A framework for leveraging machine learning tools to estimate personalized survival curves.

Area of Science:

• Biostatistics
• Survival Analysis
• Machine Learning

Background:

• Estimating conditional survival functions is crucial in time-to-event analysis, often complicated by censoring and truncation.
• Existing methods like Cox models and machine learning approaches have limitations, including focus on risk stratification or complex weighting requirements.

Purpose of the Study:

• To develop a new method for estimating conditional survival functions that bypasses the complexities of censoring and truncation.
• To enable the use of standard regression and classification techniques for survival data analysis.

Main Methods:

• Decomposition of the conditional survival function into observable regression models.
• Application of flexible regression and classification methods to these models.
• Empirical assessment of the proposed estimation procedures.

Main Results:

• The proposed decomposition effectively handles censoring and truncation without specialized survival data methods.
• Demonstrated applicability using data from an HIV vaccine trial.

Conclusions:

• This novel approach simplifies conditional survival function estimation by utilizing standard regression models.
• Offers a flexible and accessible alternative for analyzing time-to-event data in various scientific fields.