topic Splines in Statistical Procedures

Splines

Romain69100 — Mon, 17 Oct 2022 18:11:45 GMT

Hi sas users!

I am currently working on a study assessing the link between the Lymphocyte/Monocyte ratio and the progression free survival.

I would like to use this ratio as a spline in my model (Truncated power basis).

So far i did this:

proc phreg data = RELEVANCE_LMR;
effect slmr = spline(d_ratio_l_m_c / basis=tpf(noint) degree=3 knotmethod=EQUAL(1));
model D_TIME_PFS_IRC_EMA*D_CENS_PFS_IRC_EMA(1) = slmr/;
output out=mydatcub Xbeta=mylinpredcub;
run;

d_ratio_l_m_c represents my ratio which is centered on the median value (2.5). I would like to have only one knot corresponding to the median value (2.5), however i cannot how to retrieve the knot position after performing the "knotmethod=EQUAL(1)". Do you have any idea to find it?

Thanks in advance for your help

Re: Splines

sbxkoenk — Mon, 17 Oct 2022 18:23:11 GMT

Hello,

Put the details option in your spline specification, just like here :

proc orthoreg data=sashelp.gas;
   effect spl = spline(eqratio / knotmethod=equal(3) details);
   class fuel;
   model nox = spl | fuel;
   effectplot / obs extend=data;
   ods output SliceFitPlot=sp;
run;

That DETAILS option is documented here :

SAS/STAT 15.2 User's Guide

https://go.documentation.sas.com/doc/en/pgmsascdc/9.4_3.5/statug/statug_introcom_sect020.htm

DETAILS
requests tables that show the knot locations and the knots associated with each spline basis function.

Koen

Re: Splines

Romain69100 — Mon, 17 Oct 2022 18:36:21 GMT

Thank you for your answer!

Do you know how to force the knot position ?

Romain

Re: Splines

StatDave — Mon, 17 Oct 2022 19:01:58 GMT

Use KNOTMETHOD=LIST(list-of-values). But note that if you want to specify knots at percentiles, like the median, you can use KNOTMETHOD=PERCENTILELIST(list-of-percentiles). See all of the options in the EFFECT statement documentation for splines.

Re: Splines

Romain69100 — Mon, 17 Oct 2022 19:14:46 GMT

Thank you for your answer.

I just want to specify one knot which will be located at the median value of the Lymphocyte/Monocyte ratio. This value eq 2.5.

In my model, as i have centered my Ratio on the median value, i want my knot to be equal to 0. How can i specify this in the following statement?

effect slmr = spline(d_ratio_l_m_c / basis=tpf(noint) degree=3 knotmethod=EQUAL(1))

Re: Splines

StatDave — Mon, 17 Oct 2022 19:27:06 GMT

Use the LIST() option as I mentioned. Replace EQUAL(1) with LIST(0) in KNOTMETHOD=.

Re: Splines

Romain69100 — Mon, 17 Oct 2022 19:42:11 GMT

Tanks i tried your solution. Unfortunately i do not have an estimation for the last coefficient. Please find a screenshot below.

Analysis of Maximum Likelihood Estimates
Parameter / / DF / Parameter Estimate / Standard Error / Chi-Square / Pr > ChiSq / Hazard Ratio / Label
slmr / 1 / 1 / -0.01131 / 0.02359 / 0.2299 / 0.6316 /. / slmr 1
slmr / 2 / 1 / 0.00113 / 0.0009973 / 1.2939 0.2553 / . / slmr 2
slmr / 3 / 1 / -0.0000127 / 9.49004E-6 1.7897 / 0.1810 / . / slmr 3
slmr / 4 / 0 / 0/ . / . / . / . /slmr 4

Re: Splines

StatDave — Mon, 17 Oct 2022 19:53:38 GMT

Generally speaking, any time a parameter is shown with zero DF it is due to that design column being collinear with some combination of the others. This typically necessitates simplifying your model. In your case, maybe a lower degree or a different basis, etc. As always, check your log to see if there are any helpful messages.

Re: Splines

Romain69100 — Mon, 17 Oct 2022 19:57:19 GMT

Strangly with R software i was able to get estimate with this cubic spline..

Re: Splines

StatDave — Mon, 17 Oct 2022 20:08:22 GMT

You would have to verify that the design variables created by your R code are exactly the same and that the model definition is the same to be sure of that conclusion. There could easily be some difference in the respective details.