Help using Base SAS procedures

Proc Genmod: how many levels?

Reply
N/A
Posts: 1

Proc Genmod: how many levels?

Hi all,

I want to analyse two things:

--> I. if the reporting of an item differs between different surgical specialisms - nationwide.

--> II. if the outcome of a surgery differs depending on the reported items.

Since my data is cohort, I would like to produce risk ratio's.

We have data from many hospitals, so I expect clustering within hospitals.

However for my first analysis I do also expect clustering within specialisms in the hospitals. While for my second analysis I expect clustering of specialisms across hospitals, since we know from history that the outcome from nature differs significantly between surgical specialisms.

My questions are:

1. Can proc genmod handle clustering on 2 levels? And if yes, did I programme this correctly?

2. Is my code for the first analysis (see below) correct now? Or do I overcorrect by using Specialisms as independant variables (dummy variables with SPEC1 as a reference) AND using them as a level as well?

3. Is my code for second analysis (see below, option A) correct now? Or should I introduce specialism as covariate in my analysis, like I specified in option B?

Can anyone help me out on this?

Thanks!

Code for first analysis:

proc genmod descending data= test3;

class specialism hospital;

model reporting= SPEC2 SPEC3 SPEC4 SPEC5/ dist=binomial link=log;

repeated subject = special(hospital) /type=exch;

estimate 'SPEC2 vs SPEC1' SPEC2 1;

estimate 'SPEC3 vs SPEC1' SPEC3 1;

estimate 'SPEC4 vs SPEC1' SPEC4 1;

estimate 'SPEC5 vs SPEC1' SPEC5 1;

run;

Code for second analysis, option A:

proc genmod data= test3;

class specialism hospital;

model Outcome= reporting covariate1 / dist=binomial link=log;

repeated subject = hospital(special) /type=exch;

estimate 'reported yes no' reporting 1;

estimate 'covariate yes no' covariate1 1;

run;

Code for second analysis, option B:

proc genmod data= test3;

class hospital;

model Outcome= reporting SPEC2 SPEC3 SPEC4 SPEC5 covariate1 / dist=binomial link=log;

repeated subject = hospital /type=exch;

estimate 'reported yes no' reporting 1;

estimate 'SPEC2 vs SPEC1' SPEC2 1;

estimate 'SPEC3 vs SPEC1' SPEC3 1;

estimate 'SPEC4 vs SPEC1' SPEC4 1;

estimate 'SPEC5 vs SPEC1' SPEC5 1;

estimate 'covariate yes no' covariate1 1;

run;

Ask a Question
Discussion stats
  • 0 replies
  • 186 views
  • 0 likes
  • 1 in conversation