A flexible parametric accelerated failure time model and the extension to time-dependent acceleration factors

Area of Science:

• Biostatistics
• Survival Analysis
• Medical Statistics

Background:

• Accelerated failure time (AFT) models are valuable in medical research but standard parametric forms have limitations in capturing complex event time distributions.
• Proportional hazards models are more commonly used, leaving potential for broader AFT model application.
• Existing parametric AFT models often restrict the shapes of survival curves they can represent.

Purpose of the Study:

• To propose a general parametric accelerated failure time (AFT) model with enhanced flexibility for survival data analysis.
• To extend AFT models to incorporate time-dependent acceleration factors and handle delayed entry.
• To demonstrate the utility of the proposed flexible AFT model for causal inference and provide user-friendly software.

Main Methods:

• Developed a general parametric AFT model utilizing restricted cubic splines for flexible baseline hazard modeling.
• Incorporated methods to handle time-dependent acceleration factors and delayed entry, allowing for time-dependent covariates.
• Evaluated model performance through simulations and applied it to a breast cancer patient dataset.

Main Results:

• Simulations demonstrated substantial improvements of the proposed flexible AFT model over standard parametric AFT models.
• Analytical and simulation results confirmed the collapsibility of the proposed AFT models, supporting their use in causal inference.
• The study provides efficient Stata and R software packages for implementing the proposed methods.

Conclusions:

• The proposed flexible parametric AFT model offers significant advantages over traditional AFT models for survival data.
• The collapsibility of these models makes them well-suited for causal inference applications in medical research.
• The availability of software packages facilitates the adoption and application of these advanced AFT models in practice.