Lymph node metastasis (LNM) is one of the most important prognostic factors in endometrial cancer. Whether cancer has spread to nearby lymph nodes determines both surgical planning and the need for adjuvant therapy such as chemotherapy or radiation after surgery.
The standard approach -- surgical removal and pathological examination of pelvic and paraaortic lymph nodes -- provides definitive staging but adds operative time, risk of complications including lymphedema, and significant cost. Identifying which patients truly need lymphadenectomy before surgery would allow more targeted, personalized care.
This study aimed to build a preoperative prediction model using routinely available clinical and imaging data to estimate the probability of lymph node metastasis, potentially reducing unnecessary lymph node dissections in low-risk patients.
The study enrolled 254 patients from two hospitals: the National Cancer Center Hospital (NCCH) in Japan and Seoul University Hospital (SUH) in Korea. Patients from NCCH formed the training cohort; SUH patients provided an independent external validation cohort.
All patients underwent surgery for endometrial cancer with systematic lymph node dissection, providing definitive pathological lymph node status as the gold standard. The rate of confirmed lymph node metastasis was approximately 16% across both cohorts, reflecting typical clinical proportions.
Candidate predictors were restricted to variables available before surgery: preoperative CA125 blood levels, MRI findings including myometrial invasion (MI) depth and tumor diameter, presence of enlarged lymph nodes on imaging, and histological grade from preoperative biopsy.
The researchers evaluated multiple logistic regression models, testing different combinations of candidate variables. Model selection used the Akaike information criterion (AIC), which balances predictive accuracy against model complexity to find the most parsimonious combination of predictors.
The final model included four independent predictors: deep myometrial invasion on MRI (tumor invading more than half the myometrial wall), enlarged lymph nodes visible on imaging, elevated CA125 serum levels, and tumor diameter greater than a threshold size.
Model calibration -- how well predicted probabilities match actual outcomes -- was assessed using the Hosmer-Lemeshow goodness-of-fit test. Discrimination was measured by the area under the ROC curve (AUC), the standard metric for binary classification models.
In the full patient cohort, the model achieved an AUC of 0.80 in the training set and performed comparably in external validation at SUH. An AUC of 0.80 indicates good discriminatory ability -- substantially better than chance -- for identifying which patients have lymph node involvement.
Performance was notably stronger in the subgroup with high-grade endometrial cancer (grade 3 endometrioid, serous, clear cell, or carcinosarcoma histologies), where the AUC reached 0.84. This is clinically meaningful because high-grade cancers carry higher LNM risk and the consequences of missed metastasis are more severe.
By applying a risk threshold to the model, clinicians could identify a low-risk group in whom lymphadenectomy might be safely deferred, potentially reducing the number of unnecessary procedures by a meaningful margin while maintaining sensitivity for true metastasis detection.
Current surgical guidelines recommend lymphadenectomy for intermediate- and high-risk endometrial cancers, but the definition of "high risk" relies on preoperative assessments that are imperfect. A validated prediction model adds quantitative rigor to risk stratification beyond current qualitative criteria.
For patients classified as low risk by the model, surgeons could consider sentinel lymph node biopsy -- a less invasive alternative -- rather than full pelvic and paraaortic lymphadenectomy. This would reduce operative time, blood loss, and the risk of long-term complications like lymphedema.
The model is practical for clinical use because all four predictors -- CA125, MRI assessment of myometrial invasion and lymph nodes, and tumor diameter -- are routinely obtained in the standard preoperative workup, requiring no additional tests or cost.
This study provides a validated, externally tested predictive tool for preoperative LNM assessment in endometrial cancer. The model's good performance across two independent hospital cohorts in different countries suggests it is robust and generalizable.
Future refinements could incorporate molecular markers -- such as tumor mutational burden or circulating tumor DNA -- that are becoming increasingly available in clinical practice and may further improve prediction accuracy.
The broader goal is a shift toward truly personalized surgical staging: matching the extent of the operation to each patient's individual risk profile rather than applying uniform procedures to all. This model represents a meaningful step in that direction for endometrial cancer care.