Analysis of mortality data from cancer screening studies: looking in the right window

Appropriate statistical analysis is required to measure the impact of early detection and treatment of cancer. The current practice of using cumulative mortality ignores both (1) the delay between early treatment and the time that any averted deaths would have otherwise occurred, and (2) cessation of these delayed benefits some time after screening is discontinued.

We use time-specific mortality density ratios to estimate the mortality ratio in the "window of influence." We then use time-specific incidence density ratios to assess the extent to which the removal of polyps and other possibly precancerous lesions detected by fecal occult blood screening reduces the incidence of colorectal cancer.

Applied to a theoretical example, the current practice of using cumulative mortality substantially underestimates the reduction in mortality achievable by early treatment. If there is sufficient time for the full impact to emerge, time-specific mortality patterns provide a more accurate measure. In a previous analysis of the screening study, the reduction in cumulative incidence in the screened groups was just under 20%. In our reanalysis, yearly incidence density ratios indicate that had screening not been interrupted, there might have been a 40% reduction in incidence.

Time-specific mortality ratios provide a more sensitive measure of the effects of early detection and treatment. Measures based on cumulative mortality are diluted by inclusion of deaths that occur soon after the initiation of screening as well as deaths that occur too long after the cessation of screening.