Integrating metabolomics, clinical factors, and genetic risk for colorectal cancer risk prediction: Model development and internal-external cross-validation

Figure 1 Associations between plasma metabolites and incident colorectal cancer risk. Bars represent hazard ratios (HRs) per 1-standard deviation increment in each transformed and standardized metabolic marker. Orange bars indicate statistically significant positive associations with colorectal cancer risk (HR > 1), blue bars indicate statistically significant inverse associations with colorectal cancer risk (HR < 1), and gray bars indicate associations that were not statistically significant after false discovery rate correction. Background sectors indicate metabolic classes.

Figure 2 Risk stratification of incident colorectal cancer by metabolomic risk score. A: Observed colorectal cancer (CRC) event rates across metabolomic risk score (MetRS) percentiles in the overall population. Blue points indicate men and orange points indicate women; point size represents age quartiles (Q1-Q4); B: Cumulative CRC risk during follow-up across the low, middle, and high MetRS tertiles; C: Hazard ratios for the association between MetRS and incident CRC in the overall population and in subgroups defined by age, sex, and polygenic risk score tertiles. CI: Confidence interval; CRC: Colorectal cancer; HR: Hazard ratio; MetRS: Metabolomic risk score; PRS: Polygenic risk score; SD: Standard deviation.

Figure 3 Regional and pooled estimates of Harrell’s C-index and calibration slope for the four colorectal cancer prediction models. Region-specific estimates were pooled using random-effects meta-analysis. The calibration slope was calculated using the Cox model linear predictor. QCancer denotes the adapted QCancer model used in this study. CI: Confidence interval; MetRS: Metabolomic risk score; PRS: Polygenic risk score.

Figure 4 Decision-curve analysis of the four colorectal cancer prediction models at 5 years, 10 years, and 15 years. The models were compared with the treat-all and treat-none strategies. QCancer denotes the adapted QCancer model used in this study. A: Decision-curve analysis (DCA) at 5 years; B: DCA at 10 years; C: DCA at 15 years. DCA: Decision-curve analysis; MetRS: Metabolomic risk score; PRS: Polygenic risk score.