I have been working on another API recently so I may be a bit dull but my first impression is that the path of least resistance is to
- pull your question(s) with answers and comments disregarding activity/creation filter (use max page size)
- compare last activity date of the question. this will include new answers and edits on both the question and answers - if bumped you have your answer - pass go
- compare comment count on the question and then each answer - if any have changed you have your answer - pass go
- an obvious but unlikely hole here is if a comment is deleted and a comment is added resulting in an unchanged comment count. If this is a concern, compare comment_id
In terms of efficiency, this entails 1 (large payload) request and single iterated nested control structure for up to 100 questions *. One advantage is that if you pull with body=true, you already have your updated information.
Not as efficient as a last_activity_date that included comment activity but you work with what you have.
* This will likely break down for abnormal posts that contain >100 answers or posts with >100 comments. The only way I have found to fetch these is to perform distinct multipage pulls for answers and comments for each question, to which you would then apply the above criteria.