A detailed description of the statistical analysis is provided in the Supplementary Appendix . We followed participants from the month of return of the 1988 baseline questionnaire through June 2010 for the incidence analysis and through June 2012 for the mortality analysis. We excluded participants with a baseline history of cancer (except for nonmelanoma skin cancer), ulcerative colitis, colorectal polyps, familial polyposis syndromes, or previous lower endoscopy (Fig. S1 in the Supplementary Appendix ). We used Cox proportional-hazards models to calculate hazard ratios and 95% confidence intervals. All analyses were stratified according to age (in months), sex (in the combined cohort analysis), and calendar year of the questionnaire cycle. Multivariate models were adjusted for known or suspected risk factors for colorectal cancer, listed in Table 1 .
For the incidence analysis, to minimize the influence of endoscopies performed for the diagnostic evaluation of colorectal cancer, we examined the association of endoscopy status reported on the biennial questionnaire before the diagnosis of colorectal cancer, death from any cause, or the end of follow-up, whichever came first. We used the most recently updated information for all variables before each 2-year follow-up and treated all variables as time-varying to account for changes during follow-up. For the mortality analysis, we evaluated the association of screening sigmoidoscopy or screening colonoscopy with mortality on the basis of the endoscopy status reported up to and including the date of diagnosis of colorectal cancer, death from any cause, or the last follow-up cycle, whichever came first.
We calculated the population-attributable risk, estimated as the proportion of incident colorectal cancers that would have been prevented in our population if all participants had undergone colonoscopy (with negative results or polypectomy) at least once and risk factors had not changed.31 (link) We also conducted a case–case analysis using a logistic-regression model to examine whether specific molecular features were associated with cancer occurring within 5 years after colonoscopy. All statistical analyses were two-sided, and a P value of less than 0.05 was considered to indicate statistical significance.
For the incidence analysis, to minimize the influence of endoscopies performed for the diagnostic evaluation of colorectal cancer, we examined the association of endoscopy status reported on the biennial questionnaire before the diagnosis of colorectal cancer, death from any cause, or the end of follow-up, whichever came first. We used the most recently updated information for all variables before each 2-year follow-up and treated all variables as time-varying to account for changes during follow-up. For the mortality analysis, we evaluated the association of screening sigmoidoscopy or screening colonoscopy with mortality on the basis of the endoscopy status reported up to and including the date of diagnosis of colorectal cancer, death from any cause, or the last follow-up cycle, whichever came first.
We calculated the population-attributable risk, estimated as the proportion of incident colorectal cancers that would have been prevented in our population if all participants had undergone colonoscopy (with negative results or polypectomy) at least once and risk factors had not changed.31 (link) We also conducted a case–case analysis using a logistic-regression model to examine whether specific molecular features were associated with cancer occurring within 5 years after colonoscopy. All statistical analyses were two-sided, and a P value of less than 0.05 was considered to indicate statistical significance.
