Correcting for non-compliance bias in case-control studies to evaluate cancer screening programmes

In the evaluation of uncontrolled service screening programmes for cancer, the case-control design is sometimes used, in which people who die from the disease in question are compared with live controls with respect to screening histories. Such a design tends to yield estimates of relative mortality in voluntary participants compared with non-participants. This may bias results, since compilers and non-compilers may differ a priori in ways which are not related to screening but which nevertheless affect the risk of death from the disease. We present a simple method, employing external data from previously published randomized controlled trials of screening, of correction for this bias. We illustrate it by using data from a case-control study performed within the invited arm of the Malmö mammographic screening trial, a prospective study from the service screening programme in two counties in Sweden, and a matched case-control study of mammographic screening in Florence, Italy.