Parametric proportional hazards and accelerated failure time models
Abstract
A unified implementation of parametric proportional hazards (PH) and accelerated failure time (AFT) models for right-censored or interval-censored and left-truncated data is described. The description here is vaĺid for time-constant covariates, but the necessary modifications for handling time-varying covariates are implemented in eha. Note that only piecewise constant time variation is handled.
1 Introduction
There is a need for software for analyzing parametric proportional hazards (PH) and accelerated failure time (AFT) data, that are right or interval censored and left truncated.
2 The proportional hazards model
We define proportional hazards models in terms of an expansion of a given survivor function ,
| (1) |
where is a parameter vector used in modeling the baseline distribution, is the vector of regression parameters, and is a positive function, which helps defining a parametric family of baseline survivor functions through
| (2) |
With and defined as the density and hazard functions corresponding to , respectively, the density function corresponding to is
where
Correspondingly, the hazard function is
| (3) |
So, the proportional hazards model is
| (4) |
corresponding to (1).
2.1 Data and the likelihood function
Given left truncated and right or interval censored data , and the model (4), the likelihood function becomes
| (5) |
Here, for , are the left truncation and exit intervals, respectively, means that and right censoring at , means that and an event at , and means that and an event occurs in the interval (interval censoring) and is a vector of explanatory variables with corresponding parameter vector , .
From (5) we now get the log likelihood and the score vector in a straightforward manner.
| (6) |
and (in the following we drop the long argument list to ), for the regression parameters ,
| (7) |
and for the “baseline” parameters , in vector form,
| (8) |
For estimating standard errors, the observed information (the negative of the hessian) is useful. However, instead of the error-prone and tedious work of calculating analytic second-order derivatives, we will rely on approximations by numerical differentiation.
3 The shape–scale families
From (1) we get a shape–scale family of distributions by choosing and
However, for reasons of efficient numerical optimization and normality of parameter estimates, we use the parametrisation and , thus redefining to
| (11) |
For the calculation of the score and hessian of the log likelihood function, we need some partial derivatives of . They are found in an appendix.
3.1 The Weibull family of distributions
The Weibull family of distributions is obtained by
leading to
and
We need some first and second order derivatives of and , and they are particularly simple in this case, for they are both zero, and for we get
3.2 The EV family of distributions
The EV (Extreme Value) family of distributions is obtained by setting
leading to
The rest of the necessary functions are easily derived from this.
3.3 The Gompertz family of distributions
The Gompertz family of distributions is given by
This family is not directly possible to generate from the described shape-scale models, so it is treated separately by direct maximum likelihood.
3.4 Other families of distributions
Included in the eha package are the lognormal and the loglogistic distributions as well.
4 The accelerated failure time model
In the description of this family of models, we generate a scape-scale family of distributions as defined by the equations (2) and (11). We get
| (12) |
Given a vector of explanatory variables and a vector of corresponding regression coefficients , the AFT regression model is defined by
| (13) |
So, by defining , we are back in the framework of Section 2. We get
and
| (14) |
defining the AFT model generated by the survivor function and corresponding density and hazard .
4.1 Data and the likelihood function
Given left truncated and right or interval censored data , and the model (14), the likelihood function becomes
| (15) |
Here, for , are the left truncation and exit intervals, respectively, means that and right censoring at , means that and an event at , and means that and an event uccurs in the interval (interval censoring), and is a vector of explanatory variables with corresponding parameter vector , .
From (15) we now get the log likelihood and the score vector in a straightforward manner.
and (in the following we drop the long argument list to ), for the regression parameters ,
and
For estimating standard errors, the observed information (the negative of the hessian) is useful, so
and
and finally
The second order partial derivatives and are
| (16) |
and from (10),
| (17) |
Appendix A Some partial derivatives
The formulas will be easier to read if we remove all function arguments, i.e., :