30
$\begingroup$

Our small team was having a discussion and got stuck. Does anyone know whether Cox regression has an underlying Poisson distribution. We had a debate that maybe Cox regression with constant time at risk will have similarities with Poisson regression with a robust variance. Any ideas?

$\endgroup$
32
$\begingroup$

Yes, there is a link between these two regression models. Here is an illustration:

Suppose the baseline hazard is constant over time: $h_{0}(t) = \lambda$. In that case, the survival function is

$S(t) = \exp\left(-\int_{0}^{t} \lambda du\right) = \exp(-\lambda t)$

and the density function is

$f(t) = h(t) S(t) = \lambda \exp(-\lambda t)$

This is the pdf of an exponential random variable with expectation $\lambda^{-1}$.

Such a configuration yields the following parametric Cox model (with obvious notations):

$h_{i}(t) = \lambda \exp(x'_{i} \beta)$

In the parametric setting the parameters are estimated using the classical likelihood method. The log-likelihood is given by

$l = \sum_{i} \left\{ d_{i}\log(h_{i}(t_{i})) - t_{i} h_{i}(t_{i}) \right\}$

where $d_{i}$ is the event indicator.

Up to an additive constant, this is nothing but the same expression as the log-likelihood of the $d_{i}$'s seen as realizations of a Poisson variable with mean $\mu_{i} = t_{i}h_{i}(t)$.

As a consequence, one can obtain estimates using the following Poisson model:

$\log(\mu_{i}) = \log(t_{i}) + \beta_0 + x_{i}'\beta$

where $\beta_0 = \log(\lambda)$.

$\endgroup$
  • 10
    $\begingroup$ More generally, assuming constant hazard rates over fixed time intervals (known as a piecewise-exponential model) you can fit fairly flexible survival models in the form of poisson GLMs - if you add interactions between the piecewise constant baseline hazard and covariates, you can estiamte time-varying effects and move away from the proportionality assumption, for example. Sources: Michael Friedman "Piecewise Exponential Models for Survival Data with Covariates", Annals of Statistics N LAIRD, D OLIVIER "Covariance Analysis of Censored Survival Data Using Log-Linear Analysis Techniques" JASA $\endgroup$ – fabians Mar 10 '11 at 10:18
  • $\begingroup$ and @fabians, Thank you. Seems like a more interesting thing to look at and generate more discussion from our group! $\endgroup$ – Julie Mar 10 '11 at 13:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.