Explainable machine learning for predicting recurrence-free survival in endometrial carcinosarcoma patients

Area of Science:

• Oncology
• Machine Learning
• Medical Informatics

Background:

• Endometrial carcinosarcoma is a rare, aggressive uterine cancer with a poor prognosis.
• Its incidence is increasing, highlighting the need for personalized management strategies.

Purpose of the Study:

• To develop an explainable machine learning approach for predicting recurrence-free survival in endometrial carcinosarcoma patients.
• To leverage clinical, histopathological, and treatment data for risk stratification.

Main Methods:

• An explainable machine learning model was designed and applied to a cohort of 80 endometrial carcinosarcoma patients.
• The model utilized clinical, histopathological, chemotherapy, and surgical data.
• Patient recurrence-free survival was monitored over time.

Main Results:

• The model achieved a C-index of 70.00% (95% CI, 59.38-84.74), demonstrating reliable prediction of survival times.
• It effectively ranked patients based on individual risk scores.
• 32.5% of the patient cohort experienced recurrence.

Conclusions:

• Machine learning can assist clinicians in non-invasively and inexpensively identifying endometrial carcinosarcoma patients at high risk of recurrence.
• This study presents a novel, preliminary approach to predicting recurrence in this challenging cancer type.