Thursday, 18 November 2010

Lung cancer studies and confidence intervals

Author Chris Snowdon introduces the latest lung cancer study, in a post that reviews his research of lung cancer studies in the past, and points out that there are far fewer published now than up to five years ago.

I have found the discussion today (comments beneath the blog post) of particular interest, and helpful in explaining such mysteries as relative risks and confidence intervals.  If when testing the probability of lung cancer from SHS exposure, they get a confidence interval between 0.6 and 1.8, it spans a negative value and a positive value ... this surely means that the results could give a negative value. And yet there are people out there (mentioning no names) who want to persuade us that this is a statistically significant result involving people's actual deaths.

I think the participants in the discussion make it clearer!

Edit: A review of the new lung cancer study (Brenner, BMC Cancer, 2010) , supplied by its author in the comments to Chris's post, can be read here. The study considers exposure to passive smoke in non-smokers with lung cancer, and their exposure to paints and solvents, and to smoke-soot and exhaust. The results for passive smoke are not significant, the results for the other contaminants are significant, nearly trebling the risk of lung cancer (and the confidence intervals do not span zero!).

1 comment:

health quotes said...

The article is great and this provide more useful informations.Thanks.