Coding for restricted cubic splines for multivariable logistic regression

[anuli98]

My analysis is determining the association between weight (continuous variable) and post-operative complications (yes/no). I am doing a multivariable logistic regression analysis, adjusting for seven predictors (1 has multiple categories, 5 are yes/no, and 1 is continuous). I notice that the association between weight and complication is not exactly linear, so it is recommended that I do restricted cubic splines. I was using the following code:

proc logistic data=weight.avg_pred_val_all_data2;

effect spl_weight = spline(weight/naturalcubic knotmethod=percentiles(5));/*natural cubic splines with knots at quartiles*/

model est_prob= spl_weight/ cl;

output out=predicted_values pred=pred;

run;

Where est_prob is the probability of any complication.I then plotted the unadjusted spline association between weight and probability of any complication:

ods graphics/height=6.5in width=10in imagename="Figure2_revised_with_cubicspline" border=off;

proc sgplot data=predicted_values noautolegend;

xaxis label='Weight (kg)';

yaxis label='Any complication' values=(0 to 60 by 5);

series x=weight y=pred/ lineattrs=(color=gray pattern=solid thickness=2);

run;

My question is, how do I show the association between weight and any complication, adjusted for my seven predictors?

[PGStats]

If your full model doesn't involve interactions, have a look at proc gam.

[anuli98]

Thanks I will try that

[Ksharp]

Better post it at Stat Forum , since it is a Statistical Problem. proc logistic has already built-in Graphics to display it .Check EFFECTPLOT statement : http://blogs.sas.com/content/iml/2016/03/21/statistical-analysis-stephen-curry-shooting.html

[anuli98]

Thanks for your suggestions

[sasbee]

Hi, Is there way to output the estimate estimate from the effectplot? For example, the outcome at weight = x.