topic Re: SAS equivalent of PSPLINE in Statistical Procedures

SAS equivalent of PSPLINE

PamG — Fri, 30 Jun 2023 15:09:19 GMT

I am given a code in R and have to translate to SAS. Is there a SAS equivalent of the R-PSPLINE function in SAS? I could not find any relevant resources online. Basically I need to translate the following in SAS,

coxph(Surv(time,death)~sex+pspline(age,df=4),data=have)

Re: SAS equivalent of PSPLINE

Reeza — Fri, 30 Jun 2023 15:19:59 GMT

FYI I've moved your post to the stats forum. Check out PROC PHREG with the EFFECT & SPLINE options.
https://communities.sas.com/t5/Statistical-Procedures/How-do-I-plot-a-spline-in-a-Cox-Regression-model-PROC-PHREG/td-p/827279

I don't see a penalized option though so may need to use TPSPLINE first?

Hopefully someone else has a better answer 🙂

Re: SAS equivalent of PSPLINE

PamG — Fri, 30 Jun 2023 16:00:47 GMT

Thanks Reeza.

TPSPLINE will not work as I need to use this on survival data. The only way around I see is to create the variables in data step and then use it in MODEL statement in PROC SURVIVAL. I was hoping there is an easy way to do it.

Re: SAS equivalent of PSPLINE

PamG — Fri, 30 Jun 2023 16:01:56 GMT

I meant PROC PHREG.

Re: SAS equivalent of PSPLINE

Reeza — Fri, 30 Jun 2023 16:23:40 GMT

May not understand the problem correctly, but I was assuming using TPSPLINE to create the variables and then using them in PROC PHREG.

Re: SAS equivalent of PSPLINE

Tom — Fri, 30 Jun 2023 16:30:05 GMT

How is what the R package does different than the Penalized B-splines of proc transreg?

https://documentation.sas.com/doc/en/statcdc/14.2/statug/statug_transreg_details07.htm

Re: SAS equivalent of PSPLINE

Ksharp — Sat, 01 Jul 2023 09:31:54 GMT

Maybe @Rick_SAS could give you a hand .

Re: SAS equivalent of PSPLINE

PamG — Sat, 01 Jul 2023 14:55:52 GMT

Upon reading further it seems like it is not implemented in SAS. Not sure why given it is used so often.

https://bmcmedresmethodol.biomedcentral.com/articles/10.1186/s12874-019-0666-3

Re: SAS equivalent of PSPLINE

Ksharp — Sun, 02 Jul 2023 09:20:00 GMT

Also calling @StatDave @lvm @SteveDenham

Re: SAS equivalent of PSPLINE

Ksharp — Sun, 02 Jul 2023 09:31:39 GMT

There is a function MSPLINT() in data step, you refer to .

https://communities.sas.com/t5/Graphics-Programming/Obtaining-values-resulting-from-smoothconnect/m-p/853806

Re: SAS equivalent of PSPLINE

Ksharp — Sun, 02 Jul 2023 09:36:42 GMT

https://communities.sas.com/t5/Statistical-Procedures/CUBIC-SPLINE-YIELD-CURVES/m-p/845315

Re: SAS equivalent of PSPLINE

Rick_SAS — Mon, 03 Jul 2023 10:48:49 GMT

You should think about what the R model is trying to do. It is likely that author of the model chose penalized B splines simply to capture the potential that the survival rate has a potentially complicated relationship with the AGE variable. The specific form of the spline probably doesn't matter. In fact, if it is a good model, it should be robust to changes in the spline family. SAS supports B-splines, PB-splines, thin-plate splines, cubic splines, adaptive regression splines, and more, but in my experience, the form of the splines often has little impact on the predictive model.

For an example of two models that illustrate my points, see "Nonparametric regression for binary response data in SAS.' The example shows two completely different (nonparametric) models that give roughly the same predictive model. If they didn't give similar predictions, then one or both models are probably not good.

I encourage you to use the EFFECT statement in PROC PHREG to construct a model that uses either the BSPLINE or TPF spline basis. Compare the models' predictions to the R model. In most applications, the goal is to find a good model, not to worry about technical differences between different spline models.