There have been several diaries already regarding the US Preventive Services Task Force report that withdraws its previous recommendation in favor of annual mammography exams. I'd like to present the real data that support this decision.
First, my qualifications: I am a professor of biostatistics at a major medical school. I have been teaching medical students and residents about disease screening for over a decade, and have kept up to date on the evidence.
Second, the methology: the gold standard for knowledge in clinical medical science is the randomized study. Only by randomly assigning patients to a treatment -- or, in this case, to a screening modality -- can one avoid the bias from self-selection that plagues purely observational studies. One of the most famous examples of randomized studies appearing to contradict what we thought we knew from observational studies was when we discovered that hormone therapy for postmenopausal women was bad. (Had we really looked at the observational data we would have been more sceptical anyway, but we wanted to convince ourselves that we could do good. In fact, we were killing women.) Furthermore, to be totally free of bias, a randomized study must have complete follow up on participants and most have treatment groups that are identical in every possible respect.
Third, the evidence: The Cochrane collaboration is a renowned group of medical investigators who have been reviewing evidence for and against various treatment modalities:
http://www.cochrane.org/
The Cochrane collaboration evaluated all the studies of general population mammography screening and found three that qualified as a gold standard: One in Canada (which separately reported results for two subgroups in the study), one in Sweden, and one in the UK. I now present the results of those studies. First, do women screened with mammography die from breast cancer less than unscreened women? The relative risks below answer that question: A relative risk less than one means that screened women are less likely to die; a relative risk greater than one means that screened women are more likely to die.
Deaths attributed to Breast Cancer after 13 years of follow up
Study or subgroup Screening No screening Risk Ratio Weight Risk Ratio
n/N n/N M-H,Fixed,95% CI M-H,Fixed,95% CI
1 Adequately randomised trials
Canada subgroup a
105 deaths out of 25214 women in screening group
108 deaths out of 25216 women in no-screening group
Relative risk of death from breast cancer 0.97
95% confidence interval for relative risk [ 0.74, 1.27 ]
Canada subgroup b
107 deaths out of 19711 women in screening group
105 deaths out of 19694 women in no-screening group
Relative risk of death from breast cancer 1.02
95% confidence interval for relative risk [ 0.78, 1.33 ]
Sweden
87 deaths out of 20695 women in screening group
108 deaths out of 20783 women in no-screening group
Relative risk of death from breast cancer 0.81
95% confidence interval for relative risk [ 0.61, 1.07 ]
UK
105 deaths out of 53884 women in screening group
251 deaths out of 106956 women in no-screening gropu
Relative risk of death from breast cancer 0.83
95% confidence interval for relative risk [ 0.66, 1.04 ]
Pooled relative risk for death from breast cancer 0.90
95% confidence interval [ 0.79, 1.02 ]
A 95% confidence interval that includes one means that the results are not statistically significant -- they may be due to chance alone.
But that isn't the entire story. We don't just care about breast cancer, we want to improve overall mortality. We really haven't gained anything if someone dies from something other than breast cancer. So the studies looked at overall mortality:
Deaths due to any cause after 13 years follow-up
Canada subgroup a
413 deaths from any cause out of 25214 women in screening group
413 deaths from any cause out of 25216 women in no-screening group
Relative risk of death from any cause 1.00
95% confidence interval for relative risk [ 0.87, 1.14 ]
Canada subgroup b
734 deaths from any cause out of 19711 women in screening group
690 deaths from any cause out of 19694 women in no-screening group
Relative risk of death from any cause 1.06
95% confidence interval for relative risk [ 0.96, 1.18 ]
Sweden
2537 deaths from any cause out of 21088 women in screening group
2593 deaths from any cause out of 21195 women in no-screening group
Relative risk of death from any cause 0.98
95% confidence interval for relative risk [ 0.93, 1.04 ]
UK
960 deaths from any cause out of 53884 women in screening group
1975 deaths from any cause out of 106956 women in no-screening group
Relative risk of death from any cause 0.96
95% confidence interval for relative risk [ 0.89, 1.04 ]
Pooled relative risk for death from any cause 0.99
95% confidence interval for relative risk [ 0.95, 1.03 ]
-----
There you have it. Women in the screening group were just as likely to die as women in the no-screening group.
This is one of the most depressing things I have ever encountered in my career. I wish we had better cancer screening regimens, that actually did save lives. And mammography might well save lives for high risk women -- but we don't now how much higher a risk a women must be to make it justified, nor how much benefit mammography provides. The studies haven't been done that can tell us that.
Junk science is when one ignores the evidence. Unfortunately, we've seen plenty of it recently. We have had recent diaries correctly decrying the anti-science ideology of much of the Republican party. For us to reject the USPSTF report is just as anti-science.