This probably depends on the version of DITA OT.
I am using 1.5.4.
I think the index output is actually processed in dost.jar.
I had to change it, too.
I processed a list of topics based on, IIRC, the fullditatopiclist in temp/dita.list
I parsed the topics looking for keywords in the prolog that had an indexterm in them.
I generated output into a temp variable, with a copy to disk just for debugging, with the info I needed to create the index.xml file.
Then I processed that variable to get the summary data I needed in the index.xml file.
I had to override the Ant as well.
The target I needed to overwrite was dita.out.map.eclipse.index.