IARC 60th Anniversary - 19-21 May 2026
Session : Biomarker and Cancer Early detection
CRISP2: a comprehensive model to estimate colorectal cancer risk
JENKINS M. 2,24, DOWTY J. 2, MACINNIS R. 1,2, MILNE R. 1,2,3, WIN A. 2, BOUSSIOUTAS A. 4, BUCHANAN D. 5,6,7, CHENG I. 8, CORLEY D. 9,10, GALLINGER S. 11, GILES G. 1,2,3, GRANT R. 12, HUA X. 13, LE MARCHAND L. 14, MACRAE F. 15, PHIPPS A. 16,17, SAKODA L. 16,18, SAMADDER J. 19, SCHMIT S. 20,21, SOUTHEY M. 1,3,22, THOMAS M. 16, ANTONIOU A. 23
1 Cancer Epidemiology Division, Cancer Council Victoria, Melbourne, Australia; 2 Centre for Epidemiology and Biostatistics, Melbourne School of Population and Global Health, The University of Melbourne, Melbourne, Australia; 3 Precision Medicine, School of Clinical Sciences at Monash Health, Monash University, Melbourne, Australia; 4 Department of Medicine, Royal Melbourne Hospital and Sir Peter MacCallum Department of Oncology, University of Melbourne, Melbourne, Australia; 5 Colorectal Oncogenomics Group, Department of Clinical Pathology, The University of Melbourne, Melbourne, Australia; 6 University of Melbourne Centre for Cancer Research, Victorian Comprehensive Cancer Centre, Melbourne, Australia; 7 Genomic Medicine and Family Cancer Clinic, The Royal Melbourne Hospital, Melbourne, Australia; 8 Department of Epidemiology and Biostatistics, University of California, San Francisco, United States; 9 Division of Research, Kaiser Permanente Northern California, Oakland, United States; 10 Department of Gastroenterology, Kaiser Permanente Medical Center, San Francisco, United States; 11 Lunenfeld Tanenbaum Research Institute, Mount Sinai Hospital, University of Toronto, Toronto, Canada; 12 Division of Medical Oncology and Hematology, Princess Margaret Cancer Centre, University Health Network, Toronto, Canada; 13 Department of Cardiology, Peking University Third Hospital, Beijing, China; 14 Cancer Center, University of Hawaii, Honolulu, United States; 15 Department of Colorectal Medicine and Genetics, Royal Melbourne Hospital, Melbourne, Australia; 16 Public Health Sciences Division, Fred Hutchinson Cancer Center, Seattle, United States; 17 Department of Epidemiology, University of Washington, Seattle, United States; 18 Division of Research, Kaiser Permanente Northern California, Oakland, United States; 19 Division of Gastroenterology and Hepatology, Department of Medicine, Mayo Clinic, Phoenix, United States; 20 Genomic Medicine Institute, Cleveland Clinic, Cleveland, United States; 21 Population and Cancer Prevention Program, Case Comprehensive Cancer Center, Cleveland, United States; 22 Genetic Epidemiology Laboratory, Department of Pathology, The University of Melbourne, Melbourne, Australia; 23 Centre for Cancer Genetic Epidemiology, Department of Public Health and Primary Care, University of Cambridge, Cambridge, United Kingdom; 24 Cancer Research Centre, University of Melbourne, Melbourne, Australia
Background: Colorectal cancer risk prediction algorithms can identify individuals at high and low risk of the disease to inform tailored screening and prevention recommendations. There are several colorectal cancer risk prediction models that estimate an individual’s risk according to their age, sex, family history (yes/no), and various personal and lifestyle factors, but they generally do not include colorectal cancer genetic susceptibility loci or some other non-genetic risk factors. Discrimination of these models based on concordance (C)-statistics from external datasets range between 0.56 and 0.71.
Objectives: Our aim was to improve upon our CRISP1 model (Colorectal RISk Prediction version 1) by developing and externally validating a significantly enhanced model, CRISP2, that incorporates age, sex, more detailed family history (including ages at diagnosis and explicit relationships), major colorectal cancer susceptibility genes (DNA mismatch repair genes and MUTYH), a colorectal cancer polygenic risk score (PRS) based on 204 single-nucleotide polymorphisms, and non-genetic risk factors (intended screening uptake (fully compliant or not), smoking, alcohol consumption, and anthropometric measures).
Methods: To incorporate more detailed multigenerational family history and susceptibility genes, we updated our previous segregation analysis of population-based colorectal cancer family data from the Colon Cancer Family Registry Cohort. Participants were recruited from population cancer registries in the United States, Canada, and Australia between 1997 and 2012. We allowed the variance of the familial polygenic component to depend linearly on age. We included in the model a PRS that explains 28% of the colorectal cancer polygenic variance, a residual polygenic component accounting for other genetic/familial effects, and non-genetic risk factors (relative risks and prevalences obtained from meta-analyses and national surveys, respectively). For prospective external validation, we used a random selection of 4,700 participants unaffected at baseline and all 305 incident colorectal cancer cases from the Melbourne Collaborate Cohort Study (MCCS) that had attended wave 2 follow-up in 2003-2007 when aged 50-75 years. We generated 10-year absolute colorectal cancer risks and assessed their performance in terms of discrimination (C-statistic), calibration, and decision curve analysis. We also stratified calibration results by quantiles of risk and by prior colorectal cancer screening at baseline, assuming that 75% of prior screeners would continue to screen.
Results: The CRISP2 model was well-calibrated overall in the MCCS with the ratio of expected to observed cases being 1.07 (95% CI 0.95-1.20). Results by quintiles of risk and prior screening showed no evidence of over- or under-prediction (all P>0.26). The corresponding C-statistic was 0.75 (95% CI 0.72-0.78); an improvement on our CRISP1 model (C-statistic 0.70, 95% CI 0.67-0.73, P-diff<0.001). Decision curve analysis showed the potential net benefit of using CRISP2 model for individuals with a 10-year colorectal cancer risk of up to 6%.
Conclusions/Implications: The CRISP2 model provides valid 10-year colorectal cancer risk estimates and gave better discrimination than other published models. It can enable improved risk stratification and potentially aid individual decision making about prevention and early detection.