■ site map
BLOG: June 2010 - December 2013
II - Mammography
8. Mammography benefits: early detection
Despite the inherently high-noise images of breast tissues it produces, X-ray mammography did prove more efficient in early detection of apparently malignant growths than either clinical or breast self-examination. The term "early detection" is used nearly as often as "breast cancer" in connection with the screening mammography, yet hardly ever there is a mention of specific numbers: how much earlier?
For that, we'll have to turn to the breast cancer randomized controlled trials (RCT).
Based on the data on breast cancer stage at the detection/diagnosis in the screened vs. control population, it is possible to estimate the statistical (average) time differential between detecting breast cancer with and without mammographic X-ray screening. This time differential is called mean sojourn time (not to be confused with lead time, which is the actual time gain between detection and the outset of symptoms; it can be equal to, or smaller than sojourn time).
An attempt to come to some specific numbers with respect to how much earlier, on average, preventive X-ray screening detects breast cancer, was made by Shen and Zelen in 2001. Obviously, mean sojourn time is directly related to test sensitivity, and estimates for either depend on the available trial data and its quality. Some trials did not provide sufficient data, or were for other reasons seen as unreliable by the authors (shaded on table below).
For instance, for the
famous Two-County study they found study data only for the 70-79y group,
despite the numbers on mean sojourn time (MST) being published
for the entire study population by its leading author (Tabar) and its associates
at no less than
five different occasions.
It is important to note that the MST estimates are based on the cancer detection rate that includes all screening modalities used in a trial (M=mammography, C=clinical breast exam, S=breast self-exam). The four Swedish trials did not use physical exam, but in the Two-County trial screened women were encouraged to have monthly self-exam. The rest of trials did use physical exam in addition to mammography. Obviously, combined screening modalities will result in a higher MST than any single one.
Sensitivity of the clinical breast exam vs. mammography vary significantly with the age group and screening round, from nearly even at the 1st screen to about 1/3 at the 4th for the 40-49y age group, and about 1/2 to 1/10, respectively, for the 50-59y age group (Canadian trials data). Considering that most of the trial data was for early rounds, MST figure for mammography alone, in trials that also used clinical exam,
could be easily up to 50% lower for the 40-49y group, and up to 25% lower for the 50-59y.
Data on the breast self-exam sensitivity are very limited; it seems to be significantly lower than for clinical exam, although for both it varies very much with the degree of adherence to the proper procedure. If properly conducted, their sensitivity should be similar, but it hasn't been the case in most studies and trials - notable exception being the Canadian trial - with breast self-exam generally falling behind clinical exam in that respect. All considered, it shouldn't be unreasonable to assume, very approximately, at least 10-15% MST increase due to the use of breast self-exam, when combined with mammography (Two-County trial).
What the above table immediately shows is how little of relevant MST data from these major trials is available. The only large trial that appears complete with respect to this major aspect of screening is the Two-County trial. But the numbers differ significantly from one publication to another, and the data they were based on was never disclosed, except for the 1st screening round (which is normally the most efficient one detection-wise).
Numbers from the other two trials - Gothenburg (small sample, too low count of interval cancers) and Edinburgh (flawed randomization) - were seen as unreliable.
So, we're left with a few small pieces of data to play with: three out of the five remaining trials only provide MST data for a figure averaged over a wide age range, for all participants. Only Stockholm and Canada 1 and 2 give somewhat better - although still partial - insight into the magnitude of earlier detection due to screening. The Malmo trial, with a single screening mode (mammography) has seemingly too high MST compared to other trials, including those with two screening modes (New York HIP had dismal mammography sensitivity, only 39%, but combined with clinical exam's 47% it still should produce MST less than a third lower than a mammography-only trial).
It remains sore point of these trials that
most of them had inadequate design and/or execution, thus their data, in general, is unreliable.
Except for the aspects of sample size and obvious discrepancies between screen and off-screen detected cancers, Shen and Zelen did not address this problem.
Canada 1/2, arguably the highest quality trials, indicate MST for mammography alone of somewhat over 1 year for the 40-49y group, and somewhat over 2 years for the 50-59y group. But it still does not account for overdiagnosis, i.e. the fact that significant portion of those early detected breast cancers are quasi-disease, which would never become symptomatic. With this in mind, the more likely MST figures are somewhat less than one and two years, respectively.
In all, the available data give only a partial, fairly uncertain picture of what the actual benefit of mammography is in detecting suspicious breast tissue growth earlier. It seems that detection could come up to a year earlier in the 40-49y group, up to two years earlier in the 50-59y group, and somewhat more for women 60y and over.
But these numbers alone do not answer key question:
is this earlier
detection beneficial in terms of saving,
Let's take a closer look at it by focusing at the specific data for one of the above trials, the Malmo trial. According to Shen and Zelen, it has the highest MST of all - perhaps unrealistically high.
Swedish Malmo trial is one of the three found sufficiently reliable under closer scrutiny. Looking at its data, we'll try to come to the more specific answer on how efficient is screening mammography in earlier detection of breast cancer (or whatever appears to look like one on the mammogram) and, more importantly, how beneficial it is for the screened women in terms of survival.
Here are the figures for all breast cancer diagnosed
by stage within ten years from the beginning of Malmo trial, taken
(as a sum of "alive" and "dead" among those diagnosed with breast
According to it, a woman in the screening group had 86% and 83% better chance of being diagnosed with breast cancer at stage 0 or 1, respectively, while 17%, 4% and 31% less likely to be diagnosed at stage 2, 3 or 4, respectively.
However, less of a chance to be diagnosed in the last two stages could be more of a benefit with respect to breast cancer (BC) mortality reduction. If we look at the actually screened population, only two stage-4 and eleven stage-3 cancers were diagnosed here, compared to 32 and 27, respectively, in the control group. Reducing the number of deaths by 30% in the control group, to make them comparable with those for 70% attending screening in the treatment group, gives that a woman actually attending screening was 42% less likely to be diagnosed at stage 3, and eleven times (91%) less likely to be diagnosed at stage 4.
But most of the extra cancers in the screening group were diagnosed at stages 0 and 1. Isn't that beneficial too?
Apparently, not. That is what the mortality data in the expanded table suggests: despite as many as 177 more cancers diagnosed at stages 0 and 1, combined, at the end of 10-year period there was
only three deaths attributed to breast cancer fewer
Note that only a small portion of the extra cancers diagnosed at stage 1 or 2 in the screening group have been offset by more cancers diagnosed at later stages in the control group.
That directly suggests two things:
(1) detecting BC early as possible with screening mammography has no appreciable advantage over detecting it early enough, and
(2 most of these extra early cancers in the screened group weren't real cancers at all - rather abnormal growths that showed on a mammogram, but in the control group such growths
never developed into a malignancy.
According to it, in the first ten years of the Malmo trial, those fake cancers accounted for over 20% of all cancers diagnosed in the screening group. Considering that all of the extra cancers diagnosed in this group - 137 of them - were stage 0 or 1, of which the screened group had 176 more diagnoses in all, indicates that 3 out of 4 of those early detected "cancers", were not malignancies at all.
If so, the 83% better chance of being diagnosed at stage 1 or 2 for the screened group effectively
diminishes to some 20%, or so, better chance for the early detected abnormal growth that is actually breast cancer.
That certainly fits much better into the trial's BC mortality figures.
Did the fact that nearly half of BC deaths in the screened group - 31 in 63 - were among the non-attendants, negatively affect mortality rate in the treatment group? If we'd look only at those women who actually attended screening, would it take out one of every two breast cancer deaths, or so?
Let's take a closer look. First, by taking out non-attendants, we are making the screened group smaller. With the reported attendance rate for the trial being slightly over 70%, the actually screened group would have to increase by 40% in order to be size-comparable with the control group. This means that there would be not 32, but 45 deaths.
That would still make screening worthwhile, but there's more to it.
The much higher death rate (as the number of deaths vs. number of women) among the 6,000 non-attendees in the treatment group - 0.53%, or 71% more than in the control group as a whole - clearly indicates that it constituted a high-risk subpopulation in this respect. The reason for such a high mortality rate is obvious: 77% of cancers in this group were diagnosed at an advanced stage, 3 or 4. It is unlikely that it was due to some higher proportion of fast-growing cancers in this subgroup; with proper randomization, it should be similar to that in the rest of treatment group which attended screening.
these are women which for any of possible unknown reasons did not, or could not take as good care of their health as the others, and/or were more vulnerable for no obvious reasons.
Factors such as these cannot be effectively controlled through randomization.
There is no reason not to assume that such high-risk subpopulation existed within the control group as well. In fact, we have to assume this if we are to consider the two groups comparable, which is the basic requirement for trial validity. This high-risk sub-population could have been somewhat larger than in the screening group, or somewhat smaller; we'll assume it was similar in size, since it is plausible, and makes comparison possible.
If so, then the standard-risk sub-group making 71% of the control group would have 0.225% mortality rate. Multiplying it with 21,200 women in this group, gives 48 deaths. Three more than in the screened group which also consists only of the standard-risk participants.
The same figure can be obtained by simply deducting 32 high-risk deaths from the total of 66 in the control group, and multiplying by 1.4 (from 21,200/15,200, the latter being group size after after deducting 6,000 high-risk women) to get the number of deaths for the whole group of 21,200 standard-risk women.
So, assuming that the high-risk sub-population of similar size existed in both, screened and control group, high BC mortality rate among non-attendees within the screening group
does not significantly
affect its overall rate vs. that
The former still have only about 5% lower BC mortality risk.
But can even this small, statistically insignificant reduction be credited to early detection?
Looking at the data in the expanded table, we can see that among 32 BC deaths in the standard-risk subpopulation of the screened group, 8 (or 25%) were stage-3 and 4 diagnoses. In the control group as a whole, 31 of 65 deaths belonged to that category. Applying the 77% rate for stage 3 and 4 deaths from the non-attendees group to the 31 deaths within the high-risk subpopulation of the control group, gives that 24 of 41 deaths were within this group, and only seven within its standard-risk subpopulation.
Multiplied by 1.4 to bring them to the whole group size, gives 11 and 10 stage-3 and 4 deaths in the screened and control group, respectively. This implies that the rest of deaths - for stage 0, 1 and 2 diagnoses - total 34 in the screened, and 38 in the control group.
Does that suggest that there is a benefit, even small, from earlier diagnosis due to screening? Not really. Anyone with even basic knowledge in statistics would see at once that
this number of outcomes is too small to provide even reliable indication, let alone evidence of benefit.
Data from all eight trials combined is somewhat more reliable, but the number of trials is still insufficient. Each trial by itself is under-powered in this respect, with a significant chance of random deviations due to it. It doesn't help that only three of them were found adequate.
Best we can say based on the Malmo study numbers is: It is possible that earlier detection helps reduce breast cancer mortality, but we:
(1) don't know how much, and
(2) have no reliable indication it is significant.
What we do know, is that this possible BC mortality reduction benefit is
much smaller than the quite certain increase in rate of pseudo-cancer diagnosed as real cancers due to screening.
In the Malmo trial, there was 31% more breast cancers diagnosed in the screening group for all stages, and 42% more in the early stages (0, 1 and 2). But only a small portion of these extra diagnoses - probably less than a quarter - was actually malignant growth.
It should be noted that the follow up on Malmo trial after this period had the BC mortality reduction for the screening group more than threefold higher, making it unprecedented and unique in all trial follow-ups. At the same time, reliability of the trial data decreased significantly, leaving some doubts with respect to reliability of the reported figures.
Turning back to the advantage of earlier detection by screening mammography, it appears that there is little substance in it. Practically all of the extra diagnosed cancers are stage 0 or 1, and most of them would, according to the longer-term volume of detected cancers in the control (unscreened) group, remain asymptomatic. Unexpectedly,
this diagnostic "advantage" seems to be more likely to be causing harm, through overdiagnosis and overtreatment, than to benefit screened population.
That is probably why those few sufficiently reliable trials did not confirm significant benefit of earlier diagnosis, neither with respect to mortality reduction, nor less invasive treatment.
One piece of the mortality puzzle could be the recurrent cancers, which may be more frequent in the screened population, possibly due in part to their generally higher radiation exposure, with breast compression during screening being additional risk factor. Mortality rate for recurrent cancers is significantly higher, and would at least partly offset the benefit of earlier detection.
Another is that screening mammography is more likely to detect slow-growing cancers, for which an early detection is less critical.
Also, mammography-related overtreatment - initially, repeated mammograms, MRI and/or biopsy for false positives, then possible surgical procedure, radiation and/or hormonal treatment if something suspicious is found and treated without actually being malignant - as well as the related stress shouldn't be neglected as a factor negatively affecting survival rate in the screened population.
Finally, a simple plain number of cancers detected at stage 1, instead of stage 2, does not necessarily directly translate into reduced mortality. What matters is a type of cancer: if, for instance, screened population has higher incidence of more invasive and highly invasive cancers detected between screenings - which could be expected considering mammography radiation risk, false positives and overtreatment - that would also tend to offset statistical benefit of earlier detection.
Putting it all together, there is no doubt that X-ray mammography screening detects more abnormal (not necessarily malignant) growths, generally at an earlier stage than physical breast exam, self-exam, or no screening. But
the benefit from it in terms of breast cancer mortality reduction and, particularly, all-cause mortality reduction, is very elusive.
It is even possible that all-cause mortality is higher for the screened population, offsetting in part, or even exceeding its small BC mortality reduction.
On the other hand, it is all but certain that the increased rate of detection directly contributes to overdiagnosis and overtreatment of the screened population. Thus, the assumed "benefit of earlier detection" due to the higher screening sensitivity may have, in the real world,
turned into a net negative.
It is only due to the prevailing presumptuous attitude about screening benefits and, later, vested interests of those promoting screening and benefiting from it, that it took so long to come to this realization.