P91 Integrative model for prediction of lymph node metastasis in endometrioid endometrial carcinoma

Introduction/Background In endometrioid endometrial cancer (EEC), current clinical algorithms do not accurately predict patients with lymph node metastasis, leading to both under- and over-treatment. We aimed to develop an integrative model combining protein data with routinely available clinical information and preoperative imaging to identify EEC patients who may benefit from more aggressive surgery including lymphadenectomy. Methodology Using reverse phase protein arrays, expression profiles were generated for 176 proteins in independent training (n=243, Bergen; Norway) and test sets (n=56, Bergen, Norway and n=100, MDACC, Texas). LIMMA analysis identified significantly differently expressed proteins between cases with and without lymph node metastasis. Generalized linear models were then constructed selecting only the most informative proteins in addition to clinical data. Gene expression data from the same tumours were used for confirmatory testing. Results The main model, including fibronectin, cyclin D1 and tumour grade, predicted lymph node metastasis with AUC 0.79 (training); 0.88 (Bergen test set) and 0.83 (RNA expression data). The MRI model, along MRI including fibronectin and grade, resulted in AUC 0.83 (training and Bergen test set). Finally, in grade 1 and 2 EEC, a model was fitted using cyclin D1, SMAD1, fibronectin and beta catenin, with AUC 0.89 (training) and 0.72 (MDACC test set). High levels of fibronectin and cyclin D1 were, associated with metastatic lymph nodes (p Conclusion We show that data-driven integrative models, adding protein markers to readily available clinical information, have potential to significantly improve stratification of patients at risk for lymph node metastasis in EEC, including low-risk EEC. Disclosure Nothing to disclose.