Hazard rates to Markov simulation - what are the best procedures to use from data to simulation?

[VivianKvist]

We have censored data for survival of forest stands, with information on age class (discrete 5 year intervals- 20 classes), species type (3 classes) and location (2 classes). We are updating previous analyses.

Example of data can be accessed here

https://erda.ku.dk/public/archives/05474701b063e9937b244a5af6d51b35/published-archive.html 

I have used different procedures in SAS to estimate hazard rates implemented in a subsequent Markov simulation of forest populations.

Proc PHREG provides some good options, but I have uncertainty in how to apply the output from the PHREG to the subsequent Markov simulation (deterministic coded in SAS), where the hazard rate for each combination of classes are needed.

Fx

data - inrisks give sample of classes, dk_tota are single observations of starting ageclass (acl) and related information on class variabeles and information whether an event has occoured or not (cut =0 or 1)

proc phreg data=dk_tota plots(overlay)=survival;
class spec_type joe07;
model acl*cut(0)=joe07 spec_type / ;
baseline covariates=Inrisks out=Pred_BL2 survival=_all_/rowid=Id ;
run;

I have not found clear indications in the syntax or the examples for this use of PHREG. Maybe other procedures are more suitable.

Any good hints are appreciated.

[sbxkoenk]

Hello @VivianKvist ,

For your simulation approach, I think you need to look at the MCMC procedure (SAS/Stat).

Do you know that MCMC (Markov Chain Monte Carlo) procedure?

I do not know it very well, otherwise I would provide you with a quick-written example.

I would give you the link to the MCMC doc, but the server is temporarily unable to service my | your request due to maintenance downtime.

Good luck,

Koen

[VivianKvist]

Thanks Koen, I will look into the MCMC procedure.

[WeiChen]

Do you mean that you get estimates fropm PHREG but you don't know how to use the estimates to simulate data from the PH model?  If so, post the estimates.

[VivianKvist]

I do get estimates from PHREG and a number of other procedures (ICPHREG, LIFETEST, LIFEREG etc). It's more to get feedback on the links between hazard rates and survival functions, and a clearer indication of whether they are given as annual rates or interval rates. Thanks.

[WeiChen]

The documentation for PROC LIFEREG says the model behind many survival functions. It discuss the CLASS variable. FOr PHREG, see the "Details" section, especially about hazard ratios.