Impact of sample size on prognostic gene analysis in targeted sequencing of non–small cell lung cancer

Identification of prognostic somatically mutated genes in non-small cell lung cancer (NSCLC) is often inconsistent across studies due to cohort heterogeneity and insufficient sample sizes. While whole-exome sequencing suggests a power-law relationship between sample size and prognostic gene yield, the impact of sample size in targeted sequencing cohorts remains unclear. We analyzed 6,778 NSCLC cases from the MSK‑CHORD cohort, including 5,956 lung adenocarcinoma (LUAD) and 822 lung squamous cell carcinoma (LUSC) patients. Somatic mutation profiles and their prognostic significance were compared between early‑ and late‑stage disease. Random resampling across a range of sample sizes was used to model the relationship between cohort size and the number of prognostic somatically mutated genes identified. Median mutation counts did not differ significantly by stage in LUAD or LUSC, yet 54 LUAD and 11 LUSC genes showed stage‑dependent frequency differences, including canonical drivers such as TP53 , KRAS , and EGFR . The identified prognostic gene sets varied substantially between early‑ and late‑stage disease, with only nine shared genes in LUAD and none in LUSC. For both LUAD and LUSC, the number of prognostic genes initially increased linearly with sample size ( R > 0.985), stabilizing at approximately 700 samples across early-stage, late-stage, and other LUAD subgroups. Conversely, LUSC discovery did not reach a plateau within the available sample size. In targeted sequencing datasets, the discovery of prognostic somatically mutated genes in NSCLC scales linearly with sample size prior to reaching a plateau, independent of factors such as histological subtype, stage, sex, smoking history, and driver mutation status.