Although many of the trials were carried out decades ago, and did not necessarily produce valid or precise estimates of the reductions that might be expected from a sustained screening program, data from randomized cancer screening trials are still relied on by many task forces.
Increasingly, the focus is on non-experimental evidence, i.e. data from populations where organized screening programs have been introduced.
In the evaluation of the impact of such programs, before-after comparisons of cancer mortality rates need to take account of concomitant improvements in cancer care over these same decades. Time-, age- and place-matched comparisons, and attention to which deaths could/could not be averted by the screening program, are essential for valid estimates of benefit.
Using organized population-based programs of mammography screening for breast cancer as an example, we show that by ignoring these issues, many of the prevailing statistical approaches to the analysis of such population-based data underestimate the mortality reductions produced by these programs. Statistical approaches that can deal with these 'dilutions' will be described.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.