11 March 2010

Cost Control and Multi-Objective Optimization

Marginal Revolution | Tyler Cowen | Much cheaper, almost as good

Here is part of the problem behind health care cost control, from the Annals of Internal Medicine:
Under conditions of constrained resources, cost-saving innovations may improve overall outcomes, even when they are slightly less effective than available options, by permitting more efficient reallocation of resources. The authors systematically reviewed all MEDLINE-cited cost–utility analyses written in English from 2002 to 2007 to identify and describe cost- and quality-decreasing medical innovations that might offer favorable “decrementally” cost-effective tradeoffs—defined as saving at least $100 000 per quality-adjusted life-year lost. Of 2128 cost-effectiveness ratios from 887 publications, only 9 comparisons (0.4% of total) described 8 innovations that were deemed to be decrementally cost-effective. Examples included percutaneous coronary intervention (instead of coronary artery bypass graft) for multivessel coronary disease, repetitive transcranial magnetic stimulation (instead of electroconvulsive therapy) for drug-resistant major depression, watchful waiting for inguinal hernias, and hemodialyzer sterilization and reuse. On a per-patient basis, these innovations yielded savings from $122 to almost $12 000 but losses of 0.001 to 0.021 quality-adjusted life-years (approximately 8 hours to 1 week). These findings demonstrate the rarity of decrementally cost-effective innovations in the medical literature.
Let me just repeat that last sentence: "These findings demonstrate the rarity of decrementally cost-effective innovations in the medical literature."
That's a great empirical result, but I think there are already good theoretical reasons to be skeptical of the ability to bureaucratically make good decisions in these sorts of situations.

The kinds of cost-saving measures we're promised with most centralized health care reforms boil down to a multi-objective optimization. It's really disingenuous of Obama et al. to insist these "comparative effectiveness research" decisions are going to be made "scientifically" for two reasons. First of all, politics always, always, always butts it's head into these situations. (E.g the ritually postponed "doc fix" or the hullabaloo last year about mammograms.  On the other hand, the Fed has been totally objective and independent and avoided political pressure to ... oh, never mind.  They utterly failed at that too.) Secondly, and more importantly, there is no established scientific procedure for doing multi-objective optimization.  Such a thing not only does not, but can not exist. That's a corollary to Wolpert & Macready's No Free Lunch Theorem. (See Corne & Knowles 2003.)

Have you ever tried to do a multi-objective opt problem in even a semi-rigorous way?  You quickly come to the realization that in anything but toy problems or very limited domains there is no "right" answer.  It is extremely rare to find one solution which is dominant (in the technical sense).  You're forced to make subjective choices about relative trade-offs.  And if you're ever had to defend those choices to even one other person (like, say, your boss or research adviser *cough* personal experience *cough*) you find out that even two well-intentioned people with identical end-goals acting in the best of faith often disagree wildly about how to make those decisions.

There is no one-size-fits-all, single best "scientific" solution to these sorts of problems.  That's why centralized, one-size-fits-all medical regimes, as egalitarian as they may seem, scare me.  And by that I'm counting everything from France's socialist system all the way down to America's FDA.  In attempting to serve everyone they can only cater, in the best case, to the Everyman.  But the Everyman doesn't exist. He's fictional.  I suppose if you find yourself right in the middle of every health-related opinion, if you have median preferences and modal habits, if you make only the common choices, if you have only the common desires, if you like to live your life exactly as prescribed by average opinion then you might find yourself well served.  If not, well, tough luck.

In parting, I'll refer you to Dr RW for a more detailed discussion on the one-size-fits-all requirement of C.E.R. studies.

No comments:

Post a Comment