SAS code or macro for creating an exposure-response spline plot by Hazard Ratio

[nexterd]

Hello.

I am trying to create an exposure-response (or dose-response) plot by Hazard Ratio, which was calculated by time-dependent Cox proportional regression. An example I am attaching with is from Tomczak et al. (2016), but it was constructed by R.

So I want to find out how to draw an exact or similar plot in SAS. I've already tried some methods including the way LinusS posted in https://communities.sas.com/t5/SAS-in-Health-Care-Related/How-do-I-plot-the-spline-effect-in-PHREG/t...

But it draws something that needs to fix i.e., zeroing the CI at the reference value (in this case, it's the median) like below.

I think it explains about rate ratio/rate ratio (of the reference) not just rate ratio. Also, since it was set as using type=log in PROC SGPLOT, the intervals of y-axis are all different. Is it okay if I change it's type to linear intead of log? In my dataset, all individuals have been assigned each value of exposure and survival time to estimate HRs. And I am using SAS EG 7.1. Thank you for your help in advance.

[Rick_SAS]

Which of the following are you asking:

1. How to compute the points and CI in SAS

or

2. You have the points and CI, and you want to know how to plot them by using PROC SGPLOT. 

[nexterd]

I’ve already calculated one HR and one CI (not at each point of x-axis). But if I have to yield every point and CI before plotting, I want to know how to calculate them either. Thanks.

[mleslie]

Hi Rick, 

I have a related question regarding the computation of CI's for splines in PHREG. 

I have been successfully using the %est macro described by LinusS_ in this post (https://communities.sas.com/t5/SAS-Health-and-Life-Sciences/How-do-I-plot-the-spline-effect-in-PHREG... ) 

However, similar to the SAS graph above in this post, the confidence interval is essentially zero at the reference point, whereas similar code I have tried in R produces a continuous CI that does not approach zero width at the reference point. 

I am having difficulty finding statistical literature related to the validity of the CI estimation in SAS as opposed to what is produced in R and was wondering if you would be able to provide insight into these different methods of CI computation and whether there is a way to compute CIs in SAS that do not approach zero at the reference point for the spline. 

Thank you for your consideration and help!

[Rick_SAS]

I have no idea whether your CIs are valid or not. I suggest you post your data and SAS code if you want the community to provide feedback.

[mleslie]

Thank you for your feedback; I will create a new post with example data and code regarding my question.