We review the analytical methods of studies that asked whether a term
pregnancy transiently increases a woman's risk of breast cancer. These
analyses must separate the possible influence of a recent pregnancy f
rom that of two other correlated variables, attained age and age at la
st pregnancy. Most analyses have compared women of the same parity. To
determine, however, whether the risk of breast cancer is different th
an it would have been had the last pregnancy never occurred, it is nec
essary to compare the risk of breast cancer over time among women of p
arity N+1 with women of parity N, which has been done in a few studies
. To generate relative risk estimates that are independent of arbitrar
y coding decisions, we show that the analytical models must be more co
mplex than those in published studies. We used these models to compare
women of parity N and N+1 for breast cancer occurrence, using data fr
om the U.S. Cancer and Steroid Hormone Study. For the first 6 years af
ter delivery, first term pregnancy was associated with lower or unchan
ged risk of breast cancer, second pregnancy with higher risk, third pr
egnancy with lower risk, and fourth pregnancy with both lower and high
er risk depending on age at delivery. Given the inconsistent findings
between adjoining levels of parity and the wide confidence intervals a
round the estimates, we could not find clear evidence for or against t
he theory that term pregnancy is transiently associated with an increa
sed risk of breast cancer.