Chi-square for clustered data

[Melk]

I have some eye-level data and I just want to run a simple chi-square comparing the proportion of disease X between race. Since the data is clustered, is there an option I can use in proc freq to adjust for the correlation of eyes between patients? 

ID    EYE    X    RACE

1        R       0        A

1        L       0        A

2        R      1       W

2        L       0       W

[mkeintz]

I presume that "having the disease" is signified by both eyes having X=1, or either eye have X=1, correct?  If so, create one record per person, with a new variable NEWX, which takes the maximum value of X for each ID.  Then you can proc freq, crosstabulating newx with race.

data have;
  input ID    EYE :$1.     X    RACE :$1.;
datalines;
1        R       0        A
1        L       0        A
2        R      1       W
2        L       0       W
run;

data want (drop=eye);
  set have (where=(x=0))
      have (where=(x=1));
  by id;
  if last.id;
  rename x=new_;
run;

The data step above interleaves all the X=0 records with the X=1 records, within each by group.  So no matter the original order of x values within an id, the incoming data will present the X=1 record last, if there is an x=1 record.  Otherwise the last record for the id will be an x=0:

This program depends on the data being sorted by ID, but it doesn't care about the order of X within id, nor the order or R vs L within the id.

[Melk]

Actually each eye may or may not have the condition (disease was probably the wrong word, sorry). I am thinking it should be a GEE model for binary outcome to account for eye correlation.