I used to have this problem too! But I found a way around it that seems to work for me!
I noticed that each chapter generally started out with [header] or [h] or something like that. When I do a conversion, I go to structure detection, and where it says "chapter mark" I set this as "none".
Going down to where it says "Insert Page Breaks Before (XPath expression):
I type in the following,
//h:header or
//h:h1. This seems to work consistently for me!
Hope this helps!