Background: The current model used to preoperatively stratify endometrial cancer (EC) patients into low-and high-risk groups is based on histotype, grade, and imaging method and is not optimal. Our study aims to prove whether a new model incorporating immunohistochemical markers, L1CAM, ER, PR, p53, obtained from preoperative biopsy could help refine stratification and thus the choice of adequate surgical extent and appropriate adjuvant treatment. Materials and Methods: The following data were prospectively collected from patients operated for EC from January 2016 through August 2018: age, pre- and post-operative histology, grade, lymphovascular space invasion, L1CAM, ER, PR, p53, imaging parameters obtained from ultrasound, CT chest/abdomen, final FIGO stage, and current decision model (based on histology, grade, imaging method). Results: In total, 132 patients were enrolled. The current model revealed 48% sensitivity and 89% specificity for high-risk group determination. In myometrial invasion >50%, lower levels of ER (p = 0.024), PR (0.048), and higher levels of L1CAM (p = 0.001) were observed; in cervical involvement a higher expression of L1CAM (p = 0.001), lower PR (p = 0.014); in tumors with positive LVSI, higher L1CAM (p = 0.014); in cases with positive LN, lower expression of ER/PR (p < 0.001), higher L1CAM (p = 0.002) and frequent mutation of p53 (p = 0.008). Cut-offs for determination of high-risk tumors were established: ER <78% (p = 0.001), PR <88% (p = 0.008), and L1CAM >= 4% (p < 0.001). The positive predictive values (PPV) for ER, PR, and L1CAM were 87% (60.8-96.5%), 63% (52.1-72.8%), 83% (70.5-90.8%); the negative predictive values (NPV) for each marker were as follows: 59% (54.5-63.4%), 65%(55.6-74.0%), and 77%(67.3-84.2%). Mutation of p53 revealed PPV 94% (67.4-99.1%) and NPV 61% (56.1-66.3%). When immunohistochemical markers were included into the current diagnostic model, sensitivity improved (48.4 vs. 75.8%, p < 0.001). PPV was similar for both methods, while NPV (i.e., the probability of extremely low risk in negative test cases) was improved (66 vs. 78.9%, p < 0.001). Conclusion: We proved superiority of new proposed model using immunohistochemical markers over standard clinical practice and that new proposed model increases accuracy of prognosis prediction. We propose wider implementation and validation of the proposed model.