10 Accelerated Failure Time Models
The Cox model describes how covariates multiply the hazard. An accelerated failure time (AFT) model describes instead how they stretch or shrink time itself: a covariate that doubles the time scale makes every stage of the process, the median, the upper quartile, the time by which \(90\%\) have failed, twice as long. The model is fully parametric, so it gives survival curves, medians and predictions beyond the data directly, and its coefficients have a reading, “this treatment extends survival by \(40\%\)”, that many find more natural than a ratio of hazards. It is also one of the remedies, noted in the previous section, when hazards are not proportional.
10.1 The model
10.2 The Weibull model: both AFT and proportional hazards
10.3 Log-normal and log-logistic models
10.4 Checking and choosing a model
10.5 AFT models in software
10.6 Practice Problems
10.2 The Weibull model: both AFT and proportional hazards
10.3 Log-normal and log-logistic models
10.4 Checking and choosing a model
10.5 AFT models in software
10.6 Practice Problems
Questions on this section
Stuck on something here? Ask below and it stays attached to this topic.