Background: In the Minnesota Colon Cancer Control Study, which used guaiac slides to annually screen stool samples for blood, mortality from colorectal cancer was reduced by 33.4%. The reported sensitivity of this test for colorectal cancer was about 90%. However, results from another study estimated the sensitivity to be 25%-33%; other investigators have reported intermediate values. Given these contradictions, we examined screening sensitivity for colorectal cancer in the Minnesota study by several direct and indirect methods. Methods: In this reanalysis of data from the Minnesota study, we distinguished between sensitivity for colorectal cancer of the screening test (composed of six slides) and of the screening program (a series of such tests). We estimated screen sensitivity by adjusting the crude estimate from the final tests in each screening phase for colorectal cancer incidence in 5 years of follow-up, by modeling guaiac slide results at each screen as a function of the presence of occult blood, and by incorporating sensitive detection into a modification of a mathematical model developed by Lang and Ransohoff. Program sensitivity was estimated from the fraction of screen-detected cancers among all cancers diagnosed in screened individuals. Results: The crude estimate of program sensitivity was 89.4%, whereas the modified Lang-Ransohoff model estimates screen sensitivities at 94.1%-96.2%, consistent with the estimates from the other methods. Indirect measures, such as the association between the number of positive slides among the six slides in each set and the positive predictivity for colorectal cancer, are consistent with these estimates. Conclusions: The Minnesota study reduced mortality from colorectal cancer through use of a screening test with average screen and program sensitivities of about 90%.
Copyright 2017 Elsevier B.V., All rights reserved.