Survival parametric modeling for patients with heart failure based on Kernel learning.

Area of Science:

• Medical Statistics
• Machine Learning in Healthcare
• Survival Analysis

Background:

• Time-to-event data are prevalent in medical research, necessitating advanced analytical methods.
• Traditional linear regression models often struggle with the complexity and volume of clinical datasets.
• Survival analysis techniques, like the Accelerated Failure Time (AFT) model, offer valuable alternatives to Proportional Hazards (PH) models.

Purpose of the Study:

• To propose a Multiple Kernel Learning (MKL) method for optimizing survival outcomes within the Accelerated Failure Time (AFT) framework.
• To develop a parametric regression framework integrating kernel learning with the AFT model for clinical data analysis.
• To compare the performance of the proposed MKL-AFT model against the Frailty model using established metrics.

Main Methods:

• A novel Multiple Kernel Learning (MKL) approach was developed to enhance the Accelerated Failure Time (AFT) model.
• The methodology involved integrating kernel learning with a parametric regression framework using gradient descent optimization.
• Four parametric models and 19 distinct kernels were evaluated, with MKL combining selected kernels for optimal performance.

Main Results:

• Kernelization was demonstrated to significantly enhance model performance in survival outcome prediction.
• The Multiple Kernel Learning (MKL) approach, by combining selected kernels, showed superior results compared to individual kernels or the baseline Frailty model.
• Performance was assessed using the Concordance index (C-index) and Brier score (B-score) on both case study and independent datasets.

Conclusions:

• The proposed Multiple Kernel Learning (MKL) method effectively optimizes survival outcomes within the Accelerated Failure Time (AFT) model.
• Kernelization, particularly through MKL, offers a powerful strategy for improving the accuracy and robustness of survival analysis in medical applications.
• The findings suggest MKL is a valuable tool for handling complex medical time-to-event data, outperforming traditional methods.