Lydia at Right Reason makes a number of startling claims...
(1) A person's past wishes are [always?] outweighed by their present ones. "If a person does agree [to refusal of medical treatment including feeding tubes] but later cries, 'Water!' that wish should be answered with water, not with morphine."
(2) The present wish need not be "a fully coherent thought about 'wanting to stop dying'... Even if the person is simply feeling--rather like an animal under similar circumstances--an inarticulate and overwhelming desire for water, should not that count as a relevant stand-in for the person's 'wanting water'?"
(3) "a person in great pain will want the pain to stop... Therefore, giving pain medication under these circumstances has as one of its effects that it prevents a patient who has previously agreed to death by dehydration from changing his mind and wishing for food and water."
Working backwards... #3 is a simply atrocious argument. It effectively says that, by failing to make a desired event or process unpleasant, we thereby "prevent" the person from changing their mind and disliking said event/process. This is so patently ridiculous that I'm not sure how to respond. I think I will simply lift my jaw from the ground and move on.
#2 is a serious threat to personal autonomy. By ignoring the structure of our desires, it fetishizes the satisfaction of urges over the genuine preferences of the agent. According to #2, we should take along a keg of beer to an Alcoholics Anonymous meeting, since apparently an "inarticulate and overwhelming desire" for alcohol should indicate that this is what they really want - never mind their higher-order preferences!
Besides, an animal craving for water may not be so much an intrinsic desire for the object of water itself, as a desire that the pain of dehydration go away. In any case, desire frustrations can be avoided by either fulfilling the desire, or else getting rid of it entirely. Which is better will be determined by the person's higher-order preferences. Do I want to feel thirst and have it satisfied, or would I rather have the pain medication and thus avoid the cravings altogether? Well, I'm happy drinking, but the patients we're talking about here have expressed their preference for the latter. We should respect that preference. The fact that a patient is in pain (and wants it to stop) in no way implies that they want to live. Lydia seems to have missed this obvious point.
#1 is the really interesting issue for me. So let's forget about the earlier problems, and assume that a patient has genuinely changed their mind (or what's left of it) and wants to live a life that they had previously judged "not worth living".
There are cases where past desires become irrelevant once superceded, e.g. a child's desire to be a fireman. But other times they seem more important. Take the case of a devout atheist who strongly wishes for (almost) his entire life that no priest be present at his deathbed -- and impresses upon you that you ought to ignore any later request he may make to the contrary. But, lying on his deathbed, he has a moment of weakness, and asks you to send for a priest. Would doing so make his life go best? I think in such a case his past desire, being a 'global preference' that is central to the integrity of his life, takes precedence. The same seems true of the person with a strong and informed desire to forgo artificial life support and "die with dignity" rather than rot away in a hospital bed for the rest of their life.
But this one, at least, is not clear cut. Someone may have a genuine change of heart. If they actually change their global preferences, so that they now consider the hospitalized life to be well worth living, then it would be a grevious wrong to go ahead and kill them. That would be murder. But if their new desire is merely due to a temporary bout of 'weakness of the will', or a lower-order urge of the sort discussed in #2, then it seems more plausible to take their earlier choice as the authentic expression of the agent's preference. (How to tell the difference between these two cases is an extremely important practical question, of course -- but I'm much more interested in theory.)
Here's the central question, as I see it: Given a conflict between past and present desires, what determines which ought to take precedence? My tentative answer is appeal to the structure of one's desire-set, and in particular one's global desires. But I'd love to hear some alternative suggestions (or elaborations of the present one).
Update: Don't miss the excellent discussion in the comments below.