I discussed this topic in the XMLExchange Plus help file. (See topic Understanding XML Dictionaries -> Dictionaries Are NOT Needed to Run EDITS -> The Edit Engine Still Expects a Flat Buffer.)
Apart from the horror of all that would have to be redesigned in the Engine to accommodate this request, the multi-dimensional nature of XML would impact the logic of individual edits. For example, since Tumor data items exist in a many-to-one relationship to a Patient, the logic would have to be rewritten to loop through instances  to [n] of every Tumor item. This would ruin one of the major features of the EDITS software: its approachability for people who do not have programming backgrounds. Also, there are many thousands of legacy edits out in the field, still in use after decades, that somebody would have to rewrite.
(I'll attempt to attach a screen shot from the help file to this message; not sure if it will work.)
The Engine should not be concerned with the data source. You don't expect the Engine to know how to read and retrieve data directly from your database. The Engine's job is to give you the tools (APIs) you need to know where it can expect to find each data item in the record buffer; it's up to the calling software to prepare the data for processing accordingly.
Of course, I'm a member of the old guard, so there is no telling what the future may hold!