Textract Table Odd Behavior on 1 out of 5 pages

I'm using Textract to parse table data in PDFs. Most of the pages are parsed out correctly:

But the last page table has the columns split out in strange ways:

I've got a config-driven process that relies on the index of a column to determine where to get the needed values. It's causing problems when the index for a value in one part of the table all of a sudden changes. Any ideas on how to get around this? I haven't seen this happen previously and it's kind of a bummer.