Alternate to chi square for cases with multiple values

[heretolearn]

Hello all,

I have a question about choosing the correct statistical test for my analysis.

I am trying to compare treatment characteristics between two groups. While normally I would use a chi square test for that analysis, some of the individuals in each group received more than one treatment type. Below I have included an example:

id     treatment     group

1      A                  positive

1      B                  positive

2      A                  negative

3      C                  positive

4      B                  negative

4      C                  negative

Both id 1 and 4 received more than one treatment type. So I am not sure if I can do chi square analysis comparing treatment between the two groups when some individuals have more than one treatment value.

Any suggestions for the proper test that I should use?

[Ksharp]

It is longitude data ,Why not using PROC GLIMMIX ?

[PaigeMiller]

What hypothesis do you want to test?

[StatDave]

It is not clear from your description what is to be compared with respect to what. From the data sample, I have to guess that the positive/negative variable, GROUP, is the response and that you want to see if the treatments differ on the probability of being positive (or equivalently, of being negative). If so, then you can compare the treatments and take into account the correlation among repeated measurements on a person (ID) by fitting either a Generalized Estimating Equations (GEE) model using PROC GEE or GENMOD, or a random effects model using PROC GLIMMIX, or a conditional logistic model using PROC LOGISTIC. For example, the following statements fit a GEE model to compare the treatments on the binary response:

proc gee;
class ID treatment;
model group(event="positive") = treatment / dist=bin;
repeated subject=ID / type=exch;
run;

[SteveDenham]

Is the order of treatments fixed? For instance, if a patient starts on A and is negative, is the next treatment always B?  Does the same apply if B is negative, would the next treatment be C?  If this is the case, you have a repeated measures design where you likely have structural zeroes (for instance, patient 4, which started with B, so that there is structural zero for that patient for treatment A).  If this all describes what you have, look at the first few examples in the PROC GEE documentation, and see whether the alternating logistic regression method might work for you.

SteveDenham