Recently I've been tinkering with AIR and Google Health (GH). It's been surprisingly easy, if one can overlook the endless stream of XML returned by GH - but there is no other way, HL7 would be just as nasty looking. I don't know yet how it returns the file/image data that can be attached to the GH account.
AIR seems an ideal environment to build a desktop client to front a GH cloud-based application: it's lightweight, Javascript-compatible, and portable across platforms.

Speaking of, it seems that AIR
will be ported to mobiles as well. I would argue that the paragraph above (and not just the stronger OO features found in Flash and available to AIR) is a strong reason to do this port, although I am not sure how easy is to develop and maintain AIR applications, and also I am not sure how well do these applications perform given the several layers of virtual environments they have to execute in.
Will write more thoughts as soon as I finish the small scale project I am working on right now, 4-5 forms of reduced complexity (but with a significant amount of functionality built in; the underpinnings of this relatively simple project are amazingly complex and would have been hard to imagine only a few years ago).
I haven't looked at Google Tables yet, but read some things about YQL and can see a scenario where medical (and other) personal information (e.g. reverse phone lookups, credit history, white and yellow pages) is queriable over the web via a SQL-type of language
with the right security in place. In fact, the infrastructure already exists! So it would be just a mater of connecting the pipes.
And, I haven't even started to look at HealthVault's API (if it has one).