• What and where is “waste” in health care?

    Health care spending is confusing. Conventional wisdom is that there is an enormous amount of waste in the system, “anywhere from 30 to 50 percent of medical spending […] about $1 trillion annually,” according to David Cutler. Victor Fuchs once wrote, “Although many health services definitely improve health, in other cases even the best known techniques may have no effect.” And that was in 1966.

    In a recent paper published in The Milbank Quarterly, Robert Kaestner and Jeffrey Silber explain that,

    Fuchs’s [1966] characterization of the contribution of health services to health would now be referred to as an example of “flat-of-the-curve” medicine. […] The same “flat-of-the-curve” argument was used to characterize medical spending during the 1970s. Gruber summarized the findings from the Rand Health Insurance Experiment, which took place between roughly 1975 and 1980, as follows: “It suggests that, at least at the time of the experiment, the typical enrollee in the study was on the ‘flat of the medical effectiveness curve,’ the portion where additional care was not buying medically effective care.” […] Fuchs returned to the issue some forty years later, stating with respect to spending on medical care for those on Medicare: “The bottom line is that a considerable amount of the care delivered in the United States is ‘flat-of-the-curve’ medicine”.

    Kaestner and Silber are justifiably puzzled. They argue that if there is so much waste in the system–that has been there for decades–market forces should have driven much of it out by now. Even public programs have an incentive to eliminate inefficiencies, they say. If bundled payments and capitation are the answers, why haven’t private insurers implemented such systems more widely? In addition,

    A second reason to be skeptical of the “flat-of-the-curve” hypothesis is that much of the evidence supporting it comes from nonexperimental studies, with the notable exception of the Rand Health Insurance Experiment. The most obvious problem with nonexperimental research in this area is that causality is just as likely to go from health to spending as it is from spending to health, since more resources are typically spent on sicker people. Though obvious, this is a difficult problem to overcome empirically. But several recent studies that used credible research designs to address this empirical problem found evidence that additional spending on medical care is actually very effective (Card, Dobkin, and Maestas 2008; Chandra and Staiger 2007; Doyle 2005, 2007)

    And with that, Kaestner and Silber deliver there own results that suggest additional health spending produces additional value. Using an instrumental variables approach, separately for Medicare patients admitted to hospitals for each of many possible reasons (general surgery, vascular surgery, orthopedic surgery, CHF, AMI, stroke, and GI bleeding), they estimated the relationship between spending and thirty-day mortality, controlling for hospital characteristics and patient health factors. Further, they restricted the sample to patients experiencing in-hospital complications, which “yielded a more homogenous sample in regard to severity of illness.” (In-hospital complications are random events. Patients experiencing them receive additional treatment that is, arguably, not influenced by unobservable factors that also relate to mortality. Selecting the sample in this way helps mitigate against the possibility that mortality and treatment intensity are jointly determined by other unobserved factors, thereby increasing confidence that the estimated effect of spending on mortality is causal.)

    The found that,

    Estimates from the analysis indicated that except for AMI patients, a 10 percent increase in inpatient spending was associated with a decrease of between 3.1 and 11.3 percent in thirty-day mortality, depending on the type of patient. […]

    [O]ur results suggest that even though it may be cost effective to eliminate some portion of inpatient spending, this reduction would come at a considerable cost for survival, at least for inpatients. […]

    Based on our results, the narrowest interpretation of the “flat-of-thecurve” hypothesis—that health care spending could be reduced by 20 to 30 percent without adverse health effects—may be seriously misleading.

    How inconvenient! Notice, though, that their results don’t by themselves suggest the level of waste in the system. It could still be 20 to 30 percent or more. They’ve shown that there is still life-extending value in additional spending for many Medicare hospital patients. For such patients, marginal spending captured by the data lowers thirty-day mortality significantly.

    The sole basis for skepticism of the level of waste typically cited is the theoretical argument that such waste should be driven from the system by market forces and other cost control incentives. But this point of view, and the whole discussion, presupposes what “waste” is. It could be that the system delivers quite a large volume of services that don’t improve health (e.g. some outpatient services, administrative services, etc.). But those could still be services that patients demand or are otherwise necessary for the system to function. (Can you run a hospital without a huge billing department? No! Does billing itself improve health? No!). That is, certain types of spending can be “waste” in one sense and “necessary” or “desired” in another.

    Maybe one can’t spend another $10,000 on a Medicare hospitalization without having first incurred $3,000 of expense that doesn’t itself improve health. Maybe that’s the way we like it. I’m beginning to think it is. Collective griping is distinct from revealed preference.

    (See also “Is health spending worth the money?“)

    I will take a look at the following papers cited by Kaestner and Silber and comment on them in a subsequent post.


    Card, D., C. Dobkin, and N. Maestas. 2008. The Impact of Nearly Universal Insurance Coverage on Health Care Utilization: Evidence from Medicare. American Economic Review 98(5):2242– 58.

    Chandra, A., and D. Staiger. 2007. Productivity Spillovers in Health Care: Evidence from the Treatment of Heart Attacks. Journal of Political Economy 115(1):103–40.

    Doyle, J. 2005. Health Insurance, Treatment and Outcomes: Using Auto Accidents as Health Shocks. Review of Economics and Statistics 87(2):256–70.

    Doyle, J. 2007. Returns to Local-Area Health Care Spending: Using Health Shocks to Patients Far from Home. NBER working paper, number 13301. National Bureau of Economic Research: Cambridge, MA. Available at http://www.nber.org/papers/w13301 (accessed October 13, 2010).

    • A request: Austin, I’d love you to write an entry on how to evaluate the quality of instruments like those in the Kaestner and Silber paper.

      The first instrumental variables I was introduced to were about as randomly determined as you could get (in the Angrist tradition: lottery results, quarter of birth, sex composition of children, etc.). These instruments are sometimes whacky, but they are uncorrelated with any unobservable. So, as long as there’s some first stage, it’s a success.

      But, it seems like the majority of instruments in health econ and HSR papers aren’t like those instruments. Instead, the instruments seem to be of the flavor “I’ll use some variable close to X instead of X directly”. For instance: “Instead of using a person’s weight, I’ll use their family member’s weight” or “Instead of using a patient’s surgical spending, I’ll use that hospital’s EoL spending” (Kaestner and Silber). The problem, of course, is that although these instruments may not have any effect on the dependent variable, they could be correlated with some unobservable that does have an effect.

      How does one evaluate the exclusion restriction in these papers? I know it is impossible to do it comprehensively, but how do you think about this question?

      • @Aaron – Ultimately it’s based on theory. One can do falsification tests, sometimes. One better pass those, but that’s not conclusive. One can do overid tests, but those are game-able, according to Angrist. One better have sufficient power.

        But you can’t prove a negative. Maybe the instruments are still bad. So, you need a good, theoretical (or institutional) reason to believe exogeneity. At the same time, if you suspect instruments are bad you’ve got to have a plausible explanation why that could be. You can’t just say, “Maybe there are correlated unobservables.” You’ve got to say what they could be. Better, if that is somehow testable, you’ve got to ask the scholars to test it.

        Most instruments that get used again and again (like EOL, or M’caid policy variation, or differential distance) have been well tested or have good stories. One has to go back in the literature and find those.

    • They argue that if there is so much waste in the system–that has been there for decades–market forces should have driven much of it out by now.

      Blood letting among other things was wasteful and yet persisted for at least a hundred years. Even today, if you believe the research organ food, supplements, back surgery are all wasteful and yet persist.

      Robin Hanson has proposed that a possible explanation that we spend on health care to show that we care and we care less about outcomes and so none of it is waste as it accomplishes our goal. This would explain why people sometimes opt for more aggressive care that decreases average life expectancy.