A New Comprehensive Colorectal Cancer Risk Prediction Model Incorporating Family History, Personal Characteristics, and Environmental Factors

Abstract

Purpose: Reducing colorectal cancer incidence and mortality through early detection would improve efficacy if targeted. We developed a colorectal cancer risk prediction model incorporating personal, family, genetic, and environmental risk factors to enhance prevention.

Methods: A familial risk profile (FRP) was calculated to summarize individuals' risk based on detailed cancer family history (FH), family structure, probabilities of mutation in major colorectal cancer susceptibility genes, and a polygenic component. We developed risk models, including individuals' FRP or binary colorectal cancer FH, and colorectal cancer risk factors collected at enrollment using population-based colorectal cancer cases (N = 4,445) and controls (N = 3,967) recruited by the Colon Cancer Family Registry Cohort (CCFRC). Model validation used CCFRC follow-up data for population-based (N = 12,052) and clinic-based (N = 5,584) relatives with no cancer history at recruitment to assess model calibration [expected/observed rate ratio (E/O)] and discrimination [area under the receiver-operating-characteristic curve (AUC)].

Results: The E/O [95% confidence interval (CI)] for FRP models for population-based relatives were 1.04 (0.74-1.45) for men and 0.86 (0.64-1.20) for women, and for clinic-based relatives were 1.15 (0.87-1.58) for men and 1.04 (0.76-1.45) for women. The age-adjusted AUCs (95% CI) for FRP models for population-based relatives were 0.69 (0.60-0.78) for men and 0.70 (0.62-0.77) for women, and for clinic-based relatives were 0.77 (0.69-0.84) for men and 0.68 (0.60-0.76) for women. The incremental values of AUC for FRP over FH models for population-based relatives were 0.08 (0.01-0.15) for men and 0.10 (0.04-0.16) for women, and for clinic-based relatives were 0.11 (0.05-0.17) for men and 0.11 (0.06-0.17) for women.

Conclusions: Both models calibrated well. The FRP-based model provided better risk stratification and risk discrimination than the FH-based model.

Impact: Our findings suggest detailed FH may be useful for targeted risk-based screening and clinical management.

Figure 1:

Cumulative incidence of colorectal cancer (CRC) according to estimated 5-year absolute risk for a) population-based relatives, b) clinic-based relatives. Four groups were defined based on cut-points of 3^rd, 6^th and 9^th deciles of estimated 5-year absolute risk. The K-sample test was used to compare the cumulative incidence across groups and to calculate two-sided P values. FRP = familial risk profile

Figure 1:

Cumulative incidence of colorectal cancer (CRC) according to estimated 5-year absolute risk for a) population-based relatives, b) clinic-based relatives. Four groups were defined based on cut-points of 3^rd, 6^th and 9^th deciles of estimated 5-year absolute risk. The K-sample test was used to compare the cumulative incidence across groups and to calculate two-sided P values. FRP = familial risk profile

Figure 2.

Age-adjusted Receiver Operating Characteristic (ROC) curves for men and women. ROC curves and Age-adjusted area under the curve (AUC) were calculated as the weighted average of age-specific estimates, with weights as the proportion of CRC diagnosis in each age group (<50 and >=50 at baseline). We calculated 95% confidence intervals (in parentheses) using bootstrap approach. FRP = familial risk profile; FH = binary family history.