCR expansion, while longitudinal sampling revealed its late, branching emergence under immune pressure. Pan-cancer profiling (N = 5,989) demonstrated consistent associations with TMB and proliferative activity (Ki-67 index). Collectively, these results establish cHLA-AI as a pivotal biomarker bridging genomic instability, immune evasion, and therapeutic outcome, providing a framework for stratified immunotherapy in non-small cell lung cancer and beyond.