Hazard-based distributional regression via ordinary differential equations.

Area of Science:

• Biostatistics
• Mathematical Biology
• Computational Statistics

Background:

• Survival regression models like proportional hazards and accelerated failure time models are widely used.
• These models often rely on a shared baseline hazard, limiting their ability to capture diverse hazard shapes when parametrically specified.

Purpose of the Study:

• To propose a general class of parametric survival regression models that overcome the limitations of shared baseline hazards.
• To incorporate covariate information into survival models using autonomous systems of ordinary differential equations (ODEs).
• To provide deeper insights into how covariates influence survival dynamics and intervention efficacy.

Main Methods:

• Modeling the hazard function using autonomous systems of ordinary differential equations (ODEs).
• Incorporating covariate information via transformed linear predictors on ODE system parameters.
• Developing efficient Bayesian computational tools, including parallelized log-posterior evaluation and Markov Chain Monte Carlo (MCMC) samplers.
• Deriving conditions for posterior asymptotic normality for fast posterior approximations.

Main Results:

• The proposed framework allows for the identification of covariate values that produce qualitatively distinct hazard shapes linked to different ODE system attractors.
• Demonstrated the methodology using clinical trial data with crossing survival curves.
• Applied the approach to cancer recurrence times, revealing how patient characteristics influence treatment efficacy on hazard and survival.

Conclusions:

• The ODE-based survival regression framework offers a flexible and interpretable alternative to traditional models.
• This methodology provides enhanced understanding of covariate effects on survival dynamics and intervention outcomes.
• The developed computational tools facilitate efficient model fitting and analysis.