Abstract
BACKGROUND: Colorectal cancer risk models routinely adjust for endoscopic screening because of (i) possible confounding with other risk factors and (ii) possible alteration of the natural history of the disease due to adenoma detection and removal. METHODS: In this study, we defined a subject as screen-covered (SC) if a colonoscopy was performed in the past 10 years and not screen-covered (NSC) otherwise. We created colorectal cancer risk models separately for SC and NSC subjects (HRSC, HRNSC) and then obtained a screening-coverage adjusted HR estimate (HRfull) based on a weighted average of ln(HRSC) and ln(HRNSC), with weight equal to the proportion of SC person-time in the Nurses' Healh Study (NHS) population. We illustrate these methods on 45,538 participants of the NHS, for whom we have updated endoscopy data from 1986 to 2014. RESULTS: The adjustment effects of screening were small for most risk factors, but the HR for age became stronger after adjustment [HRunadjusted = 1.59 (1.48, 1.71); HRfull = 1.81 (1.67, 1.97)], whereas the HR for current menopausal hormone therapy became weaker [HRunadjusted Estrogen only (E) = 0.62 (0.54, 0.73); HRfull E = 0.72 (0.60, 0.87)]. CONCLUSIONS: We have demonstrated a novel approach to control for the confounding effect of screening in colorectal cancer risk models. IMPACT: Improved methods to address the impact of screening in colorectal cancer risk models reduce bias in the estimation of associations with other risk factors. Additional applications and evaluation of this method in other populations would support the utility of this approach.