On Tue, Feb 21, 2006 at 12:24:09PM +0900, Michael Glaesemann wrote: > I don't know the details of your database schema, If you want to you can look it up here: http://salaam.homeunix.com/twiki/bin/view/Gnumed/DatabaseSchema Feel free to comment ! > but I think the > relationally proper way to do would be to have a separate table for > the maximum ages for vaccinations that have them. I know, but, oh no, yet another table ;-( And it would also just lead to another form of NULL via left joins as you point out below. I specifically wanted to avoid that by something like 'infinite'::interval in some way or other such that I could always do "... where now < date_of_birth + max_age ..." and not need an "... or max_age is null ..." in all the places. > Vaccinations that > *don't* have a maximum age would not have an entry in the table. As you notice further down my predicate was wrong, actually. You found the proper predicate by yourself, though: "Do not care about the age of the patient when deciding whether to give this vaccination." > The "special value" method, e.g., "999 years" is another way of > indicated a special value, but in this case I think it's a bit > different. As I see it, the predicate for the vaccination_max_ages > table is "The vaccination 'vaccination' must be given before the > patient is 'maximum_age'". Using a special value changes this > predicate to "The vaccination 'vaccination' can be given at any time > in the patient's life." As you point out, using a sufficiently large > interval for maximum_age makes that statement very likely to be true, > but the predicate is not exactly the same. Not having an entry in > vaccination_max_ages is much closer to the idea that the vaccination > has no maximum age. Well, but there's not really a medical difference between the two AFAICT. > That's the theory, anyway. Hope this helps a bit. It confirms my thinking isn't entirely wrong. > Currently on the todo list there's a mention of adding infinite > dates, similar to infinite timestamps. Perhaps infinite intervals > could be added as well? I'd be delighted to have that happen. Karsten -- GPG key ID E4071346 @ wwwkeys.pgp.net E167 67FD A291 2BEA 73BD 4537 78B9 A9F9 E407 1346
