Colorectal cancer is the third most common cancer in the US population, with over 49,000 colorectal cancer deaths in 2008 among men and women. The literature supports many modifiable risk factors for colon cancer, which makes it an ideal context for generating risk prediction models and estimating the preventive potential of behavioral changes. Mathematical models of cancer risk estimate the effects of individual risk factors on lifetime risk of cancer, are the basis of clinical prediction tools, and inform individual and clinical decision-making. The need to develop and expand cancer risk prediction models has been documented by the National Cancer Institute;moreover, experts agree that it is critical to determine whether new models provide added benefit in comparison to established models. We developed a risk model for colon cancer using data from the Nurses'Health Study cohort, which included established risk factors for colon cancer and was similar to a recently developed model by Freedman et al. However, our model collected and updated important lifestyle, reproductive, dietary, and screening information prospectively over 25 years using validated questionnaires, and was limited to women. We calculated cumulative incidence and relative risks for various combinations of risk factors among over 80,000 average-risk women. Because our previous model was only for women, we plan to develop a parallel model for men. We will generate a log-incidence model of colon cancer risk using most known modifiable risk factors in an all- male cohort of health professionals and evaluate the performance of the model in predicting absolute risk. Before expanding our model for broader applications, our model requires validation in an independent population. This will allow us to evaluate the usefulness of the model outside of the population it was generated from. We will perform an external validation study of our model using the participants of the NIH-AARP Diet and Health Study. We will also determine if our model improves prediction compared to an already available model (Freedman model). By doing so, we will advance the field of prediction modeling of colon cancer risk. Lastly, we propose to use our model to develop a novel and comprehensive clinical risk prediction tool. A simple checklist, which can be administered online, will allow clinicians a way to educate and counsel their patients towards receiving colorectal cancer screening.
This project will result in an externally validated risk prediction model for colon cancer in men and women that will help us better understand the interplay of known risk factors for colon cancer on lifetime risk. We will also develop a clinical prediction tool that will help identify specific populations of individuals who may be at an above-average risk based on their risk factor profile, and who may warrant targeted interventions or specific screening recommendations.