Comparisons of colorectal cancer mortality between screening participants and the general population are strongly biased unless an incidence-based mortality approach is used

Abstract

Objectives: A common approach in the evaluation of screening for colorectal cancer (CRC) is comparing observed numbers of CRC deaths in screening participants with expected numbers derived from CRC mortality in the general population. We aimed to illustrate and quantify an often-overlooked bias that may occur in such studies if CRC mortality in the general population is not restricted by the date of diagnosis (whereas screening participants by definition do not have a prior CRC diagnosis).

Study design and setting: We illustrate and quantify the expected bias using cancer registry data from the United States.

Results: Unless an incidence-based mortality approach is used, expected numbers of CRC deaths in screening cohorts (and hence estimated screening effects) are substantially overestimated. Overestimation of expected CRC deaths is most severe (more than fivefold) during the first year of follow-up and rapidly decreases in the subsequent years. Nevertheless, overestimation of 5- and 10-year cumulative numbers of expected CRC deaths is still as high as 60-70% and 20-30%, respectively. Substantial bias even persists if the initial years of follow-up are excluded from the analyses.

Conclusion: Careful restriction of expected CRC deaths by an incidence-based mortality approach is indispensable for deriving valid screening effect estimates.

Keywords: Bias; Cohort study; Colorectal cancer; Incidence; Mortality; Screening.