Integrating polygenic and machine learning-derived nutrient risk scores for colorectal cancer risk stratification: Evidence from Korean adults.

Background

This study aimed to develop a machine learning (ML)-derived nutrient risk score (NRS) and examine the association of genetic susceptibility and dietary intake with colorectal cancer (CRC) risk in Korean adults.

Methods

This hospital-based case-control study included 1,420 cases and 2,840 controls. Polygenic risk scores (PRS) were used to examine genetic susceptibility. The NRS was developed by combining five algorithms, nested cross-validation (CV), and recursive feature elimination with CV. Model performance was primarily assessed using AUC. Logistic regression was used to derive the NRS and assess the association of PRS, NRS, and their combined effect on CRC risk.

Results

A total of 15 nutrients were retained to construct the NRS. In mutually adjusted analyses, both scores remained independently associated with CRC risk (IV-weighted PRS: OR = 1.50; 95% CI, 1.10-2.06; NRS: OR = 3.67; 95% CI, 2.64-5.15 for the highest versus lowest tertile). NRS showed robust discriminative performance (AUC = 0.852; 95% CI, 0.833-0.872). While no significant PRS-NRS interaction was observed, the model integrating both scores significantly enhanced individual-level risk reclassification (IDI = 0.005, Continuous NRI = 0.120, P<0.005).

Conclusion

Both PRS and NRS showed independent associations with CRC risk, and their integration significantly improved risk reclassification. NRS may serve as a useful composite metabolic marker for CRC risk stratification. Future studies incorporating longitudinal designs, larger sample sizes, and multi-omic data are warranted to further elucidate CRC risk determinants and to inform personalized risk assessment strategies.