Is screening for breast cancer with mammography justifiable?
Hoda Anton-Culver, Ph.D.
The above question posed by Gøtzsche and Olsen in their publication in The Lancet (Lancet, 355:129-134, 2000) received enormous attention from the national media, scientists, breast cancer survivors, and the general public.
In reply to a 1999 study showing no decrease in breast cancer mortality in Sweden, the authors decided to review the quality of the mammography trials and a Swedish study that analyzed pooled results from previous studies (a meta-analysis). They also performed a meta-analysis themselves.
The authors reported a systematic review of eight randomized trials of screening mammography. The following trials were included in this study: New York, Edinburgh, Canada, Malmo, Stockholm, Goteborg, Kopparberg, and Ostergotland. The authors judged that six of the eight trials were inadequate for the meta-analyses because of imbalances in selection and randomization, particularly by age, and that these trials used flawed methods, particularly as far as the randomization is concerned. Results from the two trials that they believed were correctly randomized showed that there was no effect on breast cancer mortality or on overall mortality. Therefore, the authors conclude that screening by mammography for breast cancer is unjustified.
One of The Lancet editors, Horton, published a commentary on the controversial political aspects of the Gøtzsche and Olsen publication (Lancet, 358:1284-85, 2001). In their reply to this commentary (Lancet, 358:1340-42, 2001), Gøtzsche and Olsen indicated that they obtained a Cochrane review, which confirmed their findings that mammography screening is not valuable and breast cancer mortality as an outcome measure is misleading. They also showed that screening leads to more aggressive treatment and more unnecessary surgical intervention, particularly on lesions that may not always develop into invasive breast cancer. They concluded that “any hope or claim that screening mammography with more modern technologies than applied in these trials will reduce mortality without causing too much harm will have to be tested in large well conducted randomized trials with all-cause mortality as a primary outcome.”
A committee of the Institute of Medicine of the National Academy of Sciences reviewed the same evidence as Gøtzsche and Olsen but reached the opposite conclusion (Henderson IC, Regular Mammograms Remain a Crucial Tool, NY Times, Feb. 9, 2002). According to committee member Dr. I. Craig Henderson, they concluded that, “the preponderance of the evidence suggests that if a woman without any signs or symptoms of breast cancer has mammograms at regular intervals, she will substantially decrease her risk of dying from this disease.” The reason for the differing conclusions is that by excluding certain trials, Gøtzsche and Olsen introduced new biases to their study. “When all the results are pooled, the data show a clear benefit from mammography.”
Hoda Anton-Culver, Ph.D.
It is unfortunate that the majority of the screening mammography trials included in the meta-analyses by Gøtzsche and Olsen are flawed using their strict criteria for adequate randomization and inclusion. The main issues are related statistical approaches and using age and related measures as markers of adequacy of a trial. These statistical analyses were done on a group basis and have little value at this time on the practice of mammography screening for women at the individual level. However, the analyses provide a rationale for future high quality screening trials. The results of the meta-analyses show the importance of appropriate and vigorous scrutiny of methodology, strict adherence to protocol, and unbiased randomization. These recommendations must apply to not only mammography screening but also to other methods of early detection of breast cancer.
The eight trials included in these meta-analyses are very heterogeneous in many ways, including their populations, time frame, sample size, and design. Consequently, the rationale to pool the data from all or some of the trials is questionable. In addition, because of the large sample sizes in each trial, very small differences between the screening and control groups, particularly in age and other related variables, were found to be statistically significant. The statistical differences were used by the authors to judge the adequacy of the trials, even when the absolute differences are meaningless in terms of their biological effect.
The claim that mammography screening leads to over-treatment and consequently higher breast cancer related deaths is a complex issue that involves multiple factors, including the treatment decisions. Mortality as an outcome measure of the value of mammography screening is invalid. Evaluation of mammography screening should take into account the data on the recommended clinical management of the patient, which vary widely by geographic area, insurance availability, time, socio-economic status, and specialty of the treating physician. Perhaps outcome measures other than mortality are needed for the evaluation of the effectiveness of mammography and other methods of early detection of breast cancer.
What makes trials randomized?
In randomized trials, patients are arbitrarily assigned to a treatment or control arm of the study. This way, in an ideal situation, researchers distribute factors that may influence the outcome of the study, such as differences in age, socioeconomic status, or even hidden factors, evenly between the two groups and therefore end up with the treatment being the only significant difference between the two populations.