ARCANE debates about the black art of statistics rarely have the power to kill. Here鈥檚 one that does.
Last week, an article in a prestigious medical journal damned six influential studies, all of which concluded that screening by mammography can save lives. The headlines that followed were predictable: 鈥淏reast cancer screening is a waste of time鈥� was par for the course.
Some women will have read this and shelved their New Year鈥檚 resolution to have a mammogram. Some could well have a tumour that will, as a result, go undetected. For some of them, detection may come too late.
Advertisement
That would be tragic, not least because until last week few people doubted that screening was beneficial. Mammography may not be the ideal screening tool, but most large studies have found that for women over 50 and under 70, it reduces deaths from breast cancer by around 25 per cent. In The Lancet, however, researchers from the Nordic Cochrane Centre in Copenhagen ruled that most of those studies are statistically flawed and, thus, worthless.
The flaws all stemmed from the way women were assigned to be screened or not. In each study, the women should have been randomly allocated to two groups that matched in all the major characteristics that could influence the likelihood of developing and surviving cancer. But the Danish researchers found that in six of eight large screening trials, discrepancies existed between the two groups (see p 18).
In one study, the groups differed slightly in size, in another, women in the groups were not perfectly matched by socioeconomic status, while in the rest the average ages in the two groups differed by between one and six months. Because of these biases, the researchers rejected the six studies, all of which found screening to be beneficial. The two remaining trials found no advantage, so they concluded that mammography is 鈥渦njustified鈥�.
So, was it sensible to totally reject the six studies on the grounds of what appear to be minor discrepancies? People who conduct such studies say no.
Of course, there is no substitute for good study design, but anomalies are a fact of life for researchers conducting research that includes thousands of people.
Epidemiologists contacted by New Scientist argue that all the discrepancies found in the screening trials are well understood and can be corrected for. None of the objections voiced in The Lancet paper, they say, are new or serious enough to justify trashing the studies.
So what鈥檚 going on? The Nordic Cochrane Centre is one of 15 centres worldwide dedicated to what鈥檚 called evidence-based medicine. With the vast amount of research published every week, it has become difficult for doctors to keep up with which interventions have been shown to work and which have not. Cochrane groups review healthcare research and judge which interventions are supported by the evidence.
To undertake such a difficult (and laudable) task, the centres rely on strict rules for what are acceptable research practices and what are not. Not surprisingly, this strict adherence to rules does not always cope well with the imperfections thrown up by studies of the real world.
As Jack Cuzick, head of epidemiology at Britain鈥檚 Imperial Cancer Research Fund, puts it: 鈥淭here is no such thing as context-free statistics.鈥� In other words, there is no substitute for looking at the data and thinking about what they mean. Common sense tells us, for example, that small discrepancies in age would not make a big, unpredictable impact on the results of the screening studies.
It is likely that the scientific credibility of the six rejected studies will be restored in the letters pages of The Lancet over the next few weeks. Rebuilding public confidence will be much tougher. Some鈥攊ncluding the Danes鈥攁rgue that a consensus conference is needed to go over the research and find a definitive answer. One is planned for September. Let鈥檚 hope that all the scientists involved put their data on the table and settle the matter.
And even if they don鈥檛, the national screening programmes started a decade ago in places such as Britain and Finland have been going long enough to begin showing their value. Studies of women who have died from breast cancer, which look at whether or not they had regular mammography, should begin to reveal the impact of screening.
If, then, the results show that screening is having no impact on death rates, it will be time to reconsider. In the meantime, women should keep that New Year鈥檚 resolution. There is鈥攑robably鈥攏o new evidence that breast cancer screening is a waste of time.