How to calculate (average) hazard ratios for parametric survival models based on the R/flexsurv package

statistics

This tutorial [PDF] aims to support the interpretation of parametric time-to-event models by explaining how to calculate the hazard ratio, which is a conventional effect size to evaluate clinical relevance of treatment effects.

The following distributions, which are considered most important for health economic analyses, are implemented in the flexsurv package (cf ?flexsurvreg):

  1. exp: exponential (PH model)
  2. weibull: Weibull (AFT model/PH model)
  3. gompertz: Gompertz (PH model)
  4. llogis: log-logistic (AFT model)
  5. lnorm: lognormal (AFT model)
  6. gengamma: Generalized gamma (AFT model)

A single, constant hazard ratio can only be calculated for distributions, for which the hazard ratio does not depend on time and assumes proportional hazards. For the exponential, Weibull (if PH parametrisation is used), and Gompertz distribution, a single, constant hazard ratio can be computed, but not for the remaining accelerated failure time (AFT) models.

The code is available as BitBucket GIT repository [CODE].