While our report benefits from the inclusion of a large sample size,
we recognize a number of important limitations. As with any study
using administrative data, we cannot control unmeasured confounding
factors that undoubtedly influenced not only the decision to utilize chemotherapy,
but also the duration of chemotherapy. UsingMedicare billing
data, it is difficult to precisely determine the number of cycles of
chemotherapy administered. To overcome this limitation, we used the
duration of chemotherapy as a surrogate for cycles of treatment as has
been previously described . Similarly, over the timespan of the
study, the therapeutic agents for ovarian cancer and the way these
drugs are delivered have evolved. We used a permissive definition of
chemotherapy to include treatment with any cytotoxic agent. A priori
we also recognize that a large number of patients did not undergo comprehensive
staging andmay have had occult disease.While this is a limitation
in that the reported stage is based on incomplete pathologic
assessment, these data capture a “real world” scenario of how patients
are managed surgically and how available pathologic data is used to
make decisions regarding adjuvant treatment. For some of the subset
analyses, particularly for low-risk patients, our sample size and power
were limited to detect small differences in survival. Lastly, our data
only includes elderly women and may not be generalizable to younger
patients.