External validation of non-imaging models for predicting distant metastasis in patients with endometrial cancer

To evaluate two non-imaging models designed to predict distant metastasis (stages IIIC-IV) in endometrial carcinoma (EC). Both used preoperative histological and biological findings. One used primary tumoral size, the other did not.

374 patients operated on for EC by hysterectomy and at least bilateral pelvic lymphadenectomy were included. Patient's characteristics, preoperative histological, biological findings and primary tumoral size were used to calculate for each patient two scores (one for each model) for distant metastasis. The accuracy of the models was evaluated in terms of areas under the receiver operating characteristic curves (AUCs), rates of false negatives, and number of patients in the group at low risk to predict stages IIIC-IV.

309 and 65 patients had FIGO stages IA-IIIB and IIIC-IV respectively. Thrombocytosis and leukocytosis were not significantly different between patients who had distant metastasis and those who did not. CA125 serum level was significantly higher in patients who had distant metastasis (71.2 vs 32.0U/mL, p<0.001). High-risk preoperative histology and primary tumor diameter >3cm were more frequently observed in patients who had distant metastasis (55.4% and 39.9%, p=0.02 and 21.3% and 8.5%, p=0.003). The AUC were 0.65 [0.63-0.67] and 0.68 [0.63-0.67] with 54% and 93.4% sensitivity, 64% 19.1% specificity. Two hundred and twenty nine patients (61.2%) and 62 (17.0%) were classified as low risk; among them, 30 patients (13.2%) and 4 (6.4%) had final stage IIIC or IV.

Both models turned out to have a low discrimination power in our population. However, the score using primary tumoral size permits to identify a subgroup of patients in whom metastatic probability is low and lymphadenectomy unnecessary. Preoperative CA125 level, histological findings and primary tumoral size remain prognostic factors of stages IIIC-IV and should be included in predictive models.