Comparing rate ratios

[BTAinRVA]

I need help determining the correct way to compare changes in rates of two medical procedures between two groups before and after an intervention. So the data look something like this:

                         Before                    After

                   Proc1    Proc2        Proc1    Proc2

Group #1     21609    51815      4793       7191

Group #2     11935    67953      3808       10829

The question I want to answer is this: Is there a difference between the two groups in the change of the ratio of Proc1 to Proc2 before and after the intervention?

I was thinking of something like this (code adapted from an on-line example):

data YesNo;

length Group $ 11 ;

input Group $ Period $ Proc1 Total;

Response="Proc1"; Count=Proc1 ; output;

Response="Proc2"; Count=Total-Proc1 ; output;

datalines;

Group1 Before 21069 72884

Group1 After 4793 11984

Group2 Before 11935 79528

Group2 After 3808 14637

;

proc print noobs;

var Group Period Response Count;

run;

ods graphics on;

proc freq data =YesNo order=data;

weight Count;

tables Period*Group*response / relrisk cmh chisq plots=all;

exact eqor;

run;

ods graphics off;

Is this the correct approach? I'm not sure how to interpret the output. Any help would be greatly appreciated!

Brian

[BTAinRVA]

Does anyone know what eaxctly the Breslow-Day statistic is telling me when you run the code I posted? Is that just an overall comparison of the two groups or does it compare the before/after rates for each group?

Thanks in advance!

[drb4511 · Accepted solution]

The Breslow-Day (from the CMH option) tells you that the rate ratios for your strata are the same (p>.05) or different (p<.05). It is only comparing the rate ratios (or odds ratios) for group*response for each level of period.

If you are trying to see within group differences (ex: is there a significant change in the rate for group 1), you'd need to do something different, maybe even just plotting the rates with a linear reg.

[BTAinRVA]

drb4511

Thanks for the reply. I think that answers my question.

Brian