Statistics, Radiation and Childhood Cancers

A while ago, I tweeted that medical journalists should have a working knowledge of statistics. Perhaps that should be extended to those who perform peer review for journals as well. A report on Yahoo News via HealthDay [headlined “Greater Caution Urged for X-Rays in Pregnancy, Infants”] states “There's a small increased risk of cancer for children who had X-rays before they were 3 months old and those whose mothers had X-rays while pregnant, researchers say.”

A few paragraphs later comes this: “Children whose mothers had X-rays while pregnant had a slightly increased risk for all childhood cancers and for leukemia, though the increase was not statistically significant [emphasis added]. Children who had X-rays in early infancy had a small, non-significant [emphasis added] increased risk for all childhood cancers, leukemia and lymphoma.”

For those of you who still do not get it, an increased risk that is not statistically significant is the same as saying there is no difference in risk. A difference that is not statistically significant is simply not a difference at all. It could have happened by chance. It is misleading, if not downright dishonest, to state otherwise.

The paper actually does say that the risk for lymphoma only is statistically significantly increased but the odds ratio (odds ratio 5.14, 1.27 to 20.78) is quite wide due to the small numbers of patients in the study.

This was a case-control study of 2690 childhood cancer cases and 4858 age, sex, and region matched controls from the United Kingdom. It was published in the British Medical Journal on February 10, 2011.

There is heighten awareness in the literature about the risks of radiation from diagnostic tests. Real concern is warranted. But this report adds only noise and confusion.