Prediction of Tumor Complete Response Using Genomic and Clinical Data in Solid Cancer Undergoing Definitive Radiation Therapy

In this retrospective, single-center observational study, targeted next-generation sequencing (NGS) was performed on pre-radiation therapy (RT) tumor specimens to identify genomic features associated with RT response and clinical outcomes. This study included 77 eligible patients with esophageal squamous cell carcinoma (ESCC, n = 23), non-human papilloma virus-related head and neck squamous cell carcinoma (HNSCC, n = 24), and rectal adenocarcinoma (RAC, n = 30) treated between 2014 and 2020. Treatment was employed for definitive aim (chemoradiotherapy or pre-operative chemoradiotherapy followed by radical surgery). Treatment response was assessed using clinical and pathological criteria and was dichotomized as complete response (CR) versus non-CR. Targeted NGS was performed using a customized panel composing 161 genes. A total of 1263 nonsynonymous somatic variants were identified across 153 genes. The most commonly altered genes were TP53 (70.1%), APC (41.6%), NOTCH1 (41.6%), KMT2C (39.0%), NOTCH3 (36.4%), and NOTCH2 (35.1%). Overall, 25 (32.5%) and 52 patients (67.5%) were classified as CR and non-CR, respectively. APC (P = 0.008), KMT2C (P = 0.018), and KRAS (P = 0.012) alterations were associated with non-CR, whereas CD274 (P = 0.003), MSH3 (P = 0.011), and NOTCH3 (P = 0.048) alterations were associated with CR. KRAS alterations occurred exclusively in RAC, and most concurrent APC–KMT2C alterations in non-CR tumors were also observed in RAC, suggesting that these associations may partly reflect tumor type-specific genomic backgrounds. An exploratory pooled model incorporating age and six candidate genomic variables achieved a cross-validated area under the curve of 0.867 for CR prediction. In survival analysis, alterations in CCND1, CXCL9, ERCC6, LIF, and RAD51B were independently associated with poor recurrence-free survival. Pre-RT tumor genomic profiling identified candidate features associated with RT response and recurrence-free survival in this heterogeneous cohort. The pooled genomic associations and prediction model should not be interpreted as pan-cancer biomarkers or predictors and require validation in larger, independent, cancer-specific cohort.