(1) Run your search in PubMed.I of course had to try this out. Years ago I embedded (now broken) search links in web pages, but one runs into the notification problem. It's very boring to keep clicking on a link and finding nothing has happened. Syndication is a vastly better model.
(2) Select RSS Feed from the Send to menu.
(3) Click Create Feed and copy the XML icon into your RSS Reader.
I tested this out an old search I've long been interested in: the genetics of berry aneurysms and screening for subarachnoid hemorrhage. I logged into PubMed and set up a search, then followed the directions above; I limited the search to Bloglines, my syndication (RSS, Atom) web client, and added this URL as a feed to a new folder called "PubMed searches" (you can copy this URL and try this in your favorite feed client yourself.)
On my first view I got 41 hits. After reviewing those the list was empty (as it should be), but as literature emerges that list should refresh. One wonders how the PubMed server will handle the transactional burden, since these unique searches may not scale well to thousands or millions of users. For now, however, the PubMed RSS and Bloglines combination is the neatest thing I've seen in this domain since Grateful Med.
If you'd like to see how this looks, without doing all of the above work, visit the public view of my bloglines list and click on the PubMed folder. It is very cool. You can also see the searches on my (hidden) blogroll page.
I guess I'll have to update that old, old, page of mine. (11/17/05 - I did. The page was over 6 years old.)
 Old dogs remember MEDLINE, MEDLARS, Grateful Med, etc. Really old dogs even remember the vast printed volumes of Index Medicus, and using early terminals to send query strings to Bethesda. Creak, creak ...