Multi-parameter regression survival modeling: An alternative to proportional hazards

Area of Science:

• Statistics
• Biostatistics
• Survival Analysis

Background:

• Standard parametric models typically incorporate covariates through a single distributional parameter.
• The proportional hazards model is a common example, assuming constant hazard ratios over time.

Purpose of the Study:

• To introduce and explore multi-parameter regression (MPR) modeling in survival analysis.
• To develop more flexible models offering greater insight into data generation.
• To address limitations of the proportional hazards assumption.

Main Methods:

• Developed a multi-parameter regression framework allowing covariates to affect multiple distributional parameters (e.g., scale and shape).
• Applied the approach to the two-parameter Weibull model to investigate time-dependent hazard ratios.
• Introduced a novel variable selection strategy accounting for correlated regression coefficients.

Main Results:

• Multi-parameter regression models demonstrated increased flexibility compared to standard models.
• The two-parameter Weibull model yielded time-dependent hazard ratios, relaxing the proportional hazards assumption.
• A new variable selection method was proposed for MPR models, considering coefficient correlations.

Conclusions:

• Multi-parameter regression offers a more nuanced understanding of survival data.
• The proposed methods, including a new variable selection strategy, are implemented in the R package 'mpr'.