A widely
used regression model for survival
analysis is the Cox
proportional hazard model. This model is semi-parametric;
the effects of predictors are assumed to be proportional, i.e.
constant, in time, while the baseline hazard, i.e. the risk
for a reference group, is non-parametric. Parametric
survival models assume proportional
hazards, but also use one or more parameters
for the baseline hazard (e.g. 1 parameter for the exponential
or Poisson model, 2 parameters for the Weibull model or the
Gompertz model). Another model is the log normal model, which
assumes that effects of predictors diminish during follow-up.
We illustrate
the use of survival models in Example
2: HELP Survival Model.