■ site map
BLOG: June 2010 - December 2013
II - Mammography
5. Mammography research: last decade
Within nearly a decade since Miettinen et al. tried to save the life-saving reputation of the standard mammography screening by constructing their "mortality benefit lag" hypothesis, nothing changed on the mammography research scene. Studies coming from the pro-mammography side had as their first priority not the facts, but demonstrating that screening clearly does more good than harm.
what they determined to be sufficiently reliable evidence,
continued to disagree.
This partial chronological list of mammography research within
last decade, or so, outlines the big picture of this ongoing
confrontation. Studies that found statistically significant mammography benefits are
This partial chronological list of mammography research within last decade, or so, outlines the big picture of this ongoing confrontation. Studies that found statistically significant mammography benefits are marked∎, and those finding no significant benefit by ▢.
As the above list indicates, the aftermath of the two studies questioning the actual benefit of mammography screening was dominated by population-based observational studies "documenting" the benefit. Only a single randomized controlled trial (RCT, #6) took place in this time period. The rest were observational studies and evidence reviews. That implies even higher uncertainty since, unlike RCT where - if properly designed and conducted - at least the major known factors are controlled, observational studies
do not offer even conceptual reliability
(in fact, the International Agency for Research on Cancer advises against regarding observational studies as evidence for the screening effect), and quality of evidence reviews depend on data quality and data selection criteria.
The fact that most of the studies listed above belong to even more bias-prone category than a RCT - for which we've seen that can be biased if so desired - explains much of the seemingly senseless series of positive and negative results with respect to benefits of mammographic screening.
Since none of the studies that found significant screening benefit is either RCT, or systematic evidence review, their implication of the significant screening benefit is questionable.
Observational studies, as opposed to random controlled trials, are considered to be less reliable evidence, and there are good reasons for it. Any disease-related statistics in two large population groups, even when genetically uniform, is affected by the differences in their socio-economic structure, education, cultural, environmental, lifestyle and other factors. The effect of these so called confounding factors cannot be accurately determined, so a significant degree of uncertainty always remains.
The most recent annual report of the American Cancer Society illustrates well possible magnitude of a single confounding factor: the difference in education level. According to it, the overall rate of cancer deaths for women with at least 16 years of education is
only half that for those with no more than 12 years of education.
For man, the difference is even higher: those with high level of education have 2.5 times lower cancer death rates.
Of course, education level itself does not affect one's cancer death risk. It is a multitude of factors related to it - from better lifestyle choices to better healthcare in general and cancer treatments in particular - that does.
So, for instance, the "most important bias" of the first study on the list (#1), according to the National Cancer Institute, is that it does not take into account "dramatic improvements" in the efficacy of adjuvant therapies in the screening vs. pre-screening period. How significant is that omission? A recent study on breast cancer mortality in Norway (#13) found that only about 1/3 of the total breast cancer mortality reduction is attributable to screening. In other words, if everything else is at least nearly correct in #1 study,
the actual breast cancer mortality reduction figure is nearly 15%, not 40-45% as the authors state.
But it is very unlikely that everything else is nearly
correct. In theory, it may be possible to unscramble confounding
the accuracy of the final result is uncertain due to unknown magnitude of confounding factors.
As if it wasn't bad enough, the choice of statistical model and its underlying assumptions can be highly biased as well. A 2007 study by Olsen AH, Lynge et al. presents several such models which, when applied to the same raw data, give different results (Estimating the benefits of mammography screening. The impact of study design). One of the models gives an increase in breast cancer mortality in the screening vs. pre-screening period, which indicates that bias toward overestimating the benefit is possible as well. The authors, however, assert that the model most favorable for screening - the one they used for their 2005 study (#3 ) - should be preferred, without explaining why, and without attempting to validate any of the models.
The problem with their 2005 study is that the full benefit of screening found - 25% breast cancer mortality reduction among the screened population (Copenhagen area) - takes place after only three years from its introduction, remaining unchanged for the study period. This is in a direct disagreement with the generally accepted view that breast cancer mortality benefit from screening - if exists -
can become evident at least 5 years after implementation of screening,
gradually increasing after that to its full rate.
This objections, raised in 2005 by Gøetzsche and Jørgensen, hasn't been answered by study authors as of 2010.
A more recent study on the effect of screening in Denmark (#11) found no effect of screening on breast cancer mortality rate. Breast cancer mortality decline was actually greater in the non-screened areas (2% vs. 1% annually), with the highest decline rate being among women too young to benefit from screening (34-54 years of age, 5-6% annually).
Unlike the 2005 study of screening benefit in Denmark, which used complicated statistical model, this study is based on a direct analysis of the raw data. Also, it includes data for women populations not invited to screening, which could provide important indications with respect to the presence of other possible BC mortality factors beside screening.
Despite having the quality of being based on the complete actual data, study #11, as all observational studies, still suffers from potentially significant limitations possibly imposed by the differences between population groups in comparison. For instance, it is more likely than not that parts of the country where screening was first introduced have not only better healthcare in general, and breast cancer awareness and therapeutics in particular, but also higher education and overall standard level. If such gap continues to widen after the introduction of screening, it would become a confounding factor lowering breast cancer mortality rate in the screened population, hence creating apparent benefit from screening.
In the Norwegian study (#13) mentioned above, mortality reduction due to screening is estimated simply by deducting the rate for non-screened population from that for screened population. But if the dynamics of factors influencing the rate within the two groups had significantly changed during the observed period of two decades, such estimate is not accurate.
Authors of the Danish study (#11) comment on this subject somewhat ambiguously, first stating that Denmark is a homogeneous country, with its screened and non-screened regions having similar proportion of urban and rural areas, and then implying that non-screened regions may have lower treatment standards.
But their conclusion that the breast mortality reduction benefit from screening is marginal at best seems to be well supported by the results of other recent studies not coming from the pro-screening group of researchers (Duffy, Tabar, Olsen AH, Smith, and others). In line with such conclusion are the Norwegian study (#13), as well as the European study that found similar rate of BC mortality rate in countries with and without screening programs,
with the largest decline in women below the screening age
(Monitoring the decrease in breast cancer mortality in Europe, Levi et al. 2005).
Taking statistical modeling out of picture also shows similar breast cancer mortality rate trends for screened vs. unscreened population in the UK. In nearly two decades from 1989 to 2007, mortality reduction was identical - 41% - for 40-49y age group not invited to screening and for the invited 50-59y age group, for which screening started in 1988 (Mayor, UK deaths from breast cancer fall to lowest figure for 40 years, BMJ 2009).
The data clearly indicate that screening had relatively minor role in the overall reduction in breast cancer mortality rates that started in the early-to-mid 1990s in most European countries, Canada and the U.S. And mainly unexplained inconsistencies in how these trends did develop point to a set of factors other than screening that had dominant role in bringing out the positive changes.
For instance, Sweden had distinct downtrend in the breast cancer mortality rate among women 75y and older prior to the introduction of screening, which turned into generally stagnant trend after it (graph below). On the other hand, breast cancer mortality rate for women 70y and older in the U.K. had up trend for over a decade prior to the start of organized screening, turning into downtrend a few years after it. This age group was not invited to screening in either country, thus screening had little influence on the mortality dynamics. Other factors, obviously, did cause BC mortality to go separate ways in these two socio-economically fairly similar countries.
On a wider scale, such direct discrepancies are also evident in mortality trends for the segment of female population that should be most affected by screening. As plot below shows, the downtrend nearly coincided with the start of screening in the U.K. and Netherlands (even the most staunch supporters of screening do not claim it can have immediate effect on mortality rates), had been been postponed nearly a decade in the US, while preceded start of screening by several years in Italy, and by nearly a decade in Germany (where organized screening started in 2005).
Note that it may take up to several years from the official start of organized screening to its full implementation.
In all, without statistical manipulation - which happened to be the approach of choice in all studies "confirming" lifesaving image of the standard mammography screening - the actual data indicate no major effect of screening on breast cancer mortality.
If so, how is it that screening advocates find 1 breast cancer death less for every 100 women screened over 20 years (#10)?
If that was true, screening saves nearly 2,250 lives in the U.K., every year. That is over 50% more than the 1,400 figure given in the official NHS (National Health Services, U.K.) screening invitation leaflet (Department of Health, NHS Cancer Screening Programmes. Breast screening: the facts. 2006), which itself is in all likelihood a grossly inflated figure that cannot be traced to the actual data.
The latest systematic review of evidence from
randomized controlled trials (RCT), generally regarded as the most
reliable basis for estimating the effects of mammography screening,
found that the cumulative
BC mortality reduction in the screened population is at a 15% level over 13-year period.
With the average BC mortality rate in these RCTs of 0.426% over 13-year period, 15% reduction comes to 0.064%, or 1 BC death spared for every 1565 women screened over this period of time. Averaged over a single year, it is 15% of 0.033%, or 1 in 20,000 women.
Shown graphically, this BC mortality reduction would be similar to the BC mortality graph for Malmo trial's older sub-group (bottom), except that the period there is somewhat shorter, and mortality reduction somewhat higher.
How does the NHS figure of 1,400 lives saved annually fit into the actual data? According to U.K. statistics, every year about 10,000 (according to the Office for National Statistics) to 12,000 (according to the Cancer Research UK) U.K. women dies from breast cancer. About 1.5 million women is screened every year, with women 50-70y of age invited every 3 years. Thus, about 4.5 million UK women in this age range participate in the screening.
At the response rate of about 75%, this makes for a total of about 6 million woman in this age range, with about 1.5 million not participating. The 6 million total eligible population is generally comparable to a RCT "treatment group", since RCT compliance rate is commonly also significantly lower than 100%.
Obviously, these are only approximate figures, but suitable to illustrate the magnitude of the "official" exaggerations of the life-saving benefit of organized screening.
With about 0.06% yearly mortality rate for this age group in the last few years, its annual BC deaths total is about 3,600. Since, according to NHS, 1,400 BC deaths are spared due to screening every year, the figure without it would have been about 5,000 deaths, and the corresponding mortality rate would have been 0.083% annually. It implies 28% BC mortality reduction each year,
nearly double the 15% reduction figure based on the trials.
Note that the UK annual BC mortality rate ~(0.06%) is significantly higher than the average rate in the trials (0.033%). The higher mortality rate, the more lives saved at a given mortality reduction rate (for instance, if 10 out of 100 women dies w/o screening, mortality reduction of 10% avoids one death, or 1 in 10 women; with mortality doubled, it saves two lives).
The higher U.K. mortality rate is mainly a consequence of the higher overall U.K. mortality rate, and an older population segment (50+ years of age UK population vs. 40+ in most trials). BC mortality is higher in the U.K. than Sweden, where most of the trials took place, but even more it is a consequence of important differences between the typical trial population and population at large, most of them predisposing the former for a lower BC mortality.
It is specifically stated in both Canadian trials, arguably the most meticulous trials of all: "Compared with the Canadian population, the participants were more likely to be married, have fewer children, have more education, be in a professional occupation, smoke less and have been born in North America."
Also, treatment level provided during the trials and overall awareness of the risk factors are likely higher in trial populations than in the population at large.
Another possible reason why real-life populations will generally have higher BC mortality rate is that they, unlike RCTs,
include high-risk women.
Since this particular group is likely to have higher all-cause and BC mortality rate, and lower rate of treatment efficacy (i.e. lower BC mortality reduction rate), the two will partly offset in the corresponding number of BC deaths avoided, given as the inverse of a product of BC mortality rate and BC mortality reduction rate. If the increase in the former is significantly higher than the decrease in the latter, the number of women needed to screen for one BC death avoided will be significantly smaller than the averaged RCT figure, and vice versa.
If only the actually screened population is observed, as screening proponents in the research field advocate, then 1,400 lives saved would apply to 3/4 of total annual deaths - assuming that the two groups are comparable - in which case annual mortality reduction would be 37% a year.
That, however, is not considered to be proper approach, since:
(1) it is not realistic to assume 100% response rate to screening in any real-life scenario,
(2) screened and unscreened populations are not likely to be comparable (it is generally accepted that women undergoing screening are likely to be generally healthier, have higher socio-economic status, education, better lifestyle choices, etc.), and
(3) the best data that we have, from the RCTs, is also based on screening compliance significantly below 100%.
For those reasons, we'll stay with the 28% yearly BC mortality reduction implied by U.K. leaflet.
If we, however, turn to the RCT data (#9), it tells that the BC mortality reduction rate for women over 50 is between 6% and 30% in two adequately and five sub-optimally randomized trials, respectively. The weighted average for all trials is 23%, but leaning toward those two adequately randomized, we'll assume it - following the authors of the systematic review - to be around 15%.
So, if about 3,600 BC deaths annually is 85% of the deaths total that would occur without screening, the number of lives spared each year is 3600(15/85), or 635. In other words,
the U.K. leaflet probably overestimates BC mortality reduction benefit of the screening by more than twice.
In terms of percentage points, the reduction is 0.06/0.85=0.0706% annual BC mortality w/o/ screening times 15%, or 0.00106%; that gives 1 BC death spared for every 944 women in the eligible population of about 6 million, or 635 BC death less for the entire population.
Duffy et al. inflates exaggeration in the leaflet further, stating that one BC death is spared for every 200 women over 10-year period. That implies 1 BC death spared for every 2,000 women annually. Applied to the U.K. actually screened 50-70y population of about 4.5 million, that gives as many as 2,250 BC deaths spared each year. While the two, obviously, are not directly comparable, with Duffy et al, assuming 100% screening compliance, that is almost four times higher estimate than one based on the trial data which, as already pointed out
has built-in bias toward higher mortality reduction due to the general trial population bias
(healthier lifestyle, higher standard and education, higher BC awareness, etc.)
Following table summarizes this major discrepancy in the assessments of the screening effect on BC mortality.
Obviously, two key factors determining the number of BC death avoided are the reduction in the mortality and mortality rate. Their product (as ratio numbers, e.g. 15% is 0.15, and 0.033% is 0.000033) gives a number that is effectively the ratio of deaths avoided within given population (i.e. 0.15 times 0.00033 is 0.00005, or 1 in 20,000).
While BC mortality rate varies with the country and population sub-groups, the mortality reduction rate should be based on the trial data, simply because it is the best evidence of the screening effectiveness available.
How did NHS and Duffy et al. get to their figures?
As already mentioned, there is no explanation as to the origin of the NHS' figure of 1,400 BC deaths avoided annually. Extrapolating from their 1,400 lives saved figure gives nearly double the trial-based mortality rate (28% vs 15%). That implies that the UK BC mortality rate in the 50-70y female population would have been 0.083%, or 83 deaths per 100,000 w/o screening. With the trial-based 15% mortality reduction estimate, the mortality rate would have been 0.0706, or 71 in 100,000.
As for Duffy et al, they also used an inflated mortality reduction figure (30% vs. trial-based 15%), but also an exceptionally high BC mortality base (i.e. no-screen) rate, which is about 2.5 times higher than the actual UK statistics figure. Here's how the authors explain its origin.
Their 30% BC mortality reduction rate is based on the 24% reduction from non-systematic 1993 evidence review (which omits the Canadian trial) by the same principal authors; it is obtained by assuming 100% compliance with screening - unsound assumption for the reasons mentioned before.
The 0.167% mortality rate in Duffy et al. results from assuming 3% risk of dying from breast cancer in the 50-70y UK population; this, however, equals the lifetime risk of dying from breast in the UK, with the actual figure for the 50+ population nearly certainly not significantly higher than what it is in the US - about 1.2% for the 55-74y population (National Cancer Institute, SEER), or
two and a half times smaller.
Duffy et al. imply that the 3% BC mortality figure is based on the UK national statistics, but do not specify the source. With their 30% BC mortality figure larger by a factor of two from the RCT based 15%, and their 3% BC death risk figure about 2.5 times the actual risk, they arrive at 1 in 2000 screened women over 50 being spared from BC death each year (from 0.3 times 0.00167 giving 0.0005, or 1 in 2000, with the 0.167% mortality figure directly resulting from their assumed 3% BC death risk). With the screened UK population of about 4.5 million, that comes to 2,250 BC death spared annually.
Using correct figures, the trial-based 15% mortality reduction and the corresponding 0.071% no-screening mortality would give 1 BC death avoided in about 9400 screened women, or slightly more than double the benefit averaged in the trials. The higher number of BC deaths avoided is a consequence of the higher U.K. vs. trial-averaged mortality. But it is still
nearly 5 times lower
mortality reduction benefit than
This somewhat lengthy "closer look" should illustrate what is it that causes major discrepancies between figures on the effectiveness of organized screening presented to the public, either by research teams or by government agencies. Simply put, the influential pro-screening side
keeps producing benefit statements and figures that suit their purpose, without much regard for good science or actual data.
This is not exactly a new practice: for decades, organized screening for breast cancer was advertized as having significant benefits, with negligible risks. But the nature of deception has changed. The pro-screening bias was started by a sincere belief that "it must be beneficial". As under closer scrutiny during the first decade of the 21st century the benefits kept shrinking, and the risks grew more and more significant, ignoring overwhelming evidence to the contrary when stating mammography "benefits"
can only be explained by a deliberate deception of the public.
Whether continuation of organized screening is justified, or not, depends exclusively on its actual benefits-to-harms score. Knowing this score is also a must for making an informed decision for every individual women.
After three decades of mammography screening being the official policy, the benefits vs. risks subject is long overdue for clarification. Following page gives a graphic illustration of what the available evidence suggests is this vital score.