■ site map
BLOG: June 2010 - December 2013
II - Mammography
3. Mammography myth: Beginning of the end
Is mammography screening saving lives? The answer was resounding "yes" ... until a small Swedish study came along, indicating that such conclusion is not supported by available population-based evidence for Sweden (Sjönell and Ståhle, Mammographic screening does not reduce breast cancer mortality, 1999).
The 0.8% decline in breast cancer mortality for women using mammography screening found in this study, was "statistically insignificant", meaning it was small enough that it could have easily happened by chance (oddly enough, the authors ended up with the mammography cost per one life saved quite similar to that of Wright and Mueller: $1.8 million).
Even assuming that the 0.8% reduction was due to the screening, with the 10-year breast cancer death rate of about 5 per 1000 (40y and older), it would imply that screening 600,000 women for 10 years reduced the death toll among them from 3000 to 2976, or by 24 - less than one tenth of what the official mantra was advertising.
At the same time, during the decade analyzed, nearly 100,000 of the screened Swedish women have had false positive, resulting in about 16,000 biopsies and 4,000 mastectomies.
While the Swedes' study has limitations, it did prompt Danish National Board of Health to ask The Nordic Cochrane Centre to independently review related data. As a result, two researchers, this time from Denmark (Gøtzsche and Olsen, Nordic Cochrane Center, Copenhagen), went on to analyze all large randomized controlled breast cancer trials on mammography screening to date.
What the Danes found was that
most of them couldn't pass reliability test.
- treatment (i.e. mammography screening) and control (no screening) groups were not properly balanced
- those declaring cause of death knew which group deceased belonged to (absence of the crucial "blindness" requirement to exclude possible biases), and/or
- some participants were excluded after the trial was underway.
In addition, data on group sizes often varied unexplainably and significantly during and after trials. Some studies had number of participants in the control group mysteriously enlarged by up to several hundreds, either during the course of a study, or in later analyses. With the number of breast cancer deaths being typically 5-10% of total deaths, such uncontrolled, "under the radar" transitions of participants in and out, or between treatment and control groups, can significantly alter mortality figure.
While the data available was often incomplete, and/or failed to provide full transparency into the actual inner workings of the trials, the two researchers felt that the evidence is sufficient to conclude that
only 2 out of 8 large breast cancer trials had sufficient methodological quality to be accepted as reliable.
This is the summary of their findings (RR=risk ratio,
CI=confidence interval, with 95% CI implying that there is 95%
probability that the actual value is within given range).
Andersson et al. The Malmo mammographic screening trial,
Five of these trials - Malmo, CNBS (1992), Stockholm (preliminary report from 1991), New York HIP and Edinburgh - were among the six analyzed by Wright and Mueller in 1995. The only two that passed the scrutiny, Malmo and CNBS, indicated no benefit in breast cancer mortality reduction from mammography screening.
In conclusion, Gøtzsche and Olsen gave straight answer to the question posed by the study title: "We conclude that screening for breast cancer with mammography is unjustified.".
They pointed out that even taking the numbers from Swedish studies as they are (from Nystrom et al. Breast cancer screening with mammography: overview of Swedish randomised trials, 1993), the absolute 12-year mortality rate reduction of 0.1% in the screened population means that 1 woman out of 1,000 would be spared due to mammography screening in this time period.
At the same time, with the total mortality risk in the screened population 6% higher, the total number of deaths at their 10% mortality rate would have been higher by six (0.6%). In other words, for every woman saved by mammography,
six more women screened by mammography would die
In addition, at ~5% false positives rate - which is rather conservative - or 60% in a 12-year period of yearly screening, for every life saved by mammography, 600 women would be exposed to distress and unnecessary additional procedures, nearly 40% of which, or well over 200, would undergo biopsy.
Where, exactly, is the benefit?
It was as if Gøtzsche and Olsen poked a bee's nest. The flurry of responses in the Lancet, where the study was published, was only the beginning. Some were supportive, but the critics were more numerous, and much louder. Hey, you don't talk like that about reputable trials by even more reputable academic figures which, by the way, had cost the taxpayers a fortune; you don't question doctrines they helped create and support; you don't threaten big profits of the mammography industry in general, and medical radiology in particular.
And, of course, you don't imply that governmental health organizations in some of the most developed world countries, by approving mammography screening recommendation to their mature women populations for decades, did very poor job.
That's what the two Danes were up against.
Undeterred, they came up with an expanded database review, concluding it confirms and strengthens their findings (Screening for breast cancer with mammography. The Cochrane Database Syst Review, Olsen and Gøtzsche, 2001).
In January 2002, the Patient Data Query Editorial Board, an independent panel of cancer experts reviewing published evidence-based, peer-reviewed research for the U.S. National Cancer Institute (NCI), evaluated breast cancer random controlled trials and concluded that:
screening for breast cancer
does not affect overall mortality,
All hell broke loose.
∎ almost immediately, The American College of Radiology - which recommends annual mammography and MRI for women with lifetime breast cancer risk greater than 20% as early as age 25, and makes tons of money doing it - issued strong disagreement, calling the Danish study "flawed", and declaring the panel's announcement conflicting with "well-founded clinical trials" and official policy "based on accepted and rigorous scientific analysis" (wonder how did they figure out so quickly that they know all that for sure?)
∎ on January 31st 2002, The NCI (National Cancer Institute) issued more composed statement, saying that it carefully weighted all the evidence, and found no reason to change present recommendations for mammography screening
∎ on the same day, "An Open Letter to Women and Their Physicians", supporting current mammography practices and urging women to continue their screenings, was published in The New York Times, with a long list of signatories:
American Academy of Family
∎ February 2002: the U.S. Secretary of Health and Human Services issued press release reaffirming official support for current mammography recommendations by the U.S. Preventive Services Task Force (an independent panel of experts in primary care and prevention services), calling for mammography screening every 1 to 2 years for women age 40 and older (in fact, just in November 2009 the USPSTF issued different recommendations, advising against routine mammography screening in women aged 40-49, and leaving it up to the individual to opt, or not, for biennial screening)
∎ March 2002: a working group of the World Health Organization (WHO) concluded that the trials in question were not perfect, but that their deficiencies do not invalidate their results, indicating - according to the group - mortality reduction in the screened population age 50-69 of about 35%
∎ June 2002: "Global Summit on Mammographic Screening" in Milan, Italy, hosted by the European Institute of Oncology in collaboration with the WHO, the European Commission, the American Cancer Society, the American Italian Cancer Foundation, the European Society for Medical Oncology, the American Society for Clinical Oncology, the International Union Against Cancer, and the U.S. Centers for Disease Control and Prevention, called the Danes' study "unfounded", finding that breast cancer mortality reduction indicated by the trials in question seems to be 21-23%
∎ finally, in September 2002, USPSTF came up with its own statement, finding important "methodological limitations" in each of the trials in question, but rating only one as "poor", and proclaiming unlikely that the flaws would negate the overall significant mortality reduction these trials indicate; the agency reemphasized that this applies only for women aged 50 and over, and that it didn't find sufficient evidence to specify optimum screening interval for women aged 40-49 (the latter caused vitriolic reaction of pro-screening advocates, followed by loud accusations that the USPSTF recommendation recklessly endangers American women)
This grand-scale negative response by major organizations was accompanied by an avalanche of criticism of the two "offending" studies by the proponents of screening in the research field.
Sjönell and Ståhle, the small Swedish study that started it all, was labeled "disaster" (Nyström) that should have never been published. Main objections were:
(1) it did not adjust for deaths resulting from breast cancers diagnosed before the screening started, and
(2) it didn't account for the increase in breast cancer incidence.
Despite these objections being used as the basis for proclaiming the study "defective" and not worth consideration, they hardly can stand closer scrutiny. Breast cancer mortality trend is inevitably built with a lag between incidence and mortality for any given year (in other words, death, if occurs, takes place years after diagnosis). Not taking out breast cancers deaths diagnosed before the screening started does not prevent the detection of screening-induced mortality reduction, if present. It only requires long enough follow-up period, during which mortality rate would gradually decrease from its pre-screening level to that where no deaths from pre-screening period are left in the picture.
Looking at the 10-year BC mortality rates plot, we can see that most of the deaths occur within 5-6 years from diagnosis. Thus, if the reduction is significant, it can be detected 5-6 years after screening had started, and
definitely after 10 years,
which was the duration of Sjönell and Ståhle follow-up period.
Objection to their "neglect" of breast cancer incidence increase after the introduction of screening, in fact, exposes incompetence of the critics. As Sjönell and Ståhle point out, what has increased is nearly entirely the detection rate (mainly due to overdiagnosis) because of widespread screening, not the breast cancer rate itself. Relating number of deaths to this increased detection rate - as some of the critics have done - superficially lowers breast cancer mortality rate.
In all, Sjönell and Ståhle stud by their study. They also acknowledged its limitations, and that better a study - or studies - could and should be commenced.
Likewise, study by Gotzshe and Olsen hasn't been spared of criticism by screening proponents, although the critics were generally less dismissive. They mainly focused on the criteria used to determine reliability of random controlled trials, trying to minimize their importance to the point of making them irrelevant.
Dane's criteria were simple: a high-quality study has to have:
(1) proper randomization, making screened and control groups comparable in that all known significant risk factors are nearly balanced,
(2) few and unbiased post-randomization exclusions, and
(3) reliable outcome data (e.g. determining cause of death blinded to study group)
According to the degree to which these three criteria were fulfilled, a study is ranked as high-quality, medium-quality, poor or flawed.
Only two trials - Canadian and Malmo - passed as medium-quality, with the rest of them ending as poor or flawed. As authors noted,
the very fact that trials with the longest
interval between screenings, the fewest screens per
participant, those that started systematic screening of the control
group after as few as
clearly indicates questionable validity.
Closer inspection did reveal significant shortcomings in one or more of the three criteria for all but two trials.
Nevertheless, as a part of the effort concentrated on discrediting mammography doubters and reaffirm the orthodox view of it as highly beneficial, a few studies quickly surfaced with the numbers appropriate to the purpose. One of them, with the data from five Swedish studies updated to 1996, came up with 21% mortality reduction for the screened populations in its 15.8 years follow up model (Long-term effects of mammography screening: updated overview of the Swedish randomised trials, Nyström et al. 2002).
The point of this study is unclear, since of the five studies it included only one (Malmo I) has been found acceptable, three were among those labeled as unreliable, and the last one, a continuation of the Malmo study (Malmo II) was fairly obscured (omitted from 1993 meta analysis of Swedish studies by the same authors), and after closer inspection also ranked inadequate by Gøtzsche and Nielsen (Screening for breast cancer with mammography, 2009).
Even with the Canadian study excluded, breast cancer mortality reduction was only 15% in the 50-59y group, and 31% in the 60-69y group (2003 article by Baines, Mammography Screening: Are Women Really Giving Informed Consent?, cites this study with 27% reduction for the latter, 12% for the former, and 9% for the 40-49y group; it is possible that the numbers were adjusted in the study errata from August 2002).
In addition, new evidence of flaws surfaced, as Nyström et al. opted to bypass the original trial mortality data and use the Swedish case-of-death register instead. They found only 10% breast cancer mortality reduction in the Östergötland part of the Two-County trial, for which as much as 24% mortality reduction was reported (The Swedish two-county trial twenty years later. Updated mortality results and new insights from long term follow-up, Tabár, Duffy, Smith et al. 2000). In this study, the authors not blinded for screening status reported
10 fewer deaths in the screened population, and 23 more deaths in the control group than what was found in the register -
an error that is mathematically impossible to occur by chance.
Such blatant irregularities clearly imply substandard quality, worse yet - tampering with the results, and should have disqualified this study from the analysis.
Nevertheless, Nyström et al. wrapped up their meta analysis and concluded that "the advantageous effect of breast screening on breast cancer mortality persists after long-term follow-up" and that "the recent criticism against the Swedish randomised controlled trials is misleading and scientifically unfounded".
Believe or not.
More studies followed; what
did they do to resolve the controversy, next...