The Role of Adjuvant Radiation Therapy for Stage I Endometrial Cancer: Does Meta-Analysis Reveal the Answer?

annual death rate from uterine cancer is much less than that from ovarian cancer and only modestly higher than that from cervical cancer, suggesting that high-risk endometrial cancer is actually a relatively rare disease (4). In Northern Europe, where most of the randomized trials were conducted, racial and other demographic characteristics differ from those in the United States, and highrisk endometrial cancer may be even less common. For example, in the Netherlands, home of the PORTEC trials, the annual death rate from endometrial cancer is about 3 per 100 000 women, compared with an annual rate of 5 per 100 000 women in the United States (4,5). We know from these data and from numerous clinical-pathologic analyses that uterine cancer has a heterogeneous presentation and that most of the recurrence (and death) risk, particularly for patients with stage I disease, resides in a relatively small subset of patients whose tumors exhibit multiple risk factors. Ideally, clinical trials would focus on these higher-risk patients for whom the margin for improvement is substantial. However, the desire to assure rapid accrual often leads investigators to broaden inclusion criteria. In fact, this has been one of the primary weaknesses of the endometrial cancer trials. Studies have tended to include large numbers of patients with low-grade or minimally invasive cancers—patients whose baseline risk of recurrence may be less than 5% to 10%. For these patients, even relatively minor side effects from radiation therapy (or lymphadenectomy or chemotherapy) will outweigh the most optimistic level of benefit from that treatment. If the proportion of patients with low-grade or minimally invasive cancers in a trial is large, treatment-related morbidity experienced by patients with low-risk tumors can easily obscure real benefits experienced by the smaller number of patients who have a greater margin for improvement. This phenomenon was well described by Kent and Haywood in their discussion of the potential pitfalls of summary statistics (6). The phenomenon is also clearly illustrated in the results of the GOG-99 trial (7). In that trial, subset analysis of a high-intermediate-risk subgroup suggested a fairly large potential benefit that was completely obscured when the entire group of patients, which on average was a relatively low-risk group, was