Researchers ask: What counts as a cancer death?
In measuring medicine's progress against cancer, the best indicator is the cancer mortality rate. That would seem to go without saying. But two Dartmouth physicians —H. Gilbert Welch, M.D., M.P.H., and William Black, M.D.—say that yardstick is not as reliable as it could be.
An article they coauthored— titled "Are Deaths within One Month of Cancer-Directed Surgery Attributed to Cancer?"— concludes that not all cancer-related deaths are properly attributed to the underlying cancer.
Welch and Black are both affiliated with DMS's Center for the Evaluative Clinical Sciences. Welch, an internist at the White River Junction, Vt., VA Medical Center, recently returned from a sabbatical at the World Health Organization's International Agency for Research on Cancer (see the adjacent story for details on his time there). And Black is a radiologist at DHMC.
Belief: Much of their research over the past decade, on the ef- ficacy of screening and diagnostic testing for disease, has called into question the belief that early detection and treatment produce unmitigated benefits for patients. In looking specifically at cancer, Welch explains, "We're trying to make judgments about how well we're doing against the disease. But who gets counted as a cancer death—people who die from the disease itself or those who die from the treatment?"
"The main purpose of this paper," Black explains, "is to show the medical community that cancer deaths are being undercounted." The article appeared in the July 17 issue of the Journal of the National Cancer Institute and was a follow-up to a study published in the February 6 issue of the same journal.
Rates: In the earlier article, titled "All-Cause Mortality in Randomized Trials of Cancer Screening," Black and Welch and their DHMC colleague David Haggstrom, M.D., concluded that analyzing only disease- specific mortality rates is likely to result in trials biased in favor of screening, because of misclassifications in the cause of death. For a more accurate analysis of the efficacy of screening, they propose including an alternative endpoint, all-cause mortality, which "depends only on an accurate determination of deaths and when they occur."
Basing their recent study on the assumption that all deaths within one month of cancer-related surgery should be attributed to the cancer, Welch and Black analyzed data from a five-year period (1994-1998), examining the recorded cause of death in patients diagnosed with one of 19 common solid tumors, such as of the breast or lung.
Of the 4,135 deaths among patients with one cancer, 1,707 (41%) were attributed to a cause other than the cancer. The number of deaths not attributed to cancer varied widely, though, from 12% for cervical cancer to 81% for laryngeal cancer.
While the number of incorrectly attributed deaths is obviously high among this study population, adding these numbers to the overall cancer death rate seems at first glance to make only a modest difference. Factoring in deaths within one month of surgery, the undercount would be approximately 1%. If the interval is increased to four months, the undercount would be 2%. And if all deaths within a year of cancer surgery are attributed to cancer, then the undercount would be 4%.
Trivial: "By itself, one percent looks trivial," Black says, "but one or two percent starts to look like a big deal in the context of 10 percent." He is referring to the 10.7% decrease in overall cancer mortality (excluding only lung cancer deaths) from 1973 to 1998. Including lung cancer skews the picture, Black says, because of the recent dramatic rise in lung cancer among people who began smoking 30 to 50 years ago. When lung cancer deaths are added, the decrease in mortality drops to 0.3%.
"A modest proportion of the reported decrease in non-lung cancer mortality could be a result of the misclassification of deaths from cancer-directed surgery," Welch and Black wrote in their article. "Equally important, this misclassification may be indicative of more widespread confusion about how to code treatment- related deaths in patients with cancer."
They do concede that no clear guidelines exist for classifying deaths related to cancer treatment. Using the World Health Organization's definition of underlying cause of death as "the disease or injury which initiated the train of morbid events leading directly to death," Welch and Black show how classification gets tricky.
A patient who had a lobectomy for an early lung cancer, for example, may be considered cured of cancer but could die of pneumonia six months later. Although the surgery increased the probability that the patient would contract pneumonia, the death would not now be counted as a lung cancer death. Similarly, some nonsurgical treatments for cancer can also increase a patient's long-term risk of death. For example, radiation treatments can increase the chance that a patient will develop vascular disease.
Welch and Black make several recommendations for ensuring a more accurate record of cancer mortality, including developing some simple rules—such as that all deaths within one month of surgery, radiation, or chemotherapy should be attributed to the related cancer. They also point out that a trend toward misclassification of cause of death has accompanied an increase in early detection of cancer.
Trends: Finally, they sound a distinctive note of alarm, concluding, "The more we look for cancer and the more we treat people with the diagnosis, the more important it will be to properly assign diagnostic and treatment-related deaths. Otherwise, observed mortality trends may make harmful interventions appear beneficial."
Catherine Tudish