A novel nutritional immune risk score model for long-term prognosis in colorectal cancer using clustering and principal component analysis

Background: Survival outcomes among patients with colorectal cancer (CRC) often differ despite identical disease stages, partly due to variations in nutritional and immune status. Malnutrition can impair immune defense, exacerbate inflammatory responses, and influence tumor progression, ultimately contributing to a poorer prognosis. However, current clinical prognostic systems rarely integrate nutritional immune indicators with tumor biomarkers, limiting the application of nutritional intervention in CRC management. This study aimed to develop a nutritional immune risk score (NIRS) model to improve long-term prognostic evaluation in patients with CRC. Methods: In this retrospective study, 892 inpatients with primary CRC who underwent curative resection in 2017 were included and followed until 2023. Unsupervised learning was applied to nutritional and tumor biomarkers for feature extraction and patient stratification. K-means clustering was used to identify subgroups, and principal component analysis was used to derive composite features, which were then used to construct the NIRS model for long-term prognostic assessment. Results: < 0.001). Locally estimated scatterplot smoothing (LOESS) analysis further demonstrated that among patients with PNI ≥ 50, survivors had smaller tumors than non-survivors, whereas the opposite pattern was observed among patients with PNI < 50. Conclusion: We developed a novel NIRS for long-term prognostic assessment in patients with CRC. The NIRS model demonstrated robust risk stratification and potential clinical utility. PNI may serve as a complementary factor to refine risk classification, and its interaction with maximum tumor diameter may improve the sensitivity and precision of prognostic assessment across different nutritional immune states.