June 18, 2014

# The screening problem

Nicely summarised by two paragraphs from a story in the Herald

In a separate breast cancer study published online by the British Medical Journal(BMJ), researchers from Norway and the United States found that mammograms carried out once every two years may reduce death risk by about 28 per cent.

About 27 deaths from breast cancer can be avoided for every 10,000 women who did mammography screening – or about one in 368, said the team after analysing data from all women in Norway aged 50 to 79 between 1986 and 2009.

The two prevention numbers — 28% of breast cancer deaths, or one breast cancer death for every 368 women screened — are the same, but they give a very different impression. [Note that this is the age range where mammography works best]

Thomas Lumley (@tslumley) is Professor of Biostatistics at the University of Auckland. His research interests include semiparametric models, survey sampling, statistical computing, foundations of statistics, and whatever methodological problems his medical collaborators come up with. He also blogs at Biased and Inefficient See all posts by Thomas Lumley »

• Martin Connelly

So … does this mean that not many people (Aged 50-7) in Norway) actually die of BC, so that 1 in 368 is actually a large % of that small number?

• Thomas Lumley

Yes, but it’s not just Norway. In NZ the proportion of deaths in women that are due to breast cancer is just under 4.5%, which is lower than many people expect. The ‘lifetime risk’ of 1 in 10 or so is for diagnosis, not death, and is assuming you don’t die of anything else before age 85.

Now, when you exclude deaths before 50 and after 80, that leaves about 400 breast cancer deaths per year. Preventing 28% of them is preventing a bit more than 110 deaths, and that would involve mammograms for the entire female population in that age range. The benefit is real, but it is not huge.

• Helen Robinson

You can also get very different impressions from the same numbers presented in different ways. So if you say ‘1 out of every 100 people undergoing this surgery die’, that sounds alarming, but if you say ‘this surgery has a 1% chance of the patient dying’, most people would write off the danger as next to nothing. Even though it’s the same statistic.