The most important sentence in that hospital price study

8/15/2014 03:28:00 PM
Aaron Carroll and Sarah Kliff are both reporting on a study titled "Variation in charges for 10 common blood tests in California hospitals: a cross-sectional analysis" published in BMJ Open. Both, I think, mostly ignored the most important sentence in the whole thing, buried pretty deep into the article:
"This study evaluated the charges for blood tests, rather than the negotiated prices paid by most insurers."
This is a study of hospital "charges" which, despite the name, are not even close to what hospitals charge for procedures. Instead "charges" are sort of a list price, from which the hospital negotiates with various insurers and other institutional payers over how much to actually pay, often on a patient-by-patient basis.

Studying actual hospital prices, as opposed to the meaningless "charges" is not easy, especially if you are interested in knowing the price of a specific routine procedure, because hospitals do not actually charge on a per-procedure basis. What I mean is that in principle the hospital does furnish the insurer with a list of all the procedures performed on a patient and the associated "charges" but they subsequently negotiate the entire bundle of procedures performed on a particular encounter, so that what Humana pays the hospital for Johnny's transcranial doppler (TCD) scan is not the same as what Humana pays the same hospital for Tommy's TCD scan. These differences arise for two reasons: one, the associated services required to perform the TCD scan is going to be a little different for each patient, and that becomes a basis for negotiation; second, Tommy and Johnny may differ in their insurance coverage--perhaps they have different plans, or perhaps they have the same plan but different underwriting characteristics. These things affect the total bills that Humana pays for seemingly identical patients with seemingly identical procedures. I've done some digging, and as best as I can tell, hospitals do not keep records of negotiated prices for each procedure--their internal pricing data contains the per-procedure "charges" and lump-sum per-patient-encounter bill totals, but never an itemized breakdown of how much an insurer actually pays for what. The way this works is you get a bill with a bunch of "charges" for each procedure, then at the bottom a one-line lump sum "adjustments" that says how much less--it is always less--they are actually billing. My point here is that not only does this study not actually compare the prices of specific procedures between hospitals, such a dataset probably does not even exist.

To really do this study correctly, we'd need to perform a modified Abowd-Kramarz-Margolis decomposition to identify hospital-specific and procedure-specific components to variation in total hospital bills. This procedure is data-intensive, and as far as I know has never been attempted. An alternative approach, and one I use in my own work, is to use hospital cost data, which unlike billing prices, is actually available on a per-procedure basis (but, be careful how you interpret that)--the downside is that costs do not give a sense of per-procedure markups (or frequently, markdowns, as there's lots of cross-subsidization going on), which is the step that adds so much complexity.

How big is the disparity between the charges this study looked at and what hospitals actually bill? I have some data that sheds a little bit of light on that. We want a sense of the difference between charges and bills for routine care rather than, say, severe trauma wounds. I do not have a data set with routine checkups, but I do have a data set with uncomplicated visits to a sickle cell clinic for pain and fever (that is for generally routine, non-complex issues) at a large urban medical center (Update: that should read "large urban pediatric medical center"--my sample is kids only). Sickle Cell is a chronic condition where the red blood cells "sickle" which causes potentially severe inflammation and tissue damage that can be life-threatening in some cases. But pain and fever are two symptoms that are relatively common so that treating these is a routine activity at a sickle cell clinic. Excluding those patients who developed more serious complications beyond just pain and fever, here is a histogram of the ratio of adjustments to total charges:
The actual prices for treatments do not resemble hospital "charges" at all.
The median reduction in charges was 76.2 percent, meaning that what the hospital actually billed insurers was 76.2 percent less than the sum of the per-procedure "charges." More than half the charge price is just bargaining!!!

And just to further give an idea exactly how much cross-subsidization goes on, note that some patients received adjustments more than 100 percent (the maximum adjustment was 107.2 percent below "charges"), so that their total bill was actually less than zero. Now, the clinic isn't giving people cash for coming to the clinic--these less-than-zero bills are being applied to outstanding and generally very large billing balances for chronic patients who visit the hospital often. Yes, in some literal sense they end up weirdly owing less after visiting the clinic than before, but the proper way of understanding this is as an adjustment to the life-cycle path of billing, since the same payer owes that prior billing balance as is paying for this clinic visit.

So enough with all the studies of "charges" already.