Review of breast cancer screening evidence
While welcoming Mike Richard's independent review (BMJ 2011;
343:d6843) of the evidence on this important policy for women, including
the publicity for women called for mammography, it is long overdue. For
many years maturing trials, proper scrutiny and new evidence has
increasingly questioned the evidence base on which the original decision
to implement a screening policy was made in the 1980's. It is obviously
incumbent on those responsible for screening policy to maintain the
policy, and the information, commensurate with a synthesis of the current
evidence at all times.
Sadly there has been no sign that this has been happening as Susan
Bewley (BMJ 2011:343:d6894) so succinctly argues. Rather there have been
consistent assertions from a majority (apparently!) of experts that the
new evidence amounts to nothing. All my attempts to discern explicitly the
nature, especially the evidence base, for this point of view have failed.
Indeed I have been asked to refrain from my polite but persistent attempts
to obtain citations for the frequent assertions in the press such as 'on
each occasion they have been comprehensively rebutted in the public domain
by various experts'. This is obviously frustrating if not annoying since
the validity of the screening programme relies entirely on scientific
evidence and debate - or it should.
As far as I can tell from this non-debate this has appeared to be a
question of believing that randomised trails are necessarily unbiased
while simultaneously seeming to favour certain trials for apparently
obscure reasons. The basic possibility of confounding in making this
contrast is so dominant that there is obvious sense in relying as far as
possible on randomised trials. But the interpretation, even of large
trials, must take account of possible residual bias in their conduct.
Therein lies the basic issue that the independent review must confront
both scientifically and wisely.
We need the most reliable unbiased estimates of the attributable
effect now of mammography on mortality from breast cancer and on
overdiagnosis. My recent review (BMJ 2010;340:c3106 ) pointed out the
range of these published estimates, which I thought was unacceptable on
current knowledge. It is and has been for too long.
Can I humbly suggest therefore that the independent review of all the
evidence takes explicit account from all the trials of at least the
1. What were the treatment actually being compared and how do they
relate to contemporary mammography screening. Who were in the 'control'
group, and what happened to them diagnostically? For example some include
physical examination in both arms, while some were offered screening when
physical examination, with or without symptoms, was not readily available
to the control group at all. And so on.
2. Were there differential opportunities for deciding on eligibility
between treatment and control arms and was this appropriately accommodated
in the report, or subsequently. Excluding extant cancer is much more
likely in a screened arm than in a control arm not seen at all.
3. Is it completely clear that designation of endpoints was performed
blind to allocation and if not could that have biased the results. Only a
mild tendency to assign borderline causes of death differentially in
favour of one's belief, knowing which group a women was allocated, can
have dramatic effects on the reported outcome.
4. Any evidence on the specific attributable effects not from large
contemporaneous randomised groups properly conducted is probably
irretrievably biased, however sophisticated the modeling might appear. The
opportunities for agenda driven analyses are altogether too tempting. Of
course observational data on population trends in mortality can provide
insight into the success or failure of screening among otherwise
comparable populations, but no effect estimates.
Independent advisors are required for this review that have never
published on breast cancer screening. They do need however considerable
expertise in epidemiological methods because that encompasses the essence
of the major current concerns. These advisors need to have a thorough,
theoretical and practical, understanding of the true nature of bias in
epidemiology - which is a core part of reliable epidemiology too often
Which is partly why the trial suggested by Mike Richards, comparing
in a randomised setting women screened aged 47-73 with women aged 50-70,
will again miss the point. Whether or not mammography now causes more good
than harm, what is point of knowing in ten years time whether this is
still true among a marginally widened age group. We know already that
only a very small effect is possible either way! It might be the largest
trial in the world but it is testing a foregone conclusion if only the
best epidemiological expertise could now provide light on the main effects
of screening from the heat of the systematic avoidance of scientific
debate. Let us hope it can, well before even more unnecessary heat is
generated. Women are now depending on it.
Klim McPherson FFPH FMedSci
Competing interests: No competing interests