A new NBER working paper by Michael Darden, Ian McCarthy, and Eric Barrette claims to have found evidence of hospital cost shifting. I’m not so sure.
I will skip the throat clearing about what hospital cost shifting is, why or when we should or should not expect it to occur in theory, and what the prior literature says in some detail about whether it occurs in practice. Go read this Upshot post and all it links to for that. I will only say that based on the prior literature, we should begin consideration of the new paper with some skepticism about claims of cost shifting — it is reasonable to believe it doesn’t happen.
Therefore, even to find, with credible methods, a modest degree of cost shifting — like $0.10 per dollar of public payer shortfall is shifted to private payers — would be a big deal.
What Darden et al. found isn’t modest. It’s huge. They estimated that more than half of Medicare payment shortfalls are recouped by jacking up prices charged to private insurers — 56 cents on the dollar, to be precise. Yow! (Right here you should wonder why hospitals would stop at 56 cents on the dollar. If your answer is, “that’s the limit of their market power,” you’re on the right track. That also means they’ve exhausted their market power and, unless they acquire more, cost shifting should halt. And right there is why more than rare claims of cost shifting aren’t credible.)
But, without even looking at methods, we should be careful about taking this figure to mean that there really is cost shifting at a rate of 56 cents on the dollar. Given prior work, a Bayesian might, at most, update his/her thinking from “there is no cost shifting” to “there could be a little cost shifting.” However, the methods might not even warrant that.
So, what’s up with the methods? Craig Garthwaite did a nice job pointing out a few issues, starting with this tweet:
There have been a bunch of tweets about the new cost shifting paper this morning. I have a couple of thoughts we should keep in mind. First, finding evidence of cost-shifting is hard at least in part because finding the right setting for testing for its presence is hard (1/4)
— Craig Garthwaite (@C_Garthwaite) February 12, 2018
Let’s unpack a few of his concerns, which I share. First of all, the headline result of massive cost shifting is based on examining how private hospital prices change due to changes in hospitals’ Medicare penalty status. That is, Medicare financially penalizes hospitals for lower quality in several ways, which the paper examines. The study found that hospitals that change from not-penalized to penalized status increase their private prices more than hospitals that don’t. From that, they back out a cost shift rate of 56 cents on the dollar.
But there’s another way to estimate what the cost shift rate is, not by looking at changes in penalty status, but by looking at what the penalty amount is. How much does a hospital cost shift when it is penalized an additional dollar? When estimated this way, the authors find small and statistically insignificant evidence of cost shifting, which is a highly credible finding on its face. (See footnote 17 of the paper.)
So, one has to ask, why might hospitals that become penalized differ from those that don’t, and in ways that are associated with increases in private prices? About this, Craig made a very good point: Hospitals were aware in advance of their risk of being penalized. Those that looked like they would be may have invested more in quality improvements. For some, those improvements didn’t translate into avoiding the penalty, but they did increase the value those hospitals were delivering. Private payers may have been willing to pay more for that additional quality. It’s possible they’re even willing to pay more for investments in quality that haven’t yet translated into actual improvements.
Separately, it’s also possible that hospitals that respond to potential penalties change their marginal costs, which would change their profit- (or revenue-) maximizing price. This could look like cost shifting, but it’s still a response to a change in quality or, more generally, cost structure. It wouldn’t be a response to Medicare reimbursement shortfalls in the sense that if Medicare just cut payments without linking them to quality, one would not expect the same change in private prices.
The BIG IDEA here is that firms (here, hospitals) respond to incentives in ways that change their production function (their costs and the value they provide). In fact, this is the point of Medicare quality penalty programs. They should affect what hospitals do and that, itself, can affect prices, as explained above. That’s not cost shifting. Not being able to control for that in a cost shifting study is a big problem. The particular approach taken in this paper likely doesn’t address this problem, leading to estimates that should not be attributed to cost shifting.
I will close by adding that this paper is highly useful for tracking down claims of cost shifting made by hospital executives and policymakers. See the main text as well as footnote 1.
PS: I should make explicit what should be obvious. I would appear to have a large interest in arguing away findings of hospital cost shifting, given my prior writing on the issue. I am so aware of that apparent conflict that I had a few colleagues without that conflict look over this post before publishing. All of their input was incorporated to their satisfaction.