topic Re: How to calculate the estimate effect (difference) and CI of continuous variables in Statistical Procedures

How to calculate the estimate effect (difference) and CI of continuous variables

Diels_O — Thu, 16 Jun 2022 15:47:44 GMT

Dear professors, I have a question of how to calculate the difference and CI of continuous variable in Clinical trails:

For example:

Outcomes Drug Placebo Effect Estimate(95%CI)
NIHSS score change, Mean(SD) 1.2(3.75) 0.8(2.01) 0.4(-0.4 to 1.1)

I want to get the difference like this

I use the PROC MEANS:

proc means data = work.stroke noprint alpha=0.05;
          class group;
          var NIHSS_24h_minus_baseline;                                
          output out=_runtemp_s1   n = n  mean = mean q1=q1 q3=q3 std=std median=median  nmiss=nmiss min=min max=max lclm=lclm_ uclm=uclm_ clm = clm_;
      run;

Can any expert tell me how to get the difference and CI or is there a macro to compute the estimate effect, risk difference, OR and CI? Thanks!

Re: How to calculate the estimate effect (difference) and CI of continuous variables

StatDave — Thu, 16 Jun 2022 13:45:37 GMT

The risk difference and odds ratio (OR) is appropriate for a binary response, not a continuous response. The output you included shows an estimate for the Drug group of 1.2 which is clearly not in the valid [0,1] range for the probability of a binary response. Clarify what your response is and what you want to estimate.

Re: How to calculate the estimate effect (difference) and CI of continuous variables

Diels_O — Thu, 16 Jun 2022 15:49:26 GMT

Thanks, Dave, I have made a change to the example, I want to calculate the difference of the continuous variable and the confidence interval.

Re: How to calculate the estimate effect (difference) and CI of continuous variables

StatDave — Thu, 16 Jun 2022 15:52:57 GMT

Then use PROC TTEST. See the Getting Started example in the TTEST documentation.

Re: How to calculate the estimate effect (difference) and CI of continuous variables

Diels_O — Thu, 16 Jun 2022 16:18:54 GMT

Thank you, and I want to ask whether PROC TTEST is appropriate for non-normal data and ordinal outcomes.

Re: How to calculate the estimate effect (difference) and CI of continuous variables

StatDave — Thu, 16 Jun 2022 16:22:59 GMT

TTEST might be adequate for a large data set, but you can always opt for a nonparametric test by using PROC NPAR1WAY instead.

Re: How to calculate the estimate effect (difference) and CI of continuous variables

Diels_O — Thu, 16 Jun 2022 16:42:09 GMT

Thanks, I have done Hodges-Lehmann Estimation to check the difference of ordinal outcomes, was it correct?

Re: How to calculate the estimate effect (difference) and CI of continuous variables

Ksharp — Fri, 17 Jun 2022 12:55:15 GMT

Check PROC FREQ.

proc freq.........
table x*y/ relrisk riskdiff ;
run;