Accelerated failure time model

The Accelerated Failure Time (AFT) model is a parametric model used in statistics, particularly in the field of survival analysis. The AFT model is used to estimate the effect of covariates on the rate at which a specified event, such as failure or death, occurs. This model assumes that the effect of the covariates accelerates or decelerates the life time of an item or individual by some constant factor.

Overview

The AFT model can be expressed in the form:

\[ \log(T) = \mathbf{X}\beta + \sigma W \]

where:

• \( T \) is the failure time.
• \( \mathbf{X} \) represents the covariates (or independent variables) associated with each observation.
• \( \beta \) is a vector of coefficients that describe the influence of the covariates.
• \( \sigma \) is the scale parameter of the model.
• \( W \) is the error term, typically assumed to follow a standard distribution such as normal, logistic, or extreme value.

In this model, the logarithm of the survival time is linearly related to the covariates. The AFT model is different from the more commonly known Proportional hazards model where the hazard function is modeled, rather than the survival time itself.

Applications

The AFT model is widely used in various fields including medicine, engineering, and biology. In medical research, it is used to analyze the survival time of patients with respect to treatments, where the covariates might include treatment plans, dosage levels, and patient characteristics. In engineering, it can be used to predict the time until failure of components or systems, where covariates might include operating conditions and material properties.

Advantages

One of the main advantages of the AFT model is its interpretability. The coefficients in the AFT model directly indicate the effect of covariates on the survival time. A positive coefficient indicates that the covariate increases the survival time, while a negative coefficient indicates a decrease in survival time.

Estimation Techniques

Estimation of the parameters in an AFT model can be performed using various methods, including:

• Maximum likelihood estimation (MLE)
• Bayesian estimation
• Rank regression

These methods help in determining the best-fit parameters that describe the data, taking into account the specified distribution of the error term.

Challenges and Limitations

While the AFT model is useful, it has limitations. The assumption of a parametric form for the survival times can be restrictive. If the true underlying distribution of the survival times does not match the chosen distribution, the model may provide biased or inaccurate estimates. Additionally, the model's performance can be significantly affected by outliers or mis-measured covariates.

See Also

• Survival analysis
• Proportional hazards model
• Cox proportional hazards model
• Kaplan-Meier estimator