0
$\begingroup$

I would like to fit a proportional hazards model with log normal baseline hazard in R. I have found several options for the semiparametric Cox proportional hazards, but I have not found a package to do parametric ph survival.

Is there a package to do this?

I have checked the packages: survival, flexsurv, and eha without sucess, but I may be missing something.

$\endgroup$
4
  • $\begingroup$ Have you looked at the survival::survreg function? $\endgroup$
    – DWin
    Commented May 10, 2020 at 16:09
  • $\begingroup$ @DWin As far as I understand "=survreg() fits accelerated failure models, not proportional hazards models. See: rstudio-pubs-static.s3.amazonaws.com/… $\endgroup$
    – logo
    Commented May 10, 2020 at 16:32
  • $\begingroup$ Some AFT models are also PH ones. The Weibull and Poisson models are both PH. I'm not sure that log normal ones can be PH. $\endgroup$
    – DWin
    Commented May 10, 2020 at 22:09
  • $\begingroup$ The Weibull (including exponential as a special case) is the only model that's both PH and AFT. $\endgroup$ Commented Feb 2, 2023 at 6:18

2 Answers 2

1
$\begingroup$

As the manual page for the phreg() function in the R eha package says:

the standard lognormal and loglogistic distributions are not closed under proportional hazards.

That is, if you try to impose a proportional hazards (PH) assumption of covariate effects on a lognormal baseline function, the resulting distribution need not be lognormal. The proposed workaround in that package is:

The lognormal and loglogistic baseline distributions are extended to a three-parameter family by adding a "proportionality" parameter (multiplying the baseline hazard function).

Nevertheless: "The lognormal and loglogistic distributions are included on an experimental basis for the moment. Use with care, results may be unreliable!"

The standard 2-parameter location-scale survival modeling with a lognormal baseline, as performed by the aftreg() function suggested in another answer, is not a proportional hazards (PH) model. It's an accelerated failure time (AFT) model. Only the Weibull family (with its special case of the exponential distribution) is closed under both PH and AFT models.

$\endgroup$
0
$\begingroup$

I just found out that it can be done using the aftreg command from the eha R package.

https://cran.r-project.org/web/packages/eha/eha.pdf

$\endgroup$
1
  • $\begingroup$ As the name aftreg suggests, this doesn't fit a proportional hazards model; it fits an accelerated failure model just as survival::survreg does. $\endgroup$ Commented Feb 2, 2023 at 6:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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