Predicting survival outcomes in advanced pancreatic cancer using machine learning methods

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Pancreatic cancer (PC) has a poor prognosis with a 5-year survival rate around 10%.
• Traditional prognostic methods may not capture complex patterns in patient data.
• Machine learning (ML) offers advanced analytical capabilities for prognostic assessments.

Purpose of the Study:

• To identify prognostic factors influencing overall survival in advanced-stage PC using ML.
• To evaluate the predictive performance of various ML algorithms for patient survival outcomes.

Main Methods:

• Utilized MATLAB for feature selection on data from 315 patients with advanced PC (2005-2023).
• Evaluated 19 clinical/laboratory features using minimum redundancy-maximum relevance, chi-square, ANOVA, and Kruskal-Wallis tests.
• Assessed 24 ML methods, including the support vector machine (SVM) kernel, for survival prediction.

Main Results:

• First-line chemotherapy emerged as the most significant predictor of survival (highest F-score).
• The SVM kernel method achieved 87% accuracy in predicting patient survival.
• Combining feature selection with the SVM kernel method improved prediction accuracy to 87.9%.

Conclusions:

• The SVM kernel method shows significant potential for predicting survival in advanced PC patients.
• Integration of feature selection techniques enhances the accuracy of ML-based survival predictions.
• Findings highlight the importance of first-line chemotherapy and the potential of ML to improve clinical decision-making and patient care.