Machine learning for epithelial ovarian cancer platinum resistance recurrence identification using routine clinical...

Area of Science:

• Oncology
• Machine Learning
• Biostatistics

Background:

• Epithelial ovarian cancer (EOC) frequently recurs after treatment.
• Identifying high-risk patients is crucial for timely intervention and improved outcomes.
• Predictive models for platinum resistance recurrence in EOC are needed.

Purpose of the Study:

• To develop and validate machine learning models for predicting platinum resistance recurrence in epithelial ovarian cancer.
• To identify key clinical and laboratory variables associated with platinum resistance recurrence.
• To compare the performance of machine learning models against traditional logistic regression.

Main Methods:

• Retrospective cohort analysis of 1,392 EOC patients treated with platinum-based chemotherapy.
• Variable selection using Lasso and multiple logistic regression.
• Development and comparison of five machine learning models (DTA, KNN, SVM, RF, XGBoost) against logistic regression.
• Internal validation using five-fold cross-validation and performance metrics (AUC, sensitivity, specificity, accuracy).

Main Results:

• Multiple logistic regression identified eight variables, while Lasso regression identified seven, associated with platinum resistance recurrence.
• The XGBoost model, using variables from multiple logistic regression, achieved the highest performance (AUC: 0.784, accuracy: 80.4%).
• A logistic regression model based on Lasso regression also showed good performance (AUC: 0.738, accuracy: 79.6%).

Conclusions:

• Successfully developed predictive models for platinum-resistant recurrence in EOC using routine clinical and laboratory data.
• The XGBoost model demonstrated superior predictive performance and is recommended for clinical use.
• Continuous model evolution is necessary to adapt to changing influencing factors over time.