■ site map
June 2010 - December 2013
II - Mammography
19. Mammography benefits: all-cause mortality reduction
Since its introduction decades ago, the assumed key benefits of screening mammography were less invasive, more efficient treatment and reduction in breast cancer mortality (BCM), both thanks to earlier detection. If true, and if mammography procedure itself does not increase all-cause mortality risk, then the screened women population should also have it lower than unscreened women.
It is also assumed that if screening does reduce BC mortality, it automatically means lower overall mortality as well. Even seasoned medical professionals at the U.S. Prevention Task Force (USPTF) can mix it up. They say: "The newly updated meta-analysis by Nelson and colleagues confirms an earlier finding that screening mammography reduces mortality.", supporting that statement with figures related to breast cancer mortality reduction alone (Screening for Breast Cancer: U.S. Preventive Services Task Force Recommendation Statement, 2009).
This may have been an omission in the USPTF report, not ignorance, but many do tend to assume that reduced BC mortality directly reduces overall mortality, and that it is where the story ends. This, however, is too serious a matter to be left relying on wishful assumptions.
In other words,
the importance of the all-mortality figure is to confirm - or not - that screening does save lives.
What do the all-cause mortality figures from breast cancer randomized controlled trials (RCT) suggest?
The Nordic Cochrane center (NCC) systematic reviews, unlike USPSTF's and some others, does present this side of equation as well. Risk ratios for both, all cancers and all-cause deaths for the screened and control populations in the RCTs analyzed are given in the table summarizing the figures and study quality (bottom).
The all-cause mortality screened vs. controls is 0.99 for both, adequate and inadequate trials. In other words, screened women from all trials combined had 1% lower chance of dying in the observed period. Can it be due to the reduction in BCM? Maybe in part, but that is how far the guessing goes. Since BC deaths are only a small fraction of the total mortality figure, relatively small changes in the BCM rate
cannot be reliably traced to the total mortality figure, which can and does vary due to a number of other (uncontrolled) factors.
In the U.S., where cancer deaths make 22% of all deaths, and breast cancer deaths make 15% of all cancer deaths, BC deaths are only 3.3% of all deaths. If so, then every 10% of BC mortality reduction would reduce total mortality by 0.33%. Thus, the combined BC mortality reduction rate of 15% would reduce total mortality by 0.5% - one half of the actual all-cause mortality reduction ratio (0.99 risk ratio reduction) for the screened RCT population.
Curiously, if those numbers were true, they would imply that screening mammography helps as much in reducing mortality from all other causes of death, as with breast cancer - a universal life saver!
If 19% BC mortality reduction - the statistical average for all, adequate and inadequate trials - is used, the all-cause mortality for the screened population is still reduced by over 50% more than what results from this level of BC mortality reduction imply. And if we use the 10% BC mortality reduction from the quality trials alone, mammography screening
would be saving twice as many lives lost to all other causes, as it does for breast cancer!
Obviously, we cannot depend on these numbers alone. They are inherently too heterogeneous by their origin (i.e. trial) to be slapped together for an accurate result. We can only hope that the biases of the individual trials would be diminished in averaging them all up, but we cannot really know whether it is happening, or not.
And, again, none of the trials were designed to reliably detect changes in all-cause mortality. Even assuming that trial populations were adequately randomized for all-cause mortality - which is itself insurmountable task - it would require that all other death causes contribute unchanged proportions in the screened and control population during the period of observation. Since this practically has no chance of happening with such a heterogeneous population, nor, for that matter in a homogeneous one (since the rates fluctuate randomly for any single cause),
the all-cause mortality figure is bound to be uncontrolled factor.
In order to properly randomize participants for
mortality, the two groups would have to be homogenized not only
with respect to the risks of developing breast cancer and dying
from it, but also
with respect to any other possible cause of death. Frankly, that cannot be done.
Still, total mortality numbers in the breast cancer RCTs are
worth a closer look. Here's what they look like (trials shown
are those that had the evidence for this particular outcome
*Weight is applied by multiplying it with the corresponding RR, add up the products for all trials and divide the sum by 100
Large discrepancies in mortality rates mainly result from the differences in the average age of study populations, group size, country's death rates and trial duration, but the selection criteria from one study to another can also be a significant factor.
For instance, Canada had nearly 50% lower death rate than Sweden at the time, and the trial itself, due to selection, had still significantly lower death rate than the female population at large. On the other hand, study population in the Malmo trial had the death rate similar to that of the population at large which, with somewhat older trial population and ~10% larger group resulted in more than three times the death of the Canada 2.
However, the 0.99 total mortality screened vs. controls risk ratio changes significantly if we change the calculation method. If, instead of taking the weighted average of ratios for each trial (weighting itself has very minor effect here, with the average ratio without it being 1.00, 0.98 and 0.99 for the adequate, inadequate and all trials, respectively), we compare mortality rates directly, based on the actual deaths in the actual total of women ("subtotal" and "all trials" rows, bold blue), we get 1.18, 1.09 and 1.26 of screened vs. controls death rates ratio for adequate, inadequate and all trials combined, respectively.
In other words, measured with a different stick, the same data gives that
screened population had consistently higher all-cause mortality.
This may not be how a statistician would do it, but
it is based on the actual count of women and deaths - there can
hardly be a more relevant number. Of course, it does not
represent the actual average for the
So, what the direct numbers for all trials combined give is that the all-cause mortality for the screened population was 7.6 in 100, and 6.1 in 100 among controls. There is no cause-of-death bias here, so the only inaccuracy possible is that coming from improper randomization, random fluctuations in mortality rate (which can affect differently even two "perfectly" randomized groups, since no one can homogenize for genetics, toxic exposures, attitude, accidents, emotional wellbeing, etc.), and/or plain entry error.
This direct contradiction between the "statistical" 0.99 and true 1.26 all-cause mortality risk ratio for screened vs. control women in the breast cancer RCTs illustrates
how uncertain these numbers can be,
even when based on the same data - in addition to all uncertainties related to acquiring data itself.
Let's look at what the screened vs. controls risk ratios for
breast cancer and all cancers show when based directly on the deaths
vs. participants numbers in all breast cancer RCTs that monitored these
outcomes, with the all-cause mortality ratios from above added for
completeness (as before, data for all-cancers and all-cause
deaths is from the breast cancer RCTs that monitored these
outcomes, not necessarily all of the trials in the review).
The four adequate trials now indicate BC mortality slightly higher in the screened population, but the difference of 2% is insignificant (meaning that it does not raise above the level of random fluctuations). The three adequate trials indicate nominally identical, insignificant mortality increase for deaths from all cancers. But mortality from all causes is as much as 18% higher in the screened population.
BC mortality in the four inadequate trials is now even more reduced in the screened population: 32%, compared to 25% when calculated by averaging the ratios for each study (Table). Mortality from any cancer is only slightly lower (3%) in the screened population, but all-cause mortality is 9% higher. This last figure is alarming, even if the 32% BC mortality reduction figure is correct. Considering that BC deaths make less than 5% of all deaths, on average, it implies that
for every BC death avoided, screened population had about six deaths from other causes added.
And the all-cause mortality figures, as already noted, indicate higher total mortality for either adequate, inadequate, or all trials combined.
These figures are throwing entirely different light on the effect on mortality of the screening mammography, than when BC mortality alone is considered. Even the four inadequate trials come up with the likely significant net loss of life in the screened population. If this RCT tendency generally holds in the female populations at large, the conclusion is rather scary: Yes, screening in all likelihood does spare some breast cancer deaths, but
at a price of up to several times more women dying from other causes.
How could this be possible? Why would screened women have higher all-cause mortality rates than those not screened? Assuming proper randomization, there is only two things different for them:
(1) going through the screening procedure, and
(2) being subjected to more treatment than unscreened women.
Since there is zero research on the effect of screening related stress (primarily that caused by false positives) we can only speculate whether, or how much it contributes to all-cause mortality. For the same reason, a non-negligible effect cannot be ruled out, considering the sheer number of women that do get false positive result.
Likewise, screen radiation is a non-trivial breast cancer risk for vulnerable women, and so is breast compression at the screening. We could speculate that these are responsible for some of the extra breast cancers diagnosed among screened women. Most of those are diagnosed early, and successfully treated, but a few could slip through. However, those would primarily affect BC mortality; most likely, their effect is statistically negligible.
Not so with the effect of unnecessary treatment. Trial data show that the screened population is significantly more exposed to the invasive procedures following breast cancer diagnosis - mainly the consequence of its significantly higher diagnosis rate. It had 35% more mastectomies and lumpectomies, 20% more mastectomies and 32% more radiation therapies.
Screened population also had 4% fewer frequency of chemotherapy, and 27% less of hormonal therapy, but data on these two treatment modalities are limited to two Swedish trials, and were relatively infrequent: 19% and 9.5% average, respectively, for Malmo and Kopparberg (Screening for breast cancer with mammography, Gøtzsche and Nielsen, 2009).
Looking at the radiotherapy alone, the 32% higher treatment rate implies that for every 3 BC-diagnosed unscreened woman nearly 4 BC-diagnosed screened women
will be exposed to the increased risk of cardiovascular death due to the detrimental effect of radiation on cardiovascular system.
For early breast cancer, which is typical in the screened population, one study found that radiotherapy cuts BC mortality by 13%, but increases death rate from other causes by 21% (Favourable and unfavourable effects on long-term survival of radiotherapy for early breast cancer: An overview of the randomised trials, Lancet 2000).
How much would such risk increase change the all-cause mortality?
Assuming 5% BC mortality rate in the total mortality, and 80% BC survival rate, the number of all BC-diagnosed women is at the level of about 25% of all-cause deaths. Assuming cancer detection rate due to screening about a third higher (most of it being overdiagnosis) gives 33% for the screened population. Taking about 1% all-cause mortality rate for U.S. women around 60y of age, the 20% increase means that all-cause deaths in the no-screen population increase from 0.25 to 0.3%, and in the screened population from 0.33% to nearly 0.4%.
In other words, the increase in all-cause mortality is less than 0.02% greater for the screened women. This is only a gross approximation, but it does illustrate the magnitude - or minitude - of the extra all-cause deaths due to overdiagnosis and ensuing overtreatment.
Taking the estimated 15% reduction in BC mortality rate due to screening gives 0.75% - much larger number, which cannot be significantly offset by the increase in all-cause mortality of screened women due to overtreatment. It would require 2000 BC-diagnosed screened women - i.e. 500 more than in unscreened population -
to add a single death from all-causes due to more of invasive treatment in the screened population.
At the same time, at 15% mortality reduction rate, 45 BC deaths would have been avoided.
Even with the all-cause mortality doubled in those given invasive treatments - which is not impossible, since radiation treated women also have significantly higher incidence of lung cancer (Prochazka et al. 2002), and longer term health consequences of other forms of invasive BC treatments haven't been paid much attention to - those deaths would still be only a small fraction - about 1/10 - of the BC deaths avoided due to screening.
This is not surprising, considering that the BC-diagnosed population is only a small fraction of the entire population, and those overdiagnosed and overtreated due to screening even smaller. The extra risk is significant for BC-diagnosed women, but cannot significantly influence all-cause mortality for the entire population.
This implies that the causes for possibly higher all-cause mortality among screened women could only be related to the screening itself, or resulting from biased or corrupted trial data. As for the latter, the fact is that all large RCT trials are not only underpowered (w/o sufficient number of participants) for assessing effect of screening on total mortality, but also cannot be considered adequately randomized for it.
As for the effect of screening itself, it is still unknown what effect on the death rate has stress related to false positives. It may take months to resolve and, although it shouldn't have lasting negative effect on otherwise healthy women, there are certainly those whose resistance level is marginal, and whose health can be pushed to the downward spiral by such event.
If, just for the sake of illustration, we assume that 1 in 100 could be such a woman, that would add one all-cause death for every 20, or so, BC-diagnosed women, i.e. for every 4, or so, BC deaths, and for every 0.6 BC-deaths (at 16% BC-mortality reduction rate) averted due to screening.
In other words, for every two BC death avoided due to screening, there would be
five added from other causes!
Is that possible? At this point, we could only speculate on how probable it is, but certainly can't rule it out.
And this is not exactly a news. A 1996 meta-analysis of the five Swedish RCT found that the all-cause death risk was 5% higher in the screened population. However, study authors then "neutralized" it by making an adjustment for age, without reporting in the paper that the adjustment has been made (An overview of the Swedish randomised mammography trials: total mortality pattern and the representivity of the study cohorts, Nystrom et al.). It took the authors 3 years to acknowledge the adjustment publicly, and they probably wouldn't have done it if it wasn't for the critics who were pointing out that the numbers do not add up (Gøtzsche and Olsen, who found 6% increased all-cause death risk for the screened population, Scrabanek, and others).
The problem with such adjustment - other than it was covered up - is that it may imply that the two groups, screened and controls, were not properly randomized, i.e. were not comparable. If so, that makes all other results uncertain, including that for breast cancer mortality. And it is these very Swedish RCTs that mammography screening advocates point to as the prime evidence of the reduction in breast cancer mortality due to mammography screening.
Well, we can't have it both ways: take their results at face value when they are favorable to mammography, and make (hidden) adjustments when they are not. But study authors had very good reason to "correct" the data: it directly implies that for every 1000 women screened over 12 years, one screened women will be spared of dying from breast cancer, but
six more will die from other causes
for the net five deaths more (the simplest way to look at it is that breast cancer death make only about 1/30 of total deaths, thus 5% risk increase for the latter is effectively 30 times larger, or 150%, which compares to the 25% overall breast cancer mortality reduction found by this study).
Accidentally - or not - this coincides with the all-mortality rate obtained for all the RCTs with a direct count.
How is that possible? We've listened for decades, from the top of the official hierarchy, that screening mammography saves lives. In its 2002 report, World Health Organization goes as specific as to say that its life-extending benefit averages 2-day per women, per screening.
Now, if we look closer at how this number was obtained, we find that:
(1) it only accounts for breast cancer deaths,
(2) it uses 25% BC mortality mortality figure based on a non-systematic WHO review of trials, without assessing trial quality, and
(3) the 2-day life extension figure doesn't fit into the actual data.
If applied to the average control group mortality rate in the five major breast cancer RCTs it was based on - four Swedish and New York HIP - 0.4% for 10-year period, it gives 0.1% mortality reduction. That corresponds to 1 death less per 1,000 women, or 3.650 "live" women-days more. Averaged over 1,000 women, it gives 3.65 days over 10-year period. With annual screening, that comes to little more than one third of a day per screening, and nearly 3/4 of a day per biannual screening.
So, WHO seemed to be overestimating life saving benefit of screening mammography by nearly three to six times. More so considering that it didn't include the no-benefit Canadian trial, despite it being just as qualifying - if not more so - than the rest of them.
And, evidently, even without these flaws, its presentation of screening mammography life saving benefits
amounted to a gross misrepresentation,
due to exclusion of data on all-cancer and all-cause mortality.
WHO body may not have just purposefully misrepresented data, although there are clear indications of bias toward screening (e.g. mentioned exclusion of the Canadian trial). What could be more of a cause for its erroneous, misleading presentation of the benefits of mammography screening is a rather common phenomenon in the scientific community and society at large: once a paradigm has been established, observations contrary to it tend to be ignored, and anything supporting it is readily accepted at face value.
This "point of view" factor readily combines with a purposeful distortion of facts by those with vested interest in preserving the status quo - in this case the mammography industry (equipment, supplies), mammography medicine (radiology, treatment) and policy makers, as well as medical professionals and researchers who are either dependant on the former, or have their own career interest at stake.
As a result, any new, unorthodox and unsuitable facts are being put on hold - rejected or ignored - for as long as it takes for this complex balance of powers over scientific information to turn in favor of the new, more factual paradigm.
And the new paradigm, when it comes to the screening mammography boils down to this significantly different view of what are its benefits and risks:
yes, it does detect more abnormal tissue changes, and generally sooner, but this inevitably results in significant negatives (false positives, overdiagnosis, overtreatment) negating the benefit of early detection
it probably reduces breast cancer mortality, but less than what it was (and still is) claimed, bordering with statistically insignificant
its overall life-saving effect seems very questionable; it could even have a negative net effect at the all-cause mortality level
Or in short, the combined negatives of screening mammography seem to be outweighing its benefits. If so, the next question is, inevitably: "Is there an alternative to mammography screening?". More on next page.