A boosting first-hitting-time model for survival analysis in high-dimensional settings

Area of Science:

• Biostatistics
• Computational Biology
• Statistical Modeling

Background:

• First hitting time (FHT) models offer a parametric alternative to Cox models for time-to-event data.
• FHT models do not require the proportional hazards assumption, which is difficult to verify in high-dimensional settings.
• Integrating low-dimensional clinical and high-dimensional molecular data in prediction models is challenging with current methods.

Purpose of the Study:

• To propose a novel boosting algorithm to enhance the applicability of FHT models in high-dimensional frameworks.
• To provide a flexible parametric alternative to the Cox model for time-to-event responses.
• To enable natural integration of multi-modal data (clinical and molecular) in prediction models.

Main Methods:

• Development of a boosting algorithm tailored for FHT models.
• Application of the algorithm to high-dimensional datasets.
• Stochastic process-based modeling approach.

Main Results:

• The proposed boosting algorithm successfully extends FHT models to high-dimensional data.
• The method naturally integrates low-dimensional clinical and high-dimensional molecular information.
• Performance was validated using three real-world data examples.

Conclusions:

• The novel boosting algorithm enhances FHT models for high-dimensional prediction.
• This approach offers a powerful, flexible alternative to existing time-to-event models.
• The method facilitates integrated analysis of diverse biomedical data types.