A new machine learning model uses simple blood and nutrition markers to predict which uterine cancer patients can safely skip aggressive surgery to preserve their fertility.

For young women diagnosed with endometrial cancer, the standard treatment of a hysterectomy is a devastating blow to their hopes of having children. Progesterone therapy offers a way to preserve fertility, but doctors have lacked a reliable way to predict who will actually respond to it. This uncertainty forces patients into a high-stakes waiting game while their cancer risks growing.

Traditional oncology often treats the tumor as an isolated island. This study challenges that dogma. It proves that systemic health—specifically how well a patient’s body manages inflammation and processes nutrients—directly dictates whether hormone therapy will succeed. This shift in perspective means oncologists must start treating the patient’s metabolic health as an active component of cancer therapy, not just a background detail.

Mapping the patient’s biology

Researchers at Peking University People’s Hospital analyzed data from 329 patients with endometrial carcinoma or atypical endometrial hyperplasia treated between January 2012 and December 2025. They used machine learning to screen 12 inflammatory biomarkers and 5 nutritional scores. By combining these metrics with four basic clinical factors—including BMI and metabolic syndrome—they built a predictive nomogram.

The results show a striking jump in predictive accuracy when these systemic markers are included. The combined model achieved an impressive Area Under the Curve (AUC) of 0.915 in the training group and 0.933 in the validation group. This easily outperformed traditional clinical models that rely on tumor characteristics alone.

Key performance metrics

• The combined inflammatory-nutritional score alone reached an AUC of 0.846 in training and 0.871 in validation.
• Integrating clinical factors like BMI and menstrual history pushed the final model’s predictive power to an AUC of 0.933.
• Decision curve analysis showed the model could reduce unnecessary surgical interventions by 35%.
• Patients categorized as high-risk had significantly lower complete response rates with a log-rank P < 0.001.

Why this matters

This matters because the current alternative is bleak. When progesterone therapy fails silently, patients lose precious time while the cancer progresses, often forcing an emergency hysterectomy anyway. By identifying high-risk patients early, doctors can pivot to aggressive treatments immediately, while low-risk patients can confidently pursue pregnancy.

However, the study has clear limitations. It is a retrospective analysis from a single institution, meaning the model must still prove its worth in diverse, real-world clinics. Until multi-center trials validate these findings, clinicians should view this as a powerful proof of concept rather than an immediate diagnostic standard.

Read the full study in Translational Oncology.