© 2001 by Oxford University Press
Journal of the National Cancer Institute, Vol. 93, No. 21, 1585,
November 7, 2001
© 2001 Oxford University Press
MEMORANDUM FOR: Science Writers and Editors on the Journal Press List
A New Look at Some Old Data: Relative Risk and Absolute Risk Clarified
November 1, 2001 (EMBARGOED FOR RELEASE 4 P.M. ET November 6)
Confusing relative risk with absolute risk can lead to incorrect interpretations of the data from well-conceived and well-conducted studies.
This point is made by Thomas V. Perneger, M.D., Ph.D., at the University of Geneva Medical School and the Geneva University Hospitals, Geneva, Switzerland, in a Commentary appearing in the Nov. 7 issue of the Journal of the National Cancer Institute.
Dr. Perenger cites recent studies that seem to show that the relative risk of cancer in smokers, compared with nonsmokers, is greater in women than in men. He says that such conclusions are wrong because of the unfortunate substitution of "risk" for "relative risk," as compared with nonsmokers of the same sex. When case patients and control subjects are matched on sex, relative risks of cancer for women versus men cannot be estimated.
To illustrate his point, Dr. Perneger re-examined the data from a recent study that concluded, "The risk of bladder cancer may be higher in women than in men who smoked comparable amounts of cigarettes." Using the original data from this study, he divided the numbers of cancer patients in each sex–smoking category by the corresponding numbers of person-years to produce estimates of incidence rates of bladder cancer. These estimates contradict the original conclusions that women are at greater risk of bladder cancer than men. In fact, Dr. Perneger says that the incidence of bladder cancer is more than three times higher in men than in women, whether in ever smokers or in never smokers.
The author says that correcting misinterpretations of relative risk estimates requires epidemiologic training, cautious interpretation of results, and careful proofreading of manuscripts before submission and after editing. However, there is a second concern—the misunderstanding of statistical and biologic interaction. Statistical interaction occurs when different patterns of disease are found in subgroups of the population, such as different risks of bladder cancer associated with smoking in men and women. Researchers are often tempted to interpret such findings in biologic terms, such as a direct interaction between tobacco carcinogens and sex hormones. Dr. Perneger shows that this is imprudent: A statistical interaction is driven by the distribution of risk factors in the population and by the type of statistical model applied to the data, not by biology. This misunderstanding will be more difficult to eradicate, he says. The underlying concepts are fairly abstract and require familiarity with mathematical expressions. It has been suggested that the term "interaction" should be removed from the statistical vocabulary and replaced with something more precise. Dr. Perneger concludes that the use of a little jargon, such as "heterogeneity of effect on a multiplicative scale" may be an acceptable price to pay for better communication and that proposals to distinguish statistical and biologic interactions are most welcome.
Contact: Dr. Thomas Perneger, Switzerland, 41-223-729-012; fax: 41-223-729-016; thomas.perneger{at}hcuge.ch. (Note: on Nov. 7–10, Dr. Perneger will be away from the office but can be reached at 41-079-257-9489.)
_____________________________
Perneger TV. Sex, smoking, and cancer: a reappraisal. J Natl Cancer Inst 2001;93:1600–2.
Note: This memo to reporters is from the Journal staff and is not an official release of the National Cancer Institute (NCI) or Oxford University Press (OUP) nor does it reflect NCI or OUP policy. In addition, unless otherwise stated, all articles and items published in the Journal reflect the individual views of the authors and not necessarily the official points of view held by NCI, any other component of the U.S. government, OUP, or the organizations with which the authors are affiliated. Neither NCI nor any other component of the U.S. government nor OUP assumes any responsibility for the completeness of the articles or other items or the accuracy of the conclusions reached therein.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||