Page 325 - 20dynamics of cancer

310 CHAPTER 15

quantitative analysis that can be molded to the data. Second, simple
comparative hypotheses create the back and forth loop between predic-
tions and tests that reveal causality.
Breast cancer illustrates the importance of breadth in analysis. Breast
cancer incidence rises rapidly through midlife and rises slowly after
menopause (Figures A.1, A.2); ovarian cancer follows a similar pattern
(Figures A.7, A.8). The rate of cell division in female reproductive tis-
sues declines after menopause. It seems natural to relate cell division to
incidence, because the rate of mitosis sets one of the major risk factors
in cancer. So we may easily fit a model in which mitotic rate shapes the
incidence pattern of breast and ovarian cancer.
My broad synthesis of pattern and process in cancer quickly shows
how little we learn from the fit of the mitotic rate model to breast and
ovarian cancer incidence. On the pattern side, breast and ovarian cancer
incidence do follow changes in reproductive status, but so do cancers
of the kidney, esophagus, and larynx in both males and females (Fig-
ures A.7, A.8).
The broad look at pattern in the Appendix shows that many cancers
have rising incidence through midlife followed by a tendency of the inci-
dence curve to flatten (declining acceleration). This common incidence
trend of many cancers suggests a universal process.
What sort of universal process might explain declining acceleration
later in life? In my theory chapters, I developed a broad conceptual
framework for how various processes of progression affect incidence.
That broad framework showed that many different processes cause de-
clining acceleration with age.
My favored explanation follows from a universal aspect of multistage
progression: as individuals age, they progress stochastically through
the early stages of disease. Later in life, they have fewer steps remaining
to overt symptoms. With fewer stages remaining, incidence accelerates
more slowly with age. This progression scenario fits the data. But I also
showed that environmental or genetic heterogeneity can fit the patterns
of declining acceleration. By looking broadly at the theory, we avoid
latching onto the first good fit.
The theory leaves us with alternative plausible hypotheses, which is
all that we should expect from a quantitative framework. But with so
many alternatives, some might feel that it is too hard to match biochem-