News Release

Misclassification of death may influence perceived value of cancer screening

Peer-Reviewed Publication

Journal of the National Cancer Institute

A statistical analysis of past randomized trials of cancer screening tests suggests that misclassifications in the cause of death may have biased the trial results in favor of screening. The findings appear in the Feb. 6 issue of the Journal of the National Cancer Institute .

Disease-specific mortality is the most widely accepted end point in randomized clinical trials of cancer screening. However, the validity of this end point assumes that the cause of death can be accurately determined. An alternative end point, all-cause mortality, depends only on an accurate determination of deaths and when they occur and, therefore, is unaffected by misclassifications in the cause of death.

To demonstrate that death misclassification may have occurred in past trials, William C. Black, M.D., of Dartmouth-Hitchcock Medical Center, and colleagues compared these two mortality end points in 12 randomized studies of cancer screening for which both end points could be determined. These trials involved screening for cancer of the breast, colon, or lung.

In five of the 12 trials, the two mortality end points suggested opposite effects of screening. The authors attributed these discrepancies to two forms of bias that affect the classification of the cause of death.

In one form of bias, deaths from other causes in the screened group are falsely attributed to the target cancer because the cancer was detected by screening. This form of misclassification, called sticky-diagnosis bias, biases the disease-specific mortality results against screening.

In the second form of bias, called slippery-linkage bias, deaths from the screening process or subsequent treatment are falsely attributed to other causes. For example, if an invasive evaluation causes a patient to have a fatal heart attack, the death may be attributed to a heart attack rather than the specific disease being evaluated as a result of the screening test. This form of misclassification biases the disease-specific mortality results in favor of screening.

The analysis by Black and his colleagues suggests that both forms of bias affected the randomized screening trials, but they argue that slippery-linkage bias had a larger effect. The authors conclude that all-cause mortality should always be analyzed and reported along with disease-specific mortality to ensure that major harms or benefits of screening are not missed because of misclassification in the cause of death. In addition, the authors note that “the selection of a high-risk population would help avoid the misinterpretation of statistical significance that could make a harmful screening intervention appear to be beneficial or vice versa.”

However, in a related editorial, Helen G. Juffs, M.D., and Ian F. Tannock, M.D., of Princess Margaret Hospital in Toronto point out that studies of high-risk patients may not apply to the general population because of differences in patient characteristics, the biology of disease, and the level of risk.

They say that even a well-conducted randomized study of cancer screening in high-risk groups using all-cause mortality could not determine whether screening is beneficial to the general population. They suggest that screening programs only be implemented in the populations in which clinical trials have shown a benefit.

###

Contact: Hali Wickner, Dartmouth Medical School, (603) 650-1492, fax: (603) 650-1730; DMS.communications@dartmouth.edu

Editorial: Vince Rice, Princess Margaret Hospital, (416) 946-4501 ext.5771, fax: (416) 946-4585; vince.rice@uhn.on.ca

Black WC, Haggstrom DA, Welch HG. All-cause mortality in randomized trials of cancer screening. J Natl Cancer Inst 2002;94:167–73.

Editorial: Juffs HG, Tannock IF. Screening trials are even more difficult than we thought they were. J Natl Cancer Insti 2002;94:156–7.

Attribution to the Journal of the National Cancer Institute is requested in all news coverage.


Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.