Abstract
The correspondence of Tabar et al. and of Nyström essentially attempt to justify their statistical approach. Notably, no evidence is cited which is not linked to the Swedish trials, the focus of our comments. Two co-authors recently attempted to justify the incorporation approach using complex mathematical arguments. 1 The incorporation approach was used for the first time in 1992, 2 and it is surprising that it took so long to see a more formal explanation. The conduct and reporting of intervention studies in humans should be based on transparent methods validated by the scientific community at large. In this regard, a single publication in which it is necessary to have recourse to complicated mathematical arguments for substantiating the incorporation approach underlines the precariousness of this approach.
Tabar et al. evoke a lead time effect by which the screen detection of some cancers during the intervention period would have prevented the occurrence of clinical cancers during the post-intervention period. It would then be expected that the incidence of cancers in the screening group should decline at least transiently after termination of the intervention. However, the Goteborg trial showed that there was no slowing down of incidence trends in years following the intervention period. 3
The initial publication of the two-county trial reported a 31% reduction in breast cancer mortality. 4 However, all further reports on this trial had recourse to the incorporation method for keeping the reduction at 31%. Surprisingly, overviews that also used the incorporation approach reported mortality reductions of 22% or less for this trial.5,6 The 9% difference thus reflects the underreporting of breast cancer as the underlying cause of death in the screening group. If, in addition, the effect of incorporation approach was removed in a way similar to what we did in our article, then breast cancer mortality reductions would be in the order of 13% or less.
Nyström evokes notoriety and authority arguments that have no place in contemporary scientific discussions. Nystrom states that Swedish trials on mammography screening did not differ from large clinical trials. This is incorrect. Trials based on the left-to-nature design cannot implement blinding procedures of subjects and of health professionals that are typical of trials testing the efficacy of drugs. This limitation paves the way to cause of death misclassification and to biases due to differences in disease awareness and in patient management between randomisation groups.
We have acknowledged Nyström’s efforts for examining correlations between causes of death until 31 December 1989 reported on death certificates and by health professionals not involved in mammography trials. However, after 1989, causes of death were based on death certificates only, and it is totally unknown whether the correlation maintained until 31 December 1996, when there were about two times more deaths than on 31 December 1989. Of note, a German study on screening for cutaneous melanoma illustrates well the untoward consequences of the absence of blinding on the way doctors complete death certificates. 7
This important debate demands wider scientific scrutiny.