[R] survival parametric question

[Frank E Harrell Jr]

Prof Brian Ripley wrote:
> The survival package is a recommended package in R and contains survreg() 
> which uses the AFT definitions for Weibull survival.  This is well 
> documented, and MASS (the book) has comparisons of PH and AFT 
> parametrization for a Weibull example.
> 
> I think you mean Frank Harrell's `Design' package.  As far as I am aware 
> that has a function psm() (not PSM) which is based on survreg(), so the 
> interpretation should be the same.

Yes, at least until you run a psm Weibull fit through the pphsm 
convertor function.  I recommend that Denis run the psm fit through 
Design's Hazard and Survival functions to create S functions containing 
the analytic representation of hazard and survival functions.  There's 
also Mean and Quantile, and latex.psm.

Frank

> 
> On Fri, 26 Aug 2005, denis lalountas wrote:
> 
> 
>>Hi to all,
>>I am working on design package using survival function.
>>First using PSM and adopting a weibull specification for the baseline hazard , I have got the following results(since weibull has both PH and AFT propreties ,in addition I have used the PPHSm command):
>>
>> Value Std. Error      z        p
>>
>>(Intercept)  1.768     1.0007   1.77 7.73e-02
>>
>>SIZE        -0.707     0.0895  -7.90 2.80e-15
>>
>>REtoTA      -0.896     0.4208  -2.13 3.33e-02
>>
>>D1toEQ       0.281     0.0330   8.51 1.81e-17
>>
>>EBTtoTA     -6.706     1.0807  -6.21 5.46e-10
>>
>>SALtoTA     -3.943     0.3575 -11.03 2.78e-28
>>
>>fishes       2.619     0.4194   6.24 4.26e-10
>>
>>computers    2.781     0.2105  13.21 7.35e-40
>>
>>Log(scale)  -0.945     0.1514  -6.24 4.25e-10
>>
>>and the loglikelihood -82.0
>>
>>I dont know the specification of the weibull that Desing package uses so 
>>I can't evaluate the result.
>>
>>For comparison reasons I have estimated the same model using another 
>>spftware EasyReg
>>
>>wich gave the following results( the weibull specification has the form 
>>a(1).a(2).t^(a(2)-1):
>>
>>parameters ML estimate t-value p-value Covariates
>>
>>beta(1)       2.411460   2.136 0.03265 fishes
>>
>>beta(2)       2.710115   3.322 0.00089 computers
>>
>>beta(3)      -7.539632  -2.646 0.00815 EBTtoTA
>>
>>beta(4)      -3.720231  -2.547 0.01086 SALtoTA
>>
>>beta(5)       0.262115   1.982 0.04751 D1toEQ
>>
>>beta(6)      -0.710535  -0.515 0.60684 REtoTA
>>
>>beta(7)      -0.493369  -1.938 0.05262 LOG(SIZE)
>>
>>alpha(1)      0.485828   0.392 0.69491
>>
>>alpha(2)      2.597073   5.516 0.00000
>>
>>log(L)=-83,4
>>
>>First observe that the results are almost the same but the weibull parameters are not.
>>
>>acooring to the weibull specification that easyreg uses a(2)>0 so the baseline hazard is monotonically increases ,acording to my expectations :(the empirical uncoditional hazard increases monotonically from t=1,12 and then decreases to zero)
>>
>>My question is what is the weibull specification that R-design package uses for the baseline hazard. Second ,it is possible to plot the baseline hazard in R , in order to "see" the accelerating-decelerating effect in the AFT case.
>>
>>In addition how can simulate a model in the AFT case ( some examples of simulation are given in the design manual for the COX-PH case.
>>
>>I hope that my questions are not borring, if so sory I am a new user of R package.
>>
>>Best regards
>>
>>D.Lalountas
>>
>>University of Patras , Greece
>>
>>
>>
>>---------------------------------
>>
>>
>>	[[alternative HTML version deleted]]
>>
>>______________________________________________
>>R-help at stat.math.ethz.ch mailing list
>>https://stat.ethz.ch/mailman/listinfo/r-help
>>PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
>>
> 
> 


-- 
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University