• Quick thoughts on Reschovsky et al.

    A 2011 Health Services Research paper by Reschovsky et al. came up in the comments to Aaron’s post about single payer and wait times. I was unaware of the paper so I took a look. (If I’m not mistaken, it’s ungated.)

    Among its findings is that provider supply has weak or insignificant association with resource use (cost, not spending). The authors claim that this is at odds with other Dartmouth work on supply sensitive care. That may be so, but let’s see if there’s any educational value in kicking the tires.

    The Reschovsky work is at the individual level and the dependent variable is an overall cost measure. Supply sensitivity is an area-wide effect, and it arises with respect to use of specific health care services, not necessarily use overall. It doesn’t seem to me that it is inconsistent to find, on the one hand, use of some services (but not others) related to supply in a regional analysis and, on the other, that at an individual level, supply is not strongly related to overall cost. Another way to think of it is that, perhaps, some specialty services crowds out general practitioner use. If that’s the case, the effect on overall cost may be a wash. But that doesn’t mean some types of care aren’t supply sensitive, being over-provided to some populations and not others. I’m speculating here. What do you think?

    The other thing I wonder about is whether their cost measure assumes away some of the potential supply sensitivity effect. They use a predicted cost measure meant to reflect resource use, not actual spending. Their method reduces the variance of the cost variable in a way that may conceal some important relationships. They say that using spending would instead introduce selection bias “because treatment efficiency will influence … a patient’s actual cost.” But isn’t that the point of the supply sensitive argument? We have health systems of varying efficiency, which is related to patterns of use. Again, I’m interested in readers’ thoughts.

    There is, of course, a final issue that the authors can’t really do anything about. It’s a huge problem in health services research. It’s the assumption that diagnoses are exogenous. It doesn’t hold as much as we’d all like it to. That being the case, diagnoses as controls “over explain” cost, soaking up some of what would otherwise be considered a supply sensitivity effect. I have little doubt that diagnoses are supply sensitive just as much as care provision is. They share incentives, don’t they?

    @afrakt

    Share
    Comments closed
     
    • How much effort do specialist put into networking with general practitioners in order to get/increase referrals?

    • Austin
      I have reread this post numerous times, and still find it confusing.

      Regardless, when you refer to subspecialty crowd out, do you mean that the presence of specialty docs are displacing PCPs, and the wash is their driving supply side resource use up or down? I am losing you with what is actually doing the displacing.
      Brad

      • Imagine there is a limit to the number of doctor visits a person can have. (No, in principle there is no limit. But, come on, in practice, the vast majority of individuals are only going to consume so much care.) In that case, it is reasonable to imagine that specialist visits might displace (crowd out, stand in for, substitute for) general practitioner visits. That being the case, the market may adjust such that there are more specialists than there otherwise would have been (supply and demand) and fewer GPs.

        No disrespect intended, but think of high price, fancy restaurants replacing low price, dumpy ones in a gentrifying town. There are only so many locations. People eat out only so much. More of one may lead to less of another.

        • Makes sense,but where it gets dicey is the equivalence in the substitution. .

          A cards visit, in lieu of a PCP one, if that cardiologist serves as the primary, can be either a 1:1 wash, or if that same doc has twitchy fingers and likes technology, can be a cardiac boondoggle.

          Difft subspecialists have difft degrees of SID (?). Does an endocriniologist serving a diabetic, or a nephrologist serving a renal patient, etc., improve or worsen utilization?

          That is where my thinking was going, thus the confusion. As they say, its complicated.

          Brad

    • Thanks to Austin for alerting me to his post about the HSR article my colleagues and I published last year.

      I have no doubt that some physicians induce demand and indeed some of my other research offers evidence to suggest that physicians are more likely to provide services that offer them higher margins. That said, the Dartmouth conclusions that geographic variations are largely driven by induced demand (i.e., “supply sensitive” goods) or for that matter that greater medical spending has no beneficial health effects strike me as poor social science, based on simplistic correlations subject to confounding/endogeneity.

      To find correlations between the supply of physicians and spending/cost does not tell you whether physicians are inducing demand or whether physicians locate in areas where there is greater demand for their services. There are large differences in the prevalence of conditions between “high-cost” and “low-cost” areas, suggesting that the latter explanation cannot be discounted. These differences in health status apply to both conditions where there is no discretion in diagnosis (e.g. hip fractures, AMIs) as well as other conditions where there are arguably more physician discretion in diagnosis.

      Austin makes a couple methodological comments, both of which I think are wrong. First, he argues that we are less likely to find associations with provider supply because we use micro-level data. If anything, use of micro data is a strength of our analysis. I don’t quite follow Austin’s crowding out arguement, but since we control for the supply of specialists and primary care physicians separately, I don’t think our results would be affected. We fail to find an association between provider supply and cost because we control for other factors ( consistent with Zuckerman, et al.’s (2010) findings with respect to spending variations). Looking only at correlations of area level means is subject to ecological biases as well as confounding.

      I disagree with Austin’s comments that our use of standardized costs, rather than spending might be biasing results. Using standardized costs simply takes away aspects of the Medicare fee schedule that has providers paid more in high cost areas (NYC) than low cost areas (rural Kansas) for providing the same service, as well as other pecularities of the payment system that lead to some providers to be paid more than others for the same service. If anything, one is more likely to obtain biased results when looking at variations in spending..

      In the end (and this is based on some subsequent research not yet published) geographic variations, supply inducement and such are far more complicated than what the Dartmouth Atlas would have you believe. There are geographic variations, but they are more likely to be specific to certain services or conditions and not likely to be as determinative on location as they would have you beleive. Moreover, the reasons for unexplained variations are difficult to identify and cannot necessarily be attributed to inefficiency.

      Jim R.