Cancer risk: Role of chance overstated

See allHide authors and affiliations

Science  13 Feb 2015:
Vol. 347, Issue 6223, pp. 728
DOI: 10.1126/science.aaa6799

The recent assertion from C. Tomasetti and B. Vogelstein that most variation in cancer risk among tissues is due to “bad luck” demands close consideration, especially as they go on to argue for increased focus on early detection (“Variation in cancer risk among tissues can be explained by the number of stem cell divisions,” Reports, 2 January, p. 78). Observations from cancer epidemiology and limitations in their analysis argue strongly against this conclusion. Most cancers show considerable differences in incidence rates between distinct populations. Rates change over time, and migrants soon exhibit incidence rates similar to their host country. Each of these is consistent with a major etiologic role for environment and lifestyle. Consequently, a majority of cancers are preventable, with primary prevention achieving notable successes and promising more (1).

In their analysis, the authors correlate total stem cell divisions in selected organs or sites, and lifetime risk of a particular cancer at those sites. There is much uncertainty in the estimates of total stem cell divisions for each cancer site, and the vast age-related fluctuations in cell division for some tissues are overlooked. Of greater concern is the lifetime risk of cancers. Their analysis excludes frequent cancers with major environmental causes (such as stomach, breast, and cervix) and oversamples cancers rare in all populations (such as osteosarcomas, small intestine, and medulloblastoma). Overall, the cancer sites included account for only 34% of the cancer cases in the United States (2). The choice of the U.S. population is also arbitrary. A different population with different cancer patterns would have provided different results.

We also take issue with the statistical analysis. Despite the reported correlation of 0.81, stem cell replication is a poor predictor of incidence rates at any given cancer site. The residual standard deviation of the log rates is 0.75, so the 95% confidence limits for the log rate of any cancer site are given by the linear predictor ±1.47 (i.e., 1.96 × 0.75). Converting from a log10 scale to an absolute scale gives an error factor of 101.47=29.4; i.e., the incidence rate may be 30 times higher or 30 times lower than the value predicted by stem cell division rates alone. This residual variation is consistent with large effects of environmental and lifestyle factors.

The role of chance underlying the onset of any individual cancer has long been recognized (3). However, although important for the individual, chance has little to say about the incidence rate in a population, or differences between populations. These are far better explained by exposure to environmental and lifestyle factors, allowing important opportunities for, and supporting implementation of, primary prevention.


Navigate This Article