在泛印度肺癌队列中将基因组分类的分子簇映射到治疗结局：一种更明智的临床试验随机化策略

Background: Lung cancers exhibit extensive genomic heterogeneity, yet clinical trial randomization commonly relies on broad histologic or single-driver classifications, potentially obscuring mechanistically distinct subgroups with differential therapeutic responses. There is a need to define reproducible, genomically informed clusters that integrate co-alteration patterns and pathway-level signals to better align patients with targeted and combination strategies. Methods: A real-world clinico-genomic lung cancer dataset was curated from 5000 Indian patients undergoing comprehensive next-generation sequencing, including single-nucleotide variants, indels, copy-number alterations, and fusions across key oncogenic pathways. Tumors were first grouped into genomically defined cohorts based on dominant driver events (for example EGFR, ALK, KRAS, MET, RET, BRAF, ERBB2, and oncogene-negative), then further stratified by co-mutation signatures, tumor suppressor loss, DNA damage and cell-cycle alterations, and pathway activation patterns. Unsupervised clustering and network analysis were applied to identify higher-order molecular mechanism clusters, which were then linked to prior treatment exposures and observed clinical outcomes where available. Results: Distinct clusters emerged within and across canonical driver-defined cohorts, characterized by recurrent constellations of alterations involving DNA repair, cell-cycle control, PI3K-AKT-mTOR, RAS-MAPK, and apoptotic pathways. Several clusters showed enrichment for therapeutic resistance-associated features, including concurrent TP53, RB1, or CDKN2A/B loss in EGFR- and ALK-driven tumors, and convergent bypass signalling alterations in KRAS- and MET-driven disease. Integration of longitudinal treatment histories suggested that these mechanistic clusters correlated with differences in progression patterns and durability of response to targeted therapies and taxane- or platinum-based backbones, supporting their potential utility as strata for trial enrolment and adaptive randomization. Conclusions: Systematic analysis of genomically classified lung cancer cohorts can reveal mechanistically coherent molecular clusters that transcend single-driver labels and may better predict therapeutic trajectories. Incorporating such clusters into trial design may refine eligibility, reduce biological heterogeneity within arms, and enable more informative comparisons of targeted and combination regimens. These insights derived from an Indian population will be very valuable for multicentric trial designs. These findings support further validation of cluster-based randomization frameworks and their integration into AI-enabled clinico-molecular platforms for precision oncology trials.