Refining Risk, Reducing Burden: ENDORISK-2 and Preoperative Stratification in Endometrial Cancer

For patients with endometrial cancer (EC), balancing oncologic safety with surgical morbidity is a key concern—particularly in decisions around lymph node (LN) staging. While sentinel lymph node (SLN) biopsy has reduced procedural morbidity compared to full lymphadenectomy, it still extends surgical time and poses complications, especially when mapping fails.

The updated ENDORISK-2 model, published in the European Journal of Cancer, offers a more personalized, non-invasive tool for assessing LN metastasis (LNM) risk, using a Bayesian network enhanced with molecular classification and imaging-based myometrial invasion (MI) data.

The original ENDORISK tool estimated LNM risk using clinical and immunohistochemical (IHC) markers. In ENDORISK-2, researchers added two contemporary inputs: molecular EC subtypes (including POLE mutations and microsatellite instability) and preoperative MI assessments via expert transvaginal ultrasound (TVU) or MRI. This enhancement was validated in two independent European cohorts: Brno (n=581) and Tübingen (n=247).

Bayesian networks (BNs) offer an intuitive graphical interface, handle missing data flexibly, and incorporate expert knowledge, making them well suited for clinical settings. ENDORISK-2 preserves these features while reflecting current EC guidelines.

High Accuracy in Low-Grade Disease

ENDORISK-2 achieved strong discrimination for LNM prediction with AUCs of 0.85 (Brno) and 0.86 (Tübingen). Notably, in patients with low-grade histology, the tool classified 83% and 89% of patients, respectively, as having <10% LNM risk. False-negative rates were low—4.3% (Brno) and 2.2% (Tübingen)—suggesting the model could safely inform omission of surgical LN assessment in many cases.

The model supports flexible variable input based on available resources:

• Set 1: Grade, CA125, and 3 of 4 IHC markers (AUC up to 0.85)
• Set 2: Grade, CA125, MI assessment, and one biomarker (AUC up to 0.87)
• Set 3: Grade, CA125, molecular markers (POLE, MSI, p53), and thrombocyte count (AUC ~0.80)

This adaptability is crucial for extending use in low-resource settings, where access to molecular or imaging resources may vary.

Clinical Implications and Next Steps

With only ~10% of EC patients exhibiting LNM, the majority undergo SLN mapping with limited therapeutic benefit. ENDORISK-2’s non-invasive prediction model could allow many patients, particularly those with low-grade disease, to avoid even the minimally invasive SLN procedure—saving operative time, reducing anesthesia exposure, and avoiding downstream complications like lymphedema.

Compared to traditional machine learning models, the graphical transparency and interpretability of Bayesian networks may foster clinician trust and facilitate shared decision-making. Continuous risk estimation also enables nuanced discussions, rather than assigning patients to rigid risk groups.

While promising, the model’s current training dataset lacks patients treated with SLN-only staging and includes relatively few high-grade or non-endometrioid cases. Also, the 2023 FIGO classification update introducing a “no MI” category was not included, which limits postoperative applications.

A prospective clinical implementation trial (NCT07200466) is underway to assess ENDORISK’s real-world utility, including its impact on surgical decision-making and cost-effectiveness. Future directions include incorporating SLN-specific training data and multi-omics inputs to further refine survival prediction and treatment stratification.

Overall, ENDORISK-2 maintains high accuracy while integrating modern diagnostics, offering a robust preoperative decision aid. For patients with low-grade EC, it may help safely omit LN assessment, reducing operative morbidity without compromising oncologic safety. The model’s flexibility across resource settings and explainable AI framework supports its potential adoption in diverse clinical workflows.

Reference:
Lombaers MS, Reijnen C, Sprik A, et al. ENDORISK-2: A personalized Bayesian network for preoperative risk stratification in endometrial cancer, integrating molecular classification and preoperative myometrial invasion assessment. Eur J Cancer. 2025;231:116058. doi:10.1016/j.ejca.2025.116058