A case-cohort study was designed to correlate various histopathologic and molecular variables with distant failure in endometrial cancer by analyzing phenotypic and molecular indices in hysterectomy specimens. From an overall population of 283 patients with endometrial cancer, we selected a cohort including all 49 patients who experienced any recurrence and 76 randomly chosen patients without recurrence. Expression of nuclear proliferating cell nuclear antigen (PCNA), MIB-1 (a marker of cell proliferation), and p53 was determined with digital image analysis, and cell membrane HER-2/neu and bcl-2 were quantitated visually. Ploidy and DNA indices were determined with flow cytometry. Overall, 6 immunohistochemical and 11 flow cytometric cases were eliminated because of technical inadequacies. Distant failures were defined as primary recurrences that developed outside the pelvis or vagina. Median follow-up was 91 months. Distant failures occurred in 13% of the patients. Cervical stromal invasion, positive adnexae, myometrial invasion >50%, positive lymph nodes, positive peritoneal cytology, lymphovascular invasion, grade 3 histology, nonendometrioid subtype, p53 >33%, strong HER-2/neu membranous staining, aneuploidy, S-phase fraction > or =9%, proliferative index > or =14%, and DNA index > or =1.5 significantly (P<0.05) predicted distant failures. However, a logistic regression model identified only p53 (OR=43.73; P<0.005), lymphovascular invasion (OR=11.59; P<0.001), and cervical stromal invasion (OR=11.29; P=0.001) as cogent predictors of distant failures. Only 3% of patients without any of these three predictors developed distant failures compared with 36% of those with at least one of the three (P<0.01). Thus, locoregional therapy may be insufficient when at least one of these predictors is present.