¿ìè¶ÌÊÓƵ

On a warm day in September 1990, clinicians and statisticians from all
over the world gathered in Oxford to hear the results of their labours.
The atmosphere in the lecture theatre was expectant. The researchers, from
the US, Europe, Japan and the former Soviet Union, were about to hear findings
of a rare scale and importance.

The final results have now been published, amid much fanfare, in The
Lancet (This Week, 11 January). Richard Peto and colleagues at the Imperial
Cancer Research Fund’s Cancer Studies Unit in Oxford based their findings
on data from 75 000 women involved in 133 medical trials. They found that
doctors could save some 10 000 lives a year worldwide if they gave women
with early breast cancer certain hormonal therapies and cell-killing drugs
after surgery.

Many doctors who treat women with breast cancer had heard the preliminary
results before they were published: the implications were simply too important
for the clinicians at the Oxford meeting to keep to themselves. ‘We have
been promoting this to doctors for a long time,’ says Michael Baum, professor
of surgery at the Royal Marsden Hospital in London and one of the investigators
whose trials were included in the data.

But there is a less publicised aspect of these findings which is of
fundamental importance. The trial results have been a good advertisement
for the technique of meta-analysis – that is, the systematic, statistical
overview of all the properly designed treatment trials for a particular
disease. In the past, many doctors have been sceptical about the value of
such overviews.

The main point of overviews is that they they can detect real differences
in the outcomes of treatments for common diseases where those differences
are too small to be detected by individual trials. There have been a handful
of overviews in the recent past, looking at treatments for other diseases.
The best known is an analysis of the trials of antiplatelet drugs in patients
who have had a heart attack or stroke. Published in 1988 by the same group
of researchers at Oxford, this convinced most heart specialists that overviews
were worthwhile. The results on breast cancer are the first to convince
clinicians treating cancer.

Overviews would not be necessary if it were obvious which treatment
was best. In breast cancer, the cell-killing drugs and hormonal therapies
used in addition to surgery increase only slightly the number of women who
survive for 10 years. Such modest differences are impossible to detect in
one-off trials involving a few hundred patients and, until now doctors have
been uncertain about which treatments to use. But when data from large numbers
of women are systematically analysed, the differences emerge at levels that
are statistically highly significant.

The studies have also shown certain other treatments to be relatively
ineffectual. For example, immunotherapy after surgery makes little difference
to the chances of surviving 10 years. This knowledge spares women unnecessary
treatments.

So the results are clear enough. But many critics have asked whether
it is safe to trust the methods of meta-analysis. How, they ask, is it possible
to combine data from more than 100 disparate clinical trials, over different
periods, in different countries, on different populations of patients, and
still manage to get sensible conclusions?

Peto and his colleagues say the fundamental assumptions of overviews
are medical, not statistical, and commonsense explanations are sufficient
to understand them. First, as the breast cancer trials have shown, even
modest gains revealed by an overview can translate into a big human impact.
For a common disease, treatments that can reduce the risk of dying by only
a small amount can save large numbers of lives.

The second point is that, on this scale, findings from one group of
patients are approximately applicable to another. ‘Although the same type
of treatment would probably not produce exactly the same size of therapeutic
effect in different circumstances,’ says the group, ‘it would probably at
least produce effects that tended to point in the same direction.’

Take, for example, tamoxifen, the best-known of the hormonal therapies
vindicated by the Oxford overview. Tamoxifen seems to block the action of
the hormone oestrogen. Although doctors are still unsure how it works against
breast cancer, they know that it is effective, particularly in women over
the age of 50.

Suppose two hospitals are conducting separate trials of the drug, in
which patients are allocated at random to receive either surgery and tamoxifen,
or surgery alone. Each hospital will have a different group of patients,
surgeons at the two sites will operate differently, each medical team will
give slightly different courses of tamoxifen, and so on. How can the two
trials be compared?

Clearly, the results from the two hospitals are unlikely to be the same:
there will probably be a difference in the size of the benefit conferred
by tamoxifen on the two groups. When the data consist of just two trials,
it is possible that one will show no benefit from tamoxifen at all, because
of chance factors. In 100 trials, however, the play of chance is most unlikely
to obscure the overall trend.

So the basic rule is that it is not necessary, or even expected, to
be comparing patients in one trial directly with those in another, provided
the number of trials is large. In statistical terms, this means calculating
within every trial the number of observed deaths minus the number of expected
deaths in the treated group. If the treated group did better than the untreated
group then the observed-minus-expected (O-E) figure will be negative. If
the treated group did worse, the O-E figure will be positive.

Next, the statistician simply adds up the O-E figures. If the treatment
had no effect, chance would dictate a mixture of posi-tive and negative
values for O-E, with atotal close to zero. If, by contrast, the treatment
worked, the results might still vary but will tend to be negative. The total
would almost certainly be strongly negative.

There are, however, important caveats that distinguish a bad overview
from one that is well conducted. First, it is not good enough just to look
at the data from the trials that are published. Trials that show no clear
benefit for treatments are less likely to be published than studies that
reveal large effects, because they do not seem to be helpful to doctors.
But they are vital in terms of the completeness of the data; omitting them
may lead to a bias in the final results. For the breast cancer overview,
the investigators believe they considered data from 90 per cent of all the
published and unpublished randomised trials of these treatments that began
before 1985.

A second constraint on the rules for overviews is that all the trials
must be properly randomised. If patients within a trial have not been allocated
to one or another group randomly, then the trial is invalid because the
treated and untreated groups are not comparable.

The Oxford group’s overview of breast cancer trials is a good example
of the technique’s proper use, says Simon Thompson, from the Clinical Trials
Research Group at the London School of Hygiene and Tropical Medicine. Not
only is it wide ranging, he says, it is also of practical use to doctors.
‘With this study, you can nail down the particular effect of a treatment
regime incredibly precisely.’ Thus doctors can disentangle the benefits
of one treatment from those of another.

However, Thompson warns that some doctors may still be sceptical about
extrapolating from the relatively old data on which Peto’s overview is based.
There may be problems, he believes, in applying the results to future patients.
‘For example,’ he says, ‘a certain group of patients may not be represented
in these data but they may be represented in future.’ Yet this criticism
is a minor one; the overall results are ‘incredibly useful’.

Once the fanfare for the results has died away, however, there remains
another crucial issue to be solved: how should these results on breast cancer
treatment be translated into real benefits for women’s health? Can we tell
whether doctors are using the best treatments now, and can we measure how
widely they are applied in future?

Apart from a few pioneering schemes, there are no mechanisms for monitoring
the treatments used by individual doctors in Britain, elsewhere in Europe
or in the US. Hence the only data available are from those women who agree
to take part in medical trials, who amount to only a fraction of all patients
treated.

Figures such as mortality data are too crude to be useful. In the case
of breast cancer, says Baum, it should be possible to detect an increase
in the number of long-term survivors in as little as five years, if all
doctors start applying the results of the study now. However, he says, the
figures will be complicated by the fact that breast screening programmes
should also start to have an impact on survival at about the same time.
Peto, however, warns that it is unscientific to interpret mortality figures
as an indicator of which treatments work. There are too many variables.
The only reliable answers come from randomised trials.

As a profession, doctors still cling to the notion that they should
be free to exercise their own ‘clinical judgement’ about what is best for
each of their patients. In breast cancer, this has been justifiable until
now because of the genuine uncertainty about which treatments were best.
Baum says most specialists now apply the results of Peto’s group, but women
still end up being treated by general surgeons who have no interest in this
research. ‘There is an argument for the lay public to demand a surgeon with
an interest in breast cancer,’ says Baum.

But should it be left to the individual woman to demand the right treatment,
or should the government intervene? Michael Peckham, Britain’s new Director
of NHS Research and Development at the Department of Health, has made it
clear that he wants the results of treatment research systematically introduced
into medical practice. Peckham’s team at the department published its first
policy document last September, and spelt out as an important aim the need
‘to see that good, relevant research findings are made use of by NHS clinicians
and managers’.

Peckham, himself a cancer specialist, is well aware of the wide variations
in medical practice in some areas of treatment. Even when information on
the best treatment is published, doctors have often been slow to change
in the past, he says.

The department’s new programme of R&D ‘will identify appropriate
mechanisms for promoting the uptake of health practice methods of proven
value’, says Peckham. These will take into account the competing options
for treatment, their cost effectiveness and the preferences of the individual
patient, he says. Although he will not be drawn on specific amounts, he
also says that ‘R&D funds may need to be made available to ensure that
appropriate methods . . . are introduced’.

Meanwhile, the fact remains that Britain’s death rate for breast cancer
is the highest in the world. This is not simply because it has a high incidence
of breast cancer; Sweden and the US, for example, have high incidence rates
but lower death rates than Britain. Certain doctors and press articles have
claimed that, because of old equipment or inadequate staffing, treatment
in Britain is worse than in other countries.

Such claims are far too simplistic, say researchers. It is ‘utter nonsense’,
says Peto, to try to assess different countries’ efficiency in treating
the disease by comparing their mortality rates for it. The treatments make
only a small difference to survival chances, whereas there are much greater
differences that are still not understood between the populations of women
in different countries, he says.

Although Baum accepts that there is still substandard treatment, he
says it is ‘rubbish’ to blame Britain’s relatively high death rate for breast
cancer on old equipment or too few doctors. Instead, he says, there are
other factors that affect Britain’s figures. First, British women still
tend to ‘keep their cancer to themselves’, he says. Twenty per cent of women
with breast cancer do not approach their doctors until the cancer is advanced,
by which time the chances of surviving are poor, whatever the treatment.
In other European countries, women tend to go to their doctors earlier.

Secondly, he says, in countries like Sweden and the US incidences of
breast cancer appear to be very high, which makes their death rates from
the disease look relatively low compared to Britain’s. These apparently
high incidences may be the result of artefact. In the US, women are screened
for breast cancer ‘overzealously’, says Baum, including those in age groups
where screening has no proven benefit. In Sweden, he argues, women tend
to obey requests to come for screening more often than in Britain or other
European countries. As a result, more early tumours are detected, including
some which Baum calls ‘favourable’ – those which may not cause disease for
some 30 years.

Whatever the reasons for the high death rate, the situation is unlikely
to change dramatically in the near future. With breast cancer, as with many
common diseases, there is no obvious breakthrough on the horizon. What there
is, however, is solid evidence of moderately helpful treatments that are
too valuable to ignore.