Classification based on event in survival machine learning analysis of cardiovascular disease cohort

Area of Science:

• Cardiology
• Biostatistics
• Machine Learning

Background:

• Cardiovascular diseases pose a significant health burden.
• Predicting patient outcomes is crucial for effective treatment strategies.
• Survival analysis with a cured fraction presents unique modeling challenges.

Purpose of the Study:

• To evaluate supervised learning classification models for predicting outcomes in cardiovascular patients.
• To identify the most effective machine learning algorithm for survival analysis in this cohort.
• To assess the presence and impact of a cured fraction on prediction accuracy.

Main Methods:

• A cohort of 919 cardiovascular patients was analyzed over a maximum of 650 days.
• Survival analysis was performed, confirming a significant cured fraction (P < 0.01).
• Various machine learning algorithms, including Random Forest, SVM, logistic, and simple regression, were applied for patient status prediction (alive/dead).

Main Results:

• Random Forest demonstrated the highest overall predictive performance with an Area Under the ROC curve (AUC) of 0.934.
• Support Vector Machine (SVM) showed a lower False Positive Rate (0.263) for deceased patients.
• Logistic and simple regression models also yielded strong results with AUCs of 0.911 and 0.909, respectively.

Conclusions:

• Supervised learning models are effective for predicting outcomes in cardiovascular patients with a cured fraction.
• Random Forest is a promising method for overall survival prediction, though SVM excels in identifying deceased cases.
• Machine learning offers valuable tools for enhancing patient outcome prediction in clinical cardiology.