Editor—In a letter Rosén and Rehnqvist criticised our analysis of mammographic screening in Sweden.1 We respond here to their criticism. The Swedish Board of Health and Welfare claims that mammographic screening has been shown to be effective in Swedish high quality clinical trials. A re-evaluation of these trials casts serious doubt on that conclusion.
A meta-analysis of five Swedish clinical trials showed a significant 24% reduction in mortality from breast cancer, but the control groups in four of the trials were invited to screening 4-5 years after randomisation.2 The fifth trial showed an excess mortality in the first six years of mammographic screening.3 This finding could be of great importance as screening in the control group would lead to false “excess mortality” and explain the significant decrease in breast cancer in the study group in the meta-analysis.2
Rosén and Rehnqvist consider the two county trial to be one of the best mammographic screening trials. In both the 1985 and 1989 papers, however, it is repeatedly stated that “the control group (passive study population) was not invited to screening.”4,5 In fact, screening started in the control group in 1982-3.2 With this in mind, can we stay assured that the positive findings in the statistical analysis of the trial are not the result of a “fishing expedition” whereby different ways of using the controls and of subgrouping the data were the means by which the magic P<0.05 finally appeared? Interestingly, the design of the two county trial was not presented in the papers in 1985 or 1989. The original protocol stated that the trial should be a five year, controlled, and cluster randomised trial with two rounds of mammographic screening in the study group and no screening in the control group.
Because several assumptions critical to the reliability of a controlled clinical trial are not fulfilled, estimates of the efficacy of mammographic screening can therefore be based on the results of only one of the five Swedish trials—the Malmö trial. Those results are compatible with our analysis of the effects of mammographic screening in daily practice in 17 Swedish counties (that is, no reduction in mortality from breast cancer). As the meta-analysis of 1993 includes the two county trial, thereby accepting intervention in the control groups in violation of the original protocol, its conclusions are invalid and not reliable.2