I recently went on a journey with respect to this. Devonthink, as someone else mentioned, is great for aggregating documents; I really wanted it to work for my QDA needs but it was much better for collection & storage than the more granular analysis phase.
Dedoose seems good, but after a friend of mine had catastrophic data loss with them in years past I have been pretty suspicious of it; I know they have beefed up their infrastructure but it makes me wary. With the other big three, I think it mostly comes down to preference, norms in your discipline, and price (they’re all expensive, but depending on which license you get there may be significant differences). I suspect depending on your needs (collaboration, for example, means you’ll need to have a team using the same one as you) and UI preferences you’ll naturally gravitate towards one. Having tried them all very recently I’d say that Atlas.ti was the most aesthetically-pleasing and Mac-y for me, and I felt like the UI was most intuitive for my needs. That said, so many of the people I work with use NVivo that I may end up having to go with that even though I didn’t love working with it. (As powerful as it is, I just preferred Atlas!)
I did try out some of the open source options and felt that none were quite robust enough, but that’s changing and I’m hopeful that in a few years they will be as full-featured as these others.