Big data

Cancer screening: the importance of context

Louise Marston - 02.11.2012

This week, while Hurricane Sandy filled many front pages, several were also devoted to the publication of a study on breast cancer screening. It's nice to see a science and evidence story make the front pages for a change, but are we now any clearer on when screening is helpful?

At the heart of this story is a question we're really interested in at Nesta: how do you give patients useful information about health that makes informed health choices possible. One of the important elements for this is context for the information you are given, and a way to relate it to your personal circumstances.

It's a complex question, and as Gerd Gigerenzer pointed out in his excellent book 'Reckoning with Risk', and as David Spiegelhalter described in a recent programme on chance, humans are generally very bad at interpreting probabilities. We do better when those probabilities are expressed in terms of people rather than percentages. However, the language used in the study, although it was expressed as people, still leaves a lot to be desired in terms of context:

"It prevented 43 deaths for every 10,000 women invited to be screened, or one death per 235 women invited. Among those 10,000 women, 681 cancers would be diagnosed and 129 of those would be over-diagnosed - the mammogram would have picked up an otherwise undetectable tumour which the woman would never have known about."

Tim Harford described in a recent column some of the problems with terms like 'survival rates' and 'prevented deaths'. Why aren't we doing the maths? - 27 Oct 2012

He's also looked at how you can present these numbers in a more understandable way in Screening - it's all in the numbers on 10 Dec 2011

In effect, there is a funnel, with each new piece of information providing additional context. Of those invited for screening, some will go and some will not. Of those who go, some will pick up a problem. Of those who are told there is an abnormality, some of these will be false positives - it looked like it might be a problem, but turns out to be nothing; some will be what the recent study calls 'overdiagnosed' - genuine abnormalities that in the future will pose no threat to the woman's health; and some are abnormalities that will go become cancerous. Some proportion of this last category will be life-threatening. The information you are given about your future at each stage needs to be in the context of what is already known, and provide context on the impact of each choice as well as the risks. Some people are more cautious than others, and choices will be different. This woman describes the 'ultra-cautious decision she made, ultimately to have a bilateral mastectomy, but acknowledges that she wasn't told about all the downsides of surgery. Another preferred to acknowledge the risk but avoid surgery. She also suggests that a good way to avoid over-treatment would be to ensure there is time to absorb the diagnosis before you are asked to choose the treatment.

There are a whole series of cognitive biases that affect how we interpret all these numbers, and what decisions we make about them, along with our personal preferences. It's not just a matter of whether the information is 'correct' - context, language, visual representations and experience all play a part in how the information is understood, and how it is used to make a decision.  All these complicated factors come together when we consider how best to present factual knowledge about medicine to the people who really use it - the patients. These are some of the questions we are considering when thinking about the design of a future 'health knowledge commons' that is useful to everyone.