Effect of colonoscopy screening on risks of colorectal cancer and related death: instrumental variable estimation of per-protocol effects

Abstract

Background

We recently reported per-protocol estimates of colonoscopy screening on colorectal cancer incidence and mortality in NordICC, a large-scale randomized trial. Our results may be affected by residual confounding due to lack of detailed information on confounders. Here, we supplement our per-protocol analyses with instrumental variable (IV) estimates whose validity relies on an alternate set of assumptions but does not depend on the availability of confounder data. Individuals in the NordICC trial were randomized at a 1:2 ratio to receive either an invitation to a one-time screening colonoscopy (the invited group) or no invitation (the usual-care group). We used IV analyses to estimate bounds and point estimates of per-protocol effects of colonoscopy screening on colorectal cancer incidence and mortality after 10 years follow-up. Analyses included 28,220 participants in the invited group and 56,365 participants in the usual-care group. Participation in screening was 42%. In IV per-protocol analyses, the 10-year risk of colorectal cancer was 1.13% (95% confidence interval [CI]: 1.04, 1.23) with usual care and, depending on the assumptions, 0.66% (95% CI: 035, 0.95) to 0.74% (95% CI: 0.57, 0.95) in screened individuals (risk ratio of 0.59 [95% CI: 0.30, 0.98] to 0.65 [95% CI: 0.48, 0.87]). The risk of colorectal cancer mortality at 10 years was 0.29% (95% CI: 0.24, 0.33) in the usual-care group and 0.20 (95% CI: 0.09, 0.73) to 0.22% (95% CI: 0.08, 0.37) in the screened group (risk ratio of 0.71 [95% CI: 0.31, 2.89] to 0.79 [95% CI: 0.24, 1.42]). IV estimation of per-protocol effects suggests that colonoscopy screening reduces colorectal cancer incidence by 35 to 41% after 10 years.