Personalized approach to malignant struma ovarii: Insights from a web-based machine learning tool

Area of Science:

• Gynecologic Oncology
• Endocrinology
• Computational Biology

Background:

• Malignant struma ovarii (MSO) is a rare ovarian neoplasm originating from mature thyroid tissue.
• Diagnosis and treatment of MSO are challenging due to varied presentations and low incidence.
• Identifying prognostic factors is crucial for improving patient outcomes.

Purpose of the Study:

• To analyze prognostic factors in Malignant struma ovarii (MSO).
• To develop and validate machine learning models for predicting MSO patient survival.
• To explore therapeutic strategies for MSO.

Main Methods:

• Retrospective cohort study using the Surveillance, Epidemiology, and End Results (SEER) database.
• Cox regression analysis for prognostic variable identification.
• Development and validation of five machine learning models (including multilayer perceptron, random forest) to predict 5-year survival using ROC curve analysis.
• Kaplan-Meier survival analysis to assess therapeutic options.

Main Results:

• The study included 329 MSO patients.
• Poor prognostic factors identified: older age, unmarried status, chemotherapy, and multiple tumors.
• Multilayer perceptron demonstrated the highest predictive accuracy, followed by random forest, gradient boosting, K-nearest neighbors, and logistic regression.
• Key predictive factors included age, marital status, and tumor count.

Conclusions:

• Developed machine learning models offer a practical tool for personalized prognosis assessment in MSO.
• Findings provide a comprehensive approach to guide clinical decision-making for MSO management.
• Further research can refine these models for enhanced clinical utility.