Utility of comprehensive genomic profiling combined with machine learning for prognostic stratification in stage II/III colorectal cancer after adjuvant chemotherapy

Background and purpose: Accurate recurrence risk evaluation in patients with stage II and III colorectal cancer (CRC) remains difficult. Traditional histopathological methods frequently fall short in predicting outcomes after adjuvant chemotherapy. This study aims to evaluate the use of comprehensive genomic profiling combined with machine learning for prognostic risk stratification in patients with CRC. Methods: A machine learning model was developed using a training cohort of 52 patients with stage II/III CRC who underwent curative surgery at Fujita Health University Hospital. Genomic DNA was isolated from formalin-fixed, paraffin-embedded tissue sections and analyzed with a 160 cancer-related gene panel. The random forest algorithm was used to determine key genes affecting recurrence-free survival. The model was validated by developing a risk score with internal and external cohorts, including 44 patients from Keio University Hospital. Results: Six key genes (KRAS, KIT, SMAD4, ARID2, NF1, and FBXW7) were determined as significant prognostic risk predictors. A risk score system integrating these genes with clinicopathological factors effectively stratified patients in both internal (p < 0.001) and external cohorts (p = 0.017). Conclusions: This study reveals that machine learning, combined with comprehensive genomic profiling, significantly improves prognostic risk stratification in patients with stage II/III CRC after adjuvant chemotherapy. This approach provides a promising tool for individualized treatment strategies, warranting further validation with larger cohorts.