Article Figures & Data
Tables
- Table 1.
Summary of the previously developed CRC risk prediction models
Factors included Model Study design and sample Methods Family history Environmental factors High-risk genetic mutations Residual risk factorsa Applicability Strengths Limitations Colditz and colleagues (63) Harvard Cancer Risk Index Estimated parameters from published data and expert opinion Scoring for each risk factor based on strengths of associations from the previous logistic regression analyses FDR with colon cancer (yes/no) BMI, screening (FOBT and sigmoidoscopy), aspirin, inflammatory bowel disease, folate, vegetables, alcohol, height, physical activity, estrogen replacement, OC, red meat, fruits, fiber, saturated fat, cigarette smoking None None Predicts 10-y risk of colon cancer Ease of use May not be applicable to rectal cancer risk prediction. Does not consider family history in relatives beyond first degree and is not applicable for people with a high-risk genetic mutation. Imperiale and colleagues (84) A cross-sectional study of 1,994 asymptomatic individuals aged ≥50 years identified between 1995 and 2001 Scoring for each risk factor (method for score undefined) None Age, sex, most advanced distal nonmalignant neoplasm (no polyps; hyperplasia; tubular adenoma <1 cm; advanced lesion—tubular adenoma >1 cm, any polyp with villous histology or severe dysplasia or cancer) None None Determines the need for screening colonoscopy based on the findings of a flexible sigmoidoscopy. Predicts risk of proximal colon cancer. Improves efficiency of CRC screening Limited to individuals having a sigmoidoscopy. May not be applicable to people aged younger than 50 y. Is not applicable for people with strong family history or with a high-risk genetic mutation. Driver and colleagues (80) A prospective cohort of 21,581 men aged 40 to 84 years (Physician's Health Study); followed-up from 1982 to 2004; 485 incident CRC cases. Logistic regression None Age, smoking, alcohol, BMI [diabetes, physical activity, vegetables, cold cereal multivitamins, vitamin C, and vitamin E were considered but not included in the model] None None Provides CRC risk score. Predicts 20-y risk of CRC for men Ease of use. Based on large sample Limited to males. Is not applicable to people with a strong family history or with a high-risk genetic mutation. Freedman and colleagues (64) 2,263 cases and 2,833 controls of non-Hispanic white men and women aged ≥50 years identified between 1991–1994 (colon) and 1997–2001 (rectal) Logistic regression FDR with CRC (yes/no), number of FDR with CRC (0, 1, ≥2) Age, sex, sigmoidoscopy and colonoscopy, current leisure time activity, aspirin and NSAIDs, cigarette smoking, vegetables, BMI, and hormone replacement None None Predicts 5-, 10-, and 20-y, and lifetime risk of developing CRC for men and women older than 50 y User friendly web version available (108) Based on large sample May not be applicable to people younger than 50 y. Does not consider family history in relatives beyond first degree. Is not applicable to people with a strong family history or with a high-risk genetic mutation Wei and colleagues (65) A prospective cohort of 83,767 women aged 30 to 54 years (Nurses' Health Study); follow-up from 1976 to 2004; 701 incident colon cancer cases. Nonlinear Poisson regression FDR with CRC (yes/no) Age, sigmoidoscopy and colonoscopy, physical activity, aspirin, cigarette smoking, processed meat or red meat, folate, height, BMI, and hormone replacement None None Predicts cumulative risk of colon cancer for women aged 30 to 70 y Ease of use. Based on large sample May not be applicable to men or to any women aged older than 70 y. Does not consider family history in relatives beyond first degree. Is not applicable to people with a strong family history or with a high-risk genetic mutation. May not be applicable for rectal cancer risk Ma and colleagues (83) A prospective cohort of 28,115 men aged 40 to 69 years (Japan Public Health Center–based study–Cohort II); followed-up from 1993 to 2005 (mean, 11.0 y); 543 incident CRC cases. Cox proportional hazards model None [FDR with CRC (yes/no) was considered, but not included in the model] Age, BMI, physical activity, smoking, alcohol [diabetes was considered but not included in the model] None None Provides CRC risk score and predicts 10-y risk of CRC for Japanese men. Ease of use May not be applicable to non-Japanese or to Japanese females. Is not applicable to people with strong family history or with a high-risk genetic mutation Chen and colleagues (70) MMRpro Estimated parameters from published data Bayesian/segregation analysis FDR and SDR: specific relationship to the proband, history of CRC/EC (yes, no), age at diagnosis, MSI status, and MLH1, MSH2, MSH6 mutation status Age, race/ethnicity, MLH1, MSH2, MSH6 None Predicts probability of carrying MLH1, MSH2, and MSH6 mutations Predicts 5-y and lifetime risk of CRC and EC Included specific family history to second-degree. Uses stand-alone software package that is freely available Does not consider family history in relatives beyond second degree. May not be applicable to PMS2 mutation carriers. Does not estimate risk of second primary (metachronous) CRC. Is not applicable to MUTYH mutation carriers. Cleveland Clinic Tool (109) Unreported Unknown FDR and SDR: specific relationship to the proband, history of CRC and polyps (yes, no), age at diagnosis of CRC (<50, 50–60, ≥60), age at diagnosis of polyps (<60, ≥60) Age (<50, ≥50), sex, ethnicity, weight, height, CRC screening (colonoscopy, sigmoidoscopy, FOBT), fruit and vegetables consumption, smoking, exercise, person history of CRC and polyps None None Provides CRC risk score (average/medium/high) User friendly web version available (109) Not possible to assess as methods used for development of this tool have not been published. Does not predict cumulative risk over a specified period Abbreviations: BMI, body mass index; EC, endometrial cancer; FDR, first-degree relative; FOBT, fecal occult blood test; MSI, microsatellite instability of tumor; NSAID, nonsteroidal anti-inflammatory drugs; OC, oral contraceptive; SDR, second-degree relative.
-
↵aFamilial aggregation that is not explained by known risk factors (including genetic and environmental risk factors shared by family members).
-
- Table 2.
Summary of studies that evaluated CRC risk prediction models
Study Model evaluated Sample Features evaluated Key findings Kim and colleagues (85) Harvard Caner Risk Index (63) A prospective cohort of 38,953 men aged 40 to 70 years (Health Professionals Follow-up Study); follow-up from 1986 to 1996; 230 incident CRC cases. A prospective cohort of 52,668 women aged 40 to 70 years (Nurses' Health Study); follow-up from 1984 to 1994; 244 incident CRC cases. Calibration, discrimination, and utility Overestimated the number of CRC for men within “much below average risk” and “very much below average risk” risk categories. Well calibrated for women. Modest discrimination: c = 0.71 (95% CI, 0.68–0.74) for men and 0.67 (95%CI, 0.64–0.70) for women. Emmons and colleagues (86) Harvard Caner Risk Index (63) In-depth cognitive interviews with 9 individuals from the general population; and 9 focus groups (6 females and 3 males) aged ≥40 years Meaning of risk, perceptions about cancer, and interpretation of the results 66% extremely/very satisfied, 32% somewhat satisfied with the model format. 86% extremely/very satisfied, 13% somewhat satisfied with the information provided in the model. 3% not at all satisfied with the model. Some dissatisfied because exposures that they believed to be important were not included (e.g., poverty, toxic waste, air pollution) Difficult for participants in completing the HCRI in its paper-and-pencil form. Emmons and colleagues (87) Harvard Colorectal Cancer Risk Assessment and Communication Tool for Research (HCCRACT-R; ref. 63) A randomized control trial on 159 men and 194 women aged 40 to 70 years without previous personal history of cancer. Intervention groups: those receiving (i) presentation of both absolute and relative risk (ii) presentation of absolute risk only Control group: those without receiving any personal risk information Accuracy of risk perception, and, level of worry and satisfaction Significant changes in risk perception accuracy for both relative risk (P = 0.01) and absolute risk (P = 0.001) across intervention groups. Of those with inaccurate absolute risk perception at baseline, 54% of the participants in the group who received presentation of both absolute and relative risk, and 64% of those in the group who received presentation of absolute risk only, had correct absolute risk perception at post-test, compared with only 12% of the control group. 13% less worried, 17% more worried about getting CRC after completing the HCCRACT-R. Imperiale and colleagues (84) Imperiale (84) A cross-sectional study on 1,031 asymptomatic individuals aged ≥50 years undergoing first-time screening colonoscopy between 1999 and 2001 Discrimination Modest discrimination: c = 0.74 (SD = 0.06) Park and colleagues (88) Freedman (64) A prospective cohort of 155,345 men and 108,057 women aged 50 to 71 years (American Association of Retired Persons—diet and health study); follow-up from 1995 to 2003 (mean, 6.9 y); 2,092 male and 965 female incident CRC cases. Calibration and discrimination Well calibrated for men (E/O = 0.99; 95% CI, 0.95–1.04) and for women (E/O = 1.05; 95% CI, 0.98–1.11) overall. Overestimated risk for men with one affected relative (E/O = 1.35; 95% CI, 1.17–1.55); women with one affected relative (E/O = 1.20; 95% CI, 1.00–1.45); men with 2 affected relatives (E/O = 1.48; 95% CI, 1.00–2.19); and men who had a history of screening and polyps (E/O = 1.42; 95% CI, 1.24–1.63). Underestimated risk for men who had a history of screening with no polyps (E/O = 0.67; 95% CI, 0.62–0.72). Modest discriminatory accuracy: c = 0.61 (95% CI, 0.60–0.62) for men and 0.61 (95% CI, 0.59–0.62) for women Ma and colleagues (83) Ma (83) A prospective cohort of 18,256 men aged 40 to 59 years (Japan Public Health Center-based study—Cohort I); follow-up from 1990 to 2005 (mean, 10.1 y); 389 incident CRC cases. Calibration and discrimination Underestimation for colon cancer: O/E = 1.19 (95% CI, 1.03–1.37). Well calibrated for rectal cancer (O/E = 0.94; 95% CI, 0.78–1.12) and for CRC overall (O/E = 1.09; 95% CI, 0.98–1.23) Modest discriminatory accuracy: c = 0.64 (95% CI, 0.61–0.67) for CRC, 0.66 (95% CI, 0.62–0.70) for colon cancer, 0.62 (95% CI, 0.57–0.66) for rectal cancer Abbreviations: c, concordance statistic; E, expected; O, observed.
March 2012
Volume 21, Issue 3
Volume 21, Issue 3
Risk Prediction Models for Colorectal Cancer: A Review
Jump to section
Advertisement