Nov 2, 2006, 30 min discussion at CATO.
D = David Cutler
R = Robin Hanson
M = Michael Cannon
X = unknown others (Arnold Kling, Peter Van Doren, Jagadeesh Gokhale, present)
R: We were waiting for you to start
M: OK, start! I was telling David that I was mostly interested in having this discussion because I wanted to hear the interaction between you two because you [Robin] have the interesting theory about whether or not any of this is actually worth a damn. Not any of this, but in aggregate .... [interruption]
R: In a sense I don't feel like I have this weird position but I guess I do because I keep getting these reactions; I feel like I'm just telling people the standard empirical results of health economics but they look back at me as if "you must be crazy."
D: The cross section results are very different from the time series results.
R: Well that's something to talk about. First of all, what is the relative amount of data in each case or the number of analyses, what is the confidence in each of these analyses given the kinds of complications that are there, and then what would be the differing implications?
D: They are also picking up fundamental differing concepts. I think the cross section is picking up the marginal product of medical care, and the time series is picking up the average product of medical care.
R: So if we accepted those two results at face value here for a moment, there are still dramatic implications that most ordinary people would be shocked by if they would accept the marginal results.
D: I know this very well
R: Yes, so I feel like I'm getting that response, and for the purpose of making that point clear I'm willing to grant the average product.
D: Yes, no, I think its right actually. I think the two central facts about medical service use is that the average product is really high and the marginal product is really low. But I actually think there is probably not one single marginal product,
R: Oh, sure, of course
D: because I think there is so much inefficiency in different parts that the marginal product of a lot of stuff, particularly stuff that is well reimbursed, is very low, and the marginal product of stuff that is not reimbursed well is often very high.
R: So obviously there is a thousand categories of medical spending,
R: and so, every variation study we do has some relative weighting of those categories that it produces, and in the weightings we have seen the net result of that weighting has been relatively little, within the margin of error
D: because typically you use spending weights, and spending weights are reflected
R: And also visit weights say, or
X: I think there are some availability studies; like if you look at a region with lots of specialists and then very little, and you don't see a difference
D: and then number of doctors would be another
R: But I think of that as kind of a spending weight, that is, what you see what you are doing is saying where the money is, if you weight it by where the money is,
D: what is the marginal product,
R: So what we agree on I presume is that the vast majority of studies that have looked at variations in medicine induced by price or wealth level of nation or local doctor availability or practice variation, those variations, when you look at an aggregate health effect, you usually don't see one.
D: Yes, though actually I want to subdivide it slightly a bit. There have been a couple of recent studies about the price of medications, that if you follow them to their logical conclusion will say that health is harmed when the price goes up. So there are a couple of studies that say things like when you take people on a statin drug, and raise the cost sharing for it, even people who need it
R: Is this the study of a particular drug or of the overall category of
D: well you look at people who are on a particular drug, so they are on their statin drug, and the cost sharing is $5, and then you raise the cost-sharing to $15, and there is a generic for $5,
R: Right, so let me just stipulate right off, that when you go away from the broad aggregates and start to look at narrower categories, you can identify some categories that at least seem to be a lot more solid evidence for a positive benefit. Like
D: No, no, no, I was going to a different point. I was making a point about the variation induced by price.
R: But also within a certain category of treatment, that is price variation for heat attack say. You weren't talking about overall mortality rates, you were talking about a particular class of patients.
D: Yes, but though the class of patients is men over the age of forty who need a cholesterol treatment. So that is a large class of patients. The studies exist now that show that when you raise the price the men stop taking their statins, not just switch to generics.
R: Seeing the mortality effect directly hasn't been
D: People haven't followed people long enough to, but I suspect that in a couple of years we'll say raising cost sharing on drugs can be a bad strategy for health.
R: Well it depends on how many other drugs we aggregate; if it was just on statins then I might be willing to agree. If we throw in all the other drugs, other things could go up or down and I wouldn't have as much confidence about the aggregate effects.
D: Uh huh,
R: So my most confident claim would be about large aggregates, the broadest aggregates we see, and I would grant say, low birth weight babies as a category of particularly likelihood of positive benefit of medicine, say
R: even over broad population aggregates, say
X: Does your belief require in any way that there be a fair amount of harmful medicine, so that when you walk into the hospital, you are taking this lottery ticket that, you know
R: Well we already have somewhat independent evidence that there is a fair amount of harmful medicine, so I don't know if that is particularly controversial. Given the evidence we have of a fair amount of harmful medicine, and the evidence we have of identifiable plausible cases of beneficial medicine, then I guess
D: In a cross section, most of the variation in use is I think associated with, probably not harmful stuff but just stuff with incredibly low value. But if you look at let's say, areas of the country, if you look at Boston, which spends the most, and you compare it to the upper midwest where spending isn't so high, even let's say Medicare, the spending is very different, it turns out that's not associated with for example getting more procedures, big procedures, if you have something really bad for you.
R: Well, the RAND health insurance experiment, which I guess I would consider our most solid data,
D: Uh huh
R: they had not the largest analysis you'd like, but they did have some analysis of the differences in appropriateness,
R: and severity of diagnosis and degree of hospitalization for the marginal treatment versus the non-marginal treatment, and they saw no significant differences
D: They actually found that in people with higher cost sharing both less appropriate use and less inappropriate use,
X: well it was appropriate use, but my generic memory of this, and I'm not ...
R: The ratio was the same
D: Right, so they got rid of both
R: Right, you couldn't identify the marginal treatments as sort of sniffles and warts or something, it was just, in terms of the way the medical system looked at it
D: So one question is if they followed people for longer would they have found some adverse health impact for them
X: Yes, essentially no difference in outcomes, but if you were somebody who was doing this, it would sort of have implications, you would say, they forgo some necessary care, if you were doing pay for performance you would say "Oh" they are going to get dinged, because they didn't do some things that supposedly helped. But then you don't see the difference in outcomes, which is a real, kind of, which is more important - doing the quote right thing or having the
R: So there are two directions we can go with the conversation, one is to more carefully try to disambiguate what we know about the harmful versus beneficial medicine among the variation at a time, another is to talk about the differing policy implications if we accepted at face value the cross sectional variation and the cross time variation. I don't know if you have a preference; I just don't want us to wander off on a little path and then the time's up.
M: Why don't you pick your
X: Well how about there is the time series literature, and I think of David in that and I think of Lichtenburg and the pharmaceuticals, and then you've got the Wennberg, evidence based medicine stuff, which is cross sectional in largely Medicare database, and then one thing that is way out there that is different is Fogel that argues that is nutrition, and I don't know how to scientifically even compare that argument versus
R: One way to focus discussion is to pick particular policy recommendations and ask what these various data points have to do with those particular policy recommendations. So if we were talking about funding medical research, we could combine these two results and still endorse funding medical research because of the overall long term benefit.
D: Uh huh
R: But if we are talking about whether we should subsidize medicine or tax it, or something like that, the cross sectional variation would suggest that it wouldn't hurt to cut off the subsidy and maybe it wouldn't hurt to tax it.
D: Which I agree we should except if you think it will have some impact on the average diffusion of, the average innovation.
R: Right, so the question is how much does the rate of innovation depend on
R: the size of the community of practice. I presume we can agree that that's not clear.
D: Correct. The best evidence is a recent paper by Fikelstein that half of medical cost growth is the result of the spread of insurance, induced innovation from the spread of insurance
R: half of cost growth or half of innovation?
X: cost growth
R: So again, that doesn't pin down the innovation driver, right? In general its hard to pin down origins of innovation in any field.
D: Yes, that is correct.
M: I think she's come closest though.
R: She's not talking about wealth, she also has some tech
D: She also has some technological stuff.
M: I don't remember that bullet point, I remember the medical spending
D: She does have a bunch of technical stuff.
X: I thought that again, economists study aggregate data, but then there is how little of medicine is based on clinical trial evidence. We've got laissez faire surgery, which is guys invent stuff, and then teach their students, and then it spreads. Jerome Groopman in the New Yorker who had back surgery wrote how he thought it was science and it wasn't and he says his whole history, and he went after the back surgery guys. And then we got pharmaceuticals that's $800 million clinical trials per, we've got the two extremes, and if you combine them maybe that's why you end up with the numbers you talk about, which is the
R: Which is your presumption, that which is actually more innovative? You are talking about two different systems for innovation.
X: I'm saying surgery may not do anything, in fact its a non-scientific part of medicine.
R: If something being more scientific means its more innovative, that would be interesting to know; then we could encourage medicine to be more scientific. But we don't really know that.
D: If you require $800 million to prove the thing you are more likely to know if it's effective than if you don't require it, if you just go on the basis of experience with three patients, which is surely true.
X: If you look at medical journals, it's guys write up what they do in their practices, and somehow that's become the standard.
D: For most things, for a lot of things there are randomized trials, the big cardiovascular surgical things actually have had randomized trials of them, but what tends to happen, that tends to comes after you do the thing, you do the thing, you do the thing and then you do a big randomized trial, of course by the time you've done it the practice has changed, so that its out of date anyways.
R: So I guess if we actually cut back medicine by a big margin, like 30% of spending or something, what would, I guess, the overall judgement is, would that hurt innovation so much that it wouldn't be worth the savings of the 30% of apparently wasted spending in the short term?
D: I've on occasion tried to argue that, but I can't ever convince myself, that static inefficiency is good, if you think there are under incentives to innovate. Of course we know there are models where there are over incentives, but if you think there are under incentives, you would trade a little bit of dead weight loss today for more induced ...
R: You'd have to think this sector was more that way than other sectors that you were substituting from, because if all sectors were equally under incentivized to innovate, then you would be discouraging other sectors even more
R: in order to encourage this one, and that wouldn't be obviously beneficial.
D: Yes, that's right.
R: So is the externality of innovation higher in medicine than in other sectors I guess is would be the question.
R: We could just agree to encourage innovation more in general if we knew a way to do it, but its not clear we know how.
D: Monopoly profits are one obvious way, so this is the argument that teaching hospitals make, is that we need our monopoly profits so we can fund our research, so don't you come after us on anti-trust grounds for being high prices,
R: If it was clear they were using their profits for innovation I guess that would be ok; simply handing monopoly profits to somebody because they say they are going to innovate isn't particularly reassuring.
D: They haven't found it a very successful argument but they keep making it. I think its right though. The reason I focus a lot on average value of care is because if you look at most of the policies, people say, ok, I want to eliminate the marginal care which has low value, and then they propose something which in fact amounts to, either, in the short term, or more certainly in the longer term, being something that would effect the average value of care.
R: So let's explore that further. The variations we see are induced by things like practice variation and price changes or cultural differences across countries, those variations in the short term don't show a health difference as a result of those variations,
R: so if we had a policy who effect mimicked those variations, we could be relatively sure our short-term cost was low,
D: the only one which we would really know a policy about is the price, and I was arguing earlier maybe we don't know that one as strongly as the others, but let's leave that aside. I suppose the other thing you could do is you could say if its cultural things God knows how, but availability of resources you could actually restrict the availability of resources, though I'd be surprised if that in this room that got a majority vote.
R: You could cut the subsidy, for example, they might vote for that; eliminate the health subsidy. Taxing medicine maybe you won't go there, but eliminating the subsidy perhaps. What seems most surprising to me is that in the Washington Post every week there is an entire section on health with dozens and dozens of articles about this study and that study, and this overall result that health economists seem to have robustly figured out about this marginal effect
D: uh huh
R: doesn't make it at all into the public consciousness.
X: There is a question here. I mean you do all kinds of things on the margin. You're spending money on all kinds of things, and the test says you see no effect today, cause when you aggregate positives and negatives, they seem to be cancelling out, or maybe everything has such low value, that everything is zero.
D: I think that's maybe a more accurate statement is that at the margin a ton of stuff has a zero marginal value.
X: Right, so how can you know that in order to get a long term average increase in benefit you don't have to go through, I mean you learn by trial and error, right?
R: That's what we were just discussing.
X: Lots of errors, and over time by eliminating these subsidies, you might,
R: first, it depends on the kind of innovation system you use
X: but this data show I think, you're discussing the average effect and the marginal effect look to be positive through '96 and then after '96 we've got nothing in the margin in the time series, and so the trial and error isn't getting us better, its getting us worse.
X: We won't know that for a while.
D: Well that's one data point which is certainly clear, but nobody knows what the hell to do with it. I think as a general proposition, one of the reasons why clinical trials are not right about the world is that people do get more experienced with them, and so a lot of things that have very low value in clinical trials probably on some patients have very high value. But then my interpretation of the marginal stuff is that we then just overdo it. We then just give it to all sorts of people in whom .... You know, so an example is, ... I think a very common thing that happens at the margin is if you say why does Boston spend more than elsewhere would be, you're in the hospital for pneumonia and then we do biopsy on the lungs and the cardiologist comes around and then a virologist comes around, and they all do their thing, and then some of them find mythical things, and a fair number find things that will never kill you. A patient sick with pneumonia will die in two years anyway. So its all just sort of very low value. And then there is a little bit, which is, in the course of doing all this, they give you an infection, which ends up being worse. Although I don't suspect that's the bulk of it. I suspect the bulk is just people doing their thing, and thats
R: drug errors, treatment errors
D: it has a relatively low value.
X: How do you deal with, its not about mortality, its about hip replacements.
D: The quality of life end of it?
D: I don't think anyone has measured the quality of life
R: well enough
D: well enough
R: but attempts have been made so far
X: Fogel tries with his, you know
D: We know that over time quality of life is improving quite a lot. What I can't tell you very well, is, you know, the US spends a ton more than Canada. Its pretty clear that the mortality is not a big endpoint of that. Its less clear whether quality of life is better for Americans than for Canadians. A couple studies say yes, a couple studies say no.
X: I don't think I've even seen those.
D: There are a couple of things, they tend to look specifically, say for a particular condition, like say, people who have had a heart attack, and their quality of life, so there's been a little bit of work on sort of walking problems after a heart attack, but they are no way, they are small samples of this and that, but there's not a lot of really systematic.
R: Well, I mean, the RAND experiment actually wasn't a mortality experiment, it was a quality of life experiment,
R: so in fact our one clearest data point about cross sectional variation is a quality of life datapoint, and the health quality of life was the same, but there was a big ding in terms of restricted activity days.
X: But its so old that hospitals don't do anything like that anymore
R: It's the data you have.
D: No, but in RAND, just to follow up the point, in RAND the people use less when they pay more, and what they do, is that when they have mild pneumonia they don't go in the hospital. All those things have been taken out of the hospital in the intervening years since then. So I actually think we should do the RAND experiment again today,
R: I would endorse it, yes.
D: because in today's environment the impact would be very different. I don't know if bigger because maybe there's some other things
R: Spending has tripled since then, obviously, its a different world, but still keep doing other kinds of cross sectional variations, and we haven't seen that much difference in other cross sectional variations, so its not clear we would
D: No, but I think the price might give a different answer from the other kinds of variations if you did it right
R: If you did it now you mean?
X: OK, why do you say that?
R: RAND was a price experiment, right?
D: I think a price variation experiment done today might give a different answer than a price variation experiment done in the 70s.
R: Because of statins, or?
D: Because there is more cheaper stuff that people can take that is really helpful than in the early 70s or mid 70s, and people seem to be pretty responsive to that in price
X: That sounds plausible. I just think of examples. My favorite one, you know, I hurt my back, and doctor tells me to get an MRI. If he'd said, that's going to be a thousand dollars out of your pocket, I would at least ask the doctor what the hell difference is it going to make in my treatment whether I get this. What are you going to see on the MRI that is going to effect the treatment.
X: It turns out that we don't know anybody
D: Do the opposite of what the doctor says and you'll be doing just fine.
X: I've got to agree; there is certainly a lot more stuff.
R: Wait wait. If we have a lot of cheap stuff that is effective, than an experiment where you lowered the price would have a plausibly less difference it made, because you could substitute the cheap stuff for the more expensive stuff.
X: He's saying that because people are so price sensitive they wouldn't use really good stuff and they would see mortality effects in a RAND done today.
D: But the bulk of the reduction in spending would have no impact. Because the bulk of the reduction in spending would be the thousand dollar MRIs. But some of the spending would be associated with cheap stuff.
R: Well sure, but some would be avoiding harmful stuff too. So its not clear. You would spend fewer days in the hospital to get an infection, or like that.
M: Should a RAND two include, you know, throwing at all the subjects information about the effectiveness of different procedures?
D: Well, yes, that's part of any strategy now. I think a RAND two should have two arms. One is the price arm and the second arm is provider payment; we should compare the impact of price incentives on the demand side with price incentives on the supply side.
X: Pay for performance?
M: Well we have ten minutes before it starts so we should probably wrap up after five more minutes.
D: Am I the decider? That's really a change.
M: You are the deciderer
R: We, I guess, do we disagree? I mean this was presented as .. Do I still look like the oddball?
M: My way of saying I don't know.
D: So let me ask you a specific question. If we put everybody into MSAs, would the population health get better, stay the same, or get worse?
R: MSAs funded at the same level?
D: You mean the same level as RAND one?
R: No, I mean at the same level as the current funding of non MSAs. So you are saying, counterfactually, take all the people who are now in other plans and put them in MSAs, and then the question is whether they would contribute a similar amount.
X: The difference is instead of getting money every time I get a health treatment, I get money and can save it.
R: I can put the same premium amount into the MSA.
D: For my answer it doesn't matter whether the employer funds part of the MSA or doesn't fund part of the MSA.
R: I'd say that in the short term we shouldn't see much difference. Ill leave open the longer term innovation question, but I think in the short term no effect.
X: Do people forgo the low marginal, or do they proportionally forgo all medical care, including the ...
D: My guess would be that after five years we'd find that some stuff that is cheap people got rid of, and their health suffered, but that that was a small share of the dollars, so that over all their health would be worse.
R: Noticeably or just ..
X: Why doesn't that happen in Europe, is it just that governments make better decisions about restricting supply than people make about their own health care
D: I'm not sure I quite understand. I mean in Europe on the demand side there is very low cost sharing
X: Let's say its a given that they spend less and that they don't have worse outcomes. So what that says is that when Europe does things on the supply side, they are doing smart things. Whereas you are suggesting that if consumers do on the demand side they will do some dumb things.
D: Yes, because in Europe what they basically do is they restrict the high tech stuff, a lot of which is over used. So in my model, high tech stuff is overused, and low tech stuff is under used.
R: So your prediction is based on the same prediction that you would see a different result in the RAND experiment two, done today.
R: So you do agree that if we saw the same result in RAND two today, then the prediction would be different, that its based on that presumption that in fact the price effect is different today than it was thirty years ago.
D: Yes, and there's a little bit of a hint in the RAND one, about the low income people not taking their anti hypertensives, that suggests that if we had followed them, we'd have observed worse outcomes. But that's a seven to ten year process of having a stroke and dying and so on.
X: MSA to me seems to me that if its going to work its going to work across the lifecycle, cause you can save lots of money when you are young but introducing MSAs now and then quickly trying to test it, its not a true test in my view of MSAs cause the advantage of the MSAs is you are now cross subsidizing stuff when you are young in a pooled system, whereas you could keep all that money when you're young and then save it and then spend a lot of your own money when you're older. So to do a true test of MSAs would take 70 years. No ones going to wait for that.
M: About your prediction for moving everyone into MSAs, RAND one is pure demand side, it had no effect on the supply side, cause it was small experiment, and Fikelstein, one of her findings I think is that a reason why we believe the effect of free care or expanded coverage on spending is smaller than it probably was is because the only evidence we have is based on this demand only experiment, where she captured the supply side response. The question you just asked was not about a RAND two, it is about moving everyone into MSAs, and the kinds of care that they would pare back on.
D: I'd think they'd pick it up in a RAND two.
M: But when you say that you think the health outcomes would be worse, are you accounting for a supply side response to the fact that people have all this money and are going to be much more stingy and
D: That wasn't showing up in my answer.
D: What was showing up in my answer was since the early 70s we've invented a ton of preventive things with short term costs and long term benefits, and we know that people are overly sensitive to short term costs to long term benefits, so if you raise the short term costs people will just stop taking them.
R: RAND two needs to last ten years then.
M: Because one of the things I think we've seen happen, just with the small consumer directed health care we have so far, we're getting supply side responses in terms of information that people get. So I like to think that people are going to make smarter decisions about what they pare back on, than they did in RAND one.
D: I wish that were true.
M: I would like to think that. We gotta go.