Cancer wars
The Cancer J. 4: 1: 4-5,1991.

Two groups of renown epidemiologists recently renewed their battle on the interpretation of cancer data. On one side are the "Oxford optimists", led by Doll and Peto, claiming that the "War on Cancer" is being won. They are opposed by "Environmental pessimists", led by Devra Lee Davis from the National Research Council in Washington claiming that the War is being lost. The turmoil started with a paper by D. Lee-Davis et al. in Lancet (1) that reviews age specific world-wide trends in cancer mortality. Among other the authors conclude that: "All forms of cancer are increasing in persons over age 54 except lung and stomach. . . . These increases are not attributable solely to diagnostic artifacts or increased access to health care although both of these factors may be involved". In six major industrial countries "mortality rates increased in persons over age 54 from cancer at some specific sites not linked to cigarette smoking, including multiple myeloma, cancer of the breast, brain and other central nervous system sites, and melanoma."

In their letter to the editor, "Oxford optimists" regard the findings as "uninteresting", "well known for years," even "boring". Sir Richard Doll who claims that "The war is being won" summarized his arguments in a paper which appeared in several scientific journals (2) and was discussed in the present forum (3). Even his data do not support his optimism. While in some rare cancers like, seminoma and childhood lymphoma, patients fare better, in most cancers e.g., breast cancer and prostate cancers, the prospect remains bleak.

The driving force behind this War is the question where to allocate funding, to treatment or to prevention. For if "Washington pessimists" are right modern industry poses a threat which should be evaluated. On the other hand you don't have to be an epidemiologist in order to realize that something is wrong in your intimate environment. Like the "supermarket-tomato" that looks tasty but does not smell like one.

One aspect of the "cancer war" is particularly disturbing and should concern anybody involved in cancer treatment and research. How is it possible that well known experts reach contradictory conclusions on the same observations. Don't they apply anymore simple statistical reasoning? After studying the data, optimists would be represented by the Ho hypothesis, pessimists by H1, both groups should agree on alpha and beta values and then be tested by simple significance tests. Why don't they settle their dispute in the same way? Is this kind of logic wrong, or do they hide some vital arguments so that both groups are essentially wrong?

Cancer epidemiology is haunted by a handful of biases tarnished with incomprehensible names, e.g., lead time bias, or Will Rogers phenomenon. Most biases originate from the fact that processes studied by epidemiology are continuous and evolve with time, yet epidemiology lacks means to study them in their entire complexity. Observations have therefore to be quantized, pooled, grouped and simplified, which introduces all kinds of inaccuracies, known as biases, that derogate the reliability of the method. Instead of admitting the limitation of their tool, epidemiologists correct, adjust, stratify and so on. The present cancer war is more a battle about "how to correct and adjust", since if you apply the Oxford method you become and optimist, while Washington corrections turn you into a pessimist. The following example should illustrate what is meant.

Fig. 1 depicts a hypothetical growth curve of a breast neoplasm. The horizontal lines depict two screening methods, breast examination and mammography. The tumor is clinically detected only when crossing one of the two lines, which is known also as tumor surfacing . With mammography the tumor surfaces at the age of 50y, while simple breast examination detects it only at the age of 70y. The fraction of clinically surfaced tumors in each age group is called "age specific tumor incidence". Since the introduction of mammography, more and more patients were examined by it, and if carrying a cancer they were diagnosed earlier. Tumor incidence in young rose while in the old it declined. Not being aware of this bias we would conclude that cancer threat increased in the young and declined in the old. Epidemiologists are well aware of this bias yet it illustrates how other less obvious biases may distort epidemiological reasoning.

Should one avoid incidence estimates and base his conclusions solely on cancer mortality? Here biases are even more profound. In addition to the above, mortality statistics suffer from the fact that patients may carry several diseases. Does an intercurrent myocardial infarction accelerate the decline of a cancer patient, and how much. And if the patient dies from ventricular fibrillation, was he killed by the ailing heart, by his tumor or by both? How to assess the contribution of each factor? Most epidemiological models presume that both diseases act independently while medicine teaches the opposite. A slight improvement in the prospect of arteriosclerosis may prolong patient's life until his hidden tumor had surfaced and kills him. Such a rise in cancer mortality would be attributed to an increased cancer threat while in reality it reflects better medical care.

This is the main lesson to be learned from the present cancer war. Instead of admitting the limitation of their method, epidemiologists correct, adjust, stratify and fight each other's biases. One Oxford optimist has recently conjured a new monster, known as Meta-analysis (4) in which these shortcomings are augmented thousand fold. The war highlights also the betrayal of modern oncology which turns its interest away from the patient to worship false prophets of epidemiology.



1. Lee Davis D, Hoel D, Fox J, Lopez A. International trends in cancer mortality in France, West Germany, Italy, Japan, England and Wales, and the USA. Lancet 336, 474-81, 1990
2. Sir Richard Doll. Progress against cancer: are we winning the war? Acta Oncol 28,611-621.1989
3. Zajicek G. Progress against cancer: are we winning the war? The Cancer J. 3,1,1990
4. Mann C. Meta-Analysis in the Breech. Science 249,476-480,1990.

lead time bias

Fig. 1 Lead time bias. The curve represents cancer growth. The horizontal lines are detection thresholds of two detection methods. When the curve crossesa threshold cancer is detected clinically.

Back to index