Assymetrical confidence intervals in PHREG, how possible?

[Marjolein]

Hi all,

This afternoon I decided to plot my hazard ratio's and confidence intervals in a forest plot (horizontal bars). However, when I plot my effect sizes and confidence intervals, they are not symmmetrical (e.g. the length from the effect size to the upper and lower bound are not equal....). I have no clue how this could have happened.

I obtained by data with the PHREG procedure. The piece of syntax I used looks basically like this:

PROC PHREG data=mydata;

CLASS var1 var2 var3/ref=first;

MODEL survival*censor(0) = var1 var2 var3 / RL;

RUN;

Really hope someone can help me out!

Kind regards,

Marjolein

[Reeza · Accepted solution]

I believe this is correct. Your likely plotting on a linear scale, while the model developed is an exponential model. 

Answe is similar to this :

http://stats.stackexchange.com/questions/4713/binomial-confidence-interval-estimation-why-is-it-not-...

[Marjolein]

Ah! That makes sense!!! Thank you so much. This is quite a relief 🙂

[Rick_SAS]

By default the oddsratio plot is on a linear scale. You can use the LOGBASE=10 option on the PLOTS=ODDSRATIO option to generate the oddsratio plot on a log scale, which I think is more natural. See the discussion and example in the article "Odds ratio plots with a logarithmic scale in SAS."